# Content and statistical-roster open loops This file is the substantive content/statistical backlog, not the active-agent roster. For current execution state read root `CURRENT_WORK_STATE.md` together with `PIPELINE_STATE.json`. The former deep `AUDIT_ACTIVE_STATE.md` is frozen history. **Current execution update — 8 September 2026:** v7 is locally activated: 676 admitted events,672 family assignments,4 family holds. The fixed17-package correction cycle,565-scenario diagnostics, metadata carry-forward, two-date addendum and portable reproduction package are complete. The370-case/380-stop display is integrated with build20260908a. The companion PUBLICATION.json and dated deployment receipt distinguish release activation from verified serving. Do not relaunch closed scientific checks from the historical notes below. Six raw-F8 follow-ups remain; Cape is counted once. Remaining authoring: 306 =246 ordinary +56 separate authoring holds +4 family holds. Exact IDs and execution state are in the v7 partition and root Current Work State. Original results remain immutable; the chronology supplement is an explicit overlay. The dated decisions below are history, not competing launch instructions. **Raw-F8 source preparation — 9 September 2026:** the private six-event packet reconciles twelve actual v7 cells and prepares source-first questions; all six substantive decisions remain open. No raw value, boundary or public display was changed. Megiddo's SCS is already U despite the old queue's zero wording; its PEL is 4 and has a separate anchor question. The Aydın PDF is Grant's *Greek Captives and Mediterranean Slavery, 1260–1460* (2024), not the older recorded title. The packet preserves the distinct modern cohorts, Cape project-scope counterevidence and Megiddo temple/cumulative-gift limits. Independent treatment, main adjudication and proportionate application checks are still required. This note corrects the current reading of the historical queue, not its frozen bytes, and does not reopen the closed correction cycle wholesale. Historical checkpoint: 2026-08-30, after promotion of immutable `second_pass_v6_675`, deployment and validation of atlas authoring Batch 23, closure of Batches 24 and 25 without public additions, and deployment and validation of bounded Batches 26 through 30. The older checkpoint began after v4 promotion, corrected Hellenistic coding, assembly of the 689-row event-unit register, completion of all 333 qualified event checks and 245 relationship checks, and the single main adjudication of all 147 findings. The non-destructive verified register is frozen at `9b22d64`; the 126-obligation target-readiness inventory is frozen at `36aa3a5`; and twelve bounded source-review packets are frozen at `03bcff6`. Event-unit fairness gate added 2026-08-23. The current execution update above governs over the dated snapshots below. This memo excludes ordinary map authoring, bibliography production, and deployment. It lists only questions that can still add, remove, split, or materially reinterpret statistical events. ## Historical post-promotion checkpoint — 2026-08-30 **Holistic sequencing decision, approved by Craig 2026-09-05:** `../../HOLISTIC_PROJECT_REVIEW_2026-09-05.md` recommends reconciling and applying accepted corrections through Batch 50 before another long authoring run, while keeping existing event-unit rules, closed reviews, and immutable v6. The saved read-only checkpoint query counts 82 correction entries across 81 distinct existing event IDs in Batches 24–50; the entries include metadata and source changes as well as score/boundary work, so this is not an invalid-event count. Batch 51 is selected but source work is not started. The correction checkpoint now precedes routine authoring; approval does not itself change admission, scores, or frozen-release contents. Existing evidence gates apply. **Superseding Batch 50 addition, 2026-09-05:** its historical review is closed; application remains open. Retire the broad Egypt (639–c.750), Emishi Tōhoku (c.660–811), and Umayyad Iberia (711–c.800) parents in the successor release. Five bounded replacements await sixty-dimension independent coding, adjudicated scores, and boundary/model tests. Qays 727 and Syrian junds 741–c.744 remain candidate holds. These are documented in `v6/ATLAS_AUTHORING_BATCH_50_ADJUDICATION.md` and `v6/BATCH_50_RELEASE_CORRECTIONS.json`. The three parent authoring holds are applied; neither their retirement nor child promotion has mutated immutable v6. Use `v6/NEXT_RELEASE_CORRECTION_QUEUE.md` for the application entry and its coverage warning: the older Batch 24–30 table is not the current full inventory. - Batch 30 is closed after one bounded source packet, one completed independent review, one adjudication, and one implementation cycle. Aegina enters as one 431–406 analytical-tenure stop; the separate 424 Thyrea violence and 405 restoration are not merged into it. Fidenae remains two source-bounded events: the selective 428 exile/reinforcement/land action and the distinct 426 conquest/captive-disposition/incorporation action. The nonhistorical median-merge stress test changes only five of 581 shared k=6 labels (0.86%, ARI 0.978907), the other grouped-versus-split tests change none, and every scenario has `material_flag: false`; the split therefore follows sources without materially steering the model. All three events are public in build `20260830a`, all 867 endpoints touch land, and local/live desktop and phone audits pass all 334 frames with zero issues. Their public trait and 426-family corrections are queued for the successor immutable release. See `v6/ATLAS_AUTHORING_BATCH_30_ADJUDICATION.md`, `v6/ATLAS_AUTHORING_BATCH_30_IMPLEMENTATION_QA.md`, and `v6/deployment_20260830a_atlas_authoring_batch_30.json`. No recursive review is authorized. - Batch 29 is deterministically selected from the 298 ordinary unrendered events after excluding two family holds and 56 separate authoring holds. Its three units are the approximately 479 BCE post-Mycale refoundation and repopulation of Miletus, Antium's 468–467 BCE conquest/partial land division and integrated colony, and the 446–442 BCE Ager Coriolanus seizure followed by locally weighted Ardeate restitution or reorganization. Selection does not settle inclusion, event boundary, family, trait display, map, or prose. The bounded source/model packet is now frozen. It identifies Miletus as a likely restoration control without a demonstrated affected prior population, Antium as a qualifying but explicitly mixed and locally continuous captured- land division, and Ardea as a qualifying land seizure whose locally weighted restitution must remain inside the event boundary. The sole completed independent review and one adjudication are now closed. Miletus is held for failed atlas admission and must retire from the successor active roster while remaining non-counted aftermath context for the distinct 494 BCE event. Antium and Ardea are authorized as one conservative stop each, with public uncertainty/correction overrides and successor-release recodes. Ardea's two dates are one internal sequence, not duplicate trait cards. Authoring and local QA are now complete in build `20260829b`: Antium and Ardea enter as one stop each, Miletus remains absent, all 861 public map endpoints touch land, and desktop and phone audits pass all 331 frames with zero issues. Latrian now serves the checksum-identical candidate, and both live responsive audits pass all 331 frames with zero issues. Antium and Ardea direct links resolve; the Miletus ID opens the preserved hold reason. Batch 29 is closed; no recursive review is authorized. See `audit/select_v6_authoring_batch_29.py`, `v6/ATLAS_AUTHORING_BATCH_29_SELECTION.json`, `v6/ATLAS_AUTHORING_BATCH_29_REVIEW.md`, and the three Batch 29 evidence and geography sidecars, `v6/ATLAS_AUTHORING_BATCH_29_INDEPENDENT_REVIEW.md`, `v6/ATLAS_AUTHORING_BATCH_29_ADJUDICATION.md`, and the Batch 29 hold, correction, and public-trait records, plus `v6/ATLAS_AUTHORING_BATCH_29_IMPLEMENTATION_QA.md` and `v6/deployment_20260829b_atlas_authoring_batch_29.json`. - Batch 28 was deterministically selected from the then-current 301 ordinary unrendered events after excluding both family holds and all 55 durable authoring holds. Its three source-review units are Miletus/Ampe (494 BCE), Circeii (491 BCE), and Eretria/Ardericca (490 BCE). Its sole review and adjudication are closed. Miletus and Eretria remain bounded captured-cohort dispositions and are authorized for conservative public authoring with successor-release trait corrections. Circeii is held: Livy reports expulsion while Dionysius says nobody was expelled, so a likely Volscian control change cannot stand in for demonstrated removal. Circeii requires retirement or full redefinition and recoding in the successor release. The next finite step is authoring, QA, and deployment of Miletus and Eretria. See `v6/ATLAS_AUTHORING_BATCH_28_SELECTION.json`, `v6/ATLAS_AUTHORING_BATCH_28_REVIEW.md`, both Batch 28 source sidecars, and `v6/ATLAS_AUTHORING_BATCH_28_INDEPENDENT_REVIEW.md`, `v6/ATLAS_AUTHORING_BATCH_28_ADJUDICATION.md`, and `audit/select_v6_authoring_batch_28.py`. Miletus and Eretria are implemented as one stop each in deployed build `20260829a`; all 858 endpoints touch land and local/live desktop/phone audits pass all 329 frames with zero enumerated issues. Local, staged, installed, and live HTTPS files are byte-identical. Circeii is preserved in a standalone authoring hold and is absent from the public roster. Batch 28 is closed; see `v6/deployment_20260829a_atlas_authoring_batch_28.json`. No recursive review is authorized. - Batch 27 closes three source-critical event-unit reviews without changing immutable v6. The Barca case remains one coerced imperial transplantation sequence but changes from exact 512 BCE to approximately 515–510 BCE and requires F2/F3/F6/F7/F8 recoding. Pometia remains one destruction and captive sale event with alternative c. 502 or 495 notices—not a 503–495 continuous action—and requires structural-NA and uncertainty repairs. Velitrae remains one 494/492 conquest, settlement, and reinforcement sequence; coercion of incoming Roman colonists is not prior-resident removal or imported unfree labor. All three are authorized for one qualified public stop while the underlying corrections wait for the successor statistical release. See `v6/ATLAS_AUTHORING_BATCH_27_ADJUDICATION.md` and `v6/BATCH_27_RELEASE_CORRECTIONS.json`. All three are implemented and live as one stop each in build `20260828zb`; local and live desktop and phone audits pass all 327 frames, all 854 map endpoints touch land, and the three direct case links resolve. Local, staged, installed, and live files are byte- identical. Batch 27 is closed; see `v6/deployment_20260828zb_atlas_authoring_batch_27.json`. - Batch 26 has one public addition and two content/statistical holds. The public addition is Megabazus's c. 512 BCE deportation of named Paeonian communities to Phrygia, authored conservatively from Herodotus with F1 solid and F2–F8 limited. The next statistical release must still correct its date and actor metadata, add the 5.23 locator, and recode source-silence-derived cells before regenerating F2–F8. The Neo-Babylonian row must retire as a mixed wrapper and become a newly coded Judean 597 + 587/586 two-phase program; Ashkelon becomes a separate candidate, 582/581 remains held, Tyre is excluded, and 539 is not an event end. The Vanuatu row remains a supported migration/ancestry-turnover process but must be rebound provisionally to c. 750–350 BCE and fully recoded without inferring coercion or institutional absence from genetics. These decisions are closed in `v6/BATCH_26_RELEASE_CORRECTIONS.json`; immutable v6 is unchanged. - Immutable `second_pass_v6_675` is active locally: 675 modeled events, 748 preserved corpus rows, 73 inactive audit rows, 60 dimensions, and 40,500 score cells. Local and live build `20260830a` render 324 active events in 334 playback stops. The six Batch 23 additions and three refreshed records use the source-adjudicated v6 chronology, family, cohort, count, route, and trait decisions, plus the qualified Batch 26 Paeonian case and the three qualified Batch 27 source-critical cases, the two qualified Batch 28 Achaemenid cases, and the qualified Batch 29 Antium and Ardea cases, plus all three qualified Batch 30 Aegina/Fidenae cases. All 867 endpoints touch land; local and live desktop and phone audits pass all 334 frames with zero issues or browser errors. Build `20260830a` remains byte-identical across its recorded local/staged/installed/live copies. The monograph and 52-file data package remain the unchanged publication build `20260828z`. - Atlas authoring Batch 24 is closed after one bounded independent review and one adjudication. The second Zamua campaign, Arbakki–Naʾiri, and Matiātu remain supported, separate events, but all three are held from public authoring until a successor release corrects their source boundaries, cohort wording, families where necessary, and unknown-versus-low trait treatment. In particular, 880 Zamua changes to conquest/incorporation and Matiātu's locally returned cohort is explicitly troops rather than generalized survivors. Batch 24 adds 0 cases / 0 stops; live `20260828z` remains 313 / 323. No reviewer remains active and no recursive review is authorized. See `v6/ATLAS_AUTHORING_BATCH_24_ADJUDICATION.md`, `v6/BATCH_24_AUTHORING_HOLDS.json`, and `v6/BATCH_24_RELEASE_CORRECTIONS.json`. - The active release has 351 events not yet rendered in live `20260830a`. Of those, 349 have a reviewed primary family and two remain explicit family evidence holds: Croton and the partial Paeonian return around 498 BCE. Fifty-seven of the 349 family-classified events are separate authoring holds. The original 55 are Aydın, Menteşe, Ottoman Bithynia, 39 unrendered early Neo-Assyrian events, all eleven unrendered later Neo-Assyrian source-series events, the Neo- Babylonian wrapper, and the Vanuatu ancestry row. Six Assyrian rows have closed Batch 24/25 correction instructions; the other 44 are held only for one combined source-calibration lane, not because an event-specific defect has been inferred. Circeii is the 56th separate hold because its ancient accounts directly conflict over whether colonists were expelled. The post- Mycale Miletus restoration is the 57th because it is supported history but fails the atlas's land-taking/removal admission rule. The early roster explicitly includes the active legacy-ID Aribua/Luḫūti merged successor. The other 292 are ordinary authoring work, not new inclusion or model votes. The next ordinary batch skips both held Assyrian tranches rather than repeating the same three-event audit pattern. - The Roman-wrapper question is closed at the temporary-warning level. All 57 source-bounded Roman Republican Italy successors and all 15 Po Valley successors are now authored and appear once each. The former wrappers are absent from playback and statistics. Their old direct URLs remain neutral successor indices so saved links still explain the split; these are not warnings that research or coding remains unfinished. No additional statistical work is required to deprecate the old temporary notice. - The staged public package publishes the working monograph in HTML, Markdown, and PDF, together with the immutable v6 matrices, event register, evidence and provenance ledgers, sensitivities, atlas inventory, codebook, and bibliography inventories. Unfinished research, copyrighted source captures, scratch files, and raw model/coder journals remain excluded. This package is explicitly a working release, not a completeness claim. - The ten closed post-v5 obligations have completed their one finite application cycle. The resulting candidate has 675 active events and 748 preserved corpus rows: the broad Indo-Aryan wrapper is inactive but audit-preserved, two Assyrian IDs are corrected through durable aliases, three families change, and 117 source-adjudicated score cells change across nine retained events. All data/editorial builders reproduce and pass their validators. The bounded model comparison also reproduces: v5 and the corrected candidate both retain nine factors; their nine-factor maximum principal angle is 2.765 degrees, compared with 6.685 degrees under forced eight. No corrected event is individually material. The low same-name F2/F7/F8 comparisons are therefore an eight/nine-boundary rotation, while inherited sensitivity to weighting, imputation, documentation, and coherent event-resolution alternatives remains a reported limitation. Commit `f9b20198` is the restart checkpoint. The one scoring reviewer is closed and no recursive review is authorized. Real remaining content/statistical work: one final Editorial Suite, immutable successor promotion/validation, then one authoring/QA/deployment batch for the newly eligible held events and a refresh of the public data and monograph. - Atlas authoring Batch 22 is closed after its sole bounded ORACC/Sano review and one adjudication pass. All three campaign units are supported but held for the next immutable release. Nimme/Tummu–Ḫabḫu requires an 882/883 chronology/ID correction, family change, cohort limits, and trait recoding; Upper Tigris/Nirbu remains one phased 882 conquest/incorporation event but requires cohort-specific F2/F3/F6/F7 and unknown F8; Zamua 881 remains one event but changes to conquest/incorporation and requires military/city count limits and trait recoding. Batch 22 adds 0 cases / 0 stops; live build `20260828y` remains 307 / 317 and frozen v5 is unchanged. Hooke the 2nd is closed, no agent remains active, and no recursive review is authorized. The next action is one successor-release application of the ten closed post-v5 obligations in `v5/NEXT_RELEASE_CORRECTION_QUEUE.md`, followed by the affected model reruns. See `v5/ATLAS_AUTHORING_BATCH_22_ADJUDICATION.md`, `v5/BATCH_22_AUTHORING_HOLDS.json`, and `v5/BATCH_22_RELEASE_CORRECTIONS.json`. - Atlas authoring Batch 21 is closed after its sole bounded source review and one adjudication pass. All three event units are historically supported, but all three are durable next-release correction holds rather than public cases. Amenhotep II Year 9 requires chronology/count/route and cohort-trait corrections; Muršili II's Arzawa removal requires count interpretation and trait corrections; Ashurnasirpal II's Nipur–Paṣāte phase requires a corrected 882/883 chronology/identifier, a change to `conquest_and_incorporation`, and trait corrections. The next immutable release must apply those records and rerun the affected models before authoring. Batch 21 adds 0 cases / 0 stops; public build `20260828y` remains 307 / 317 and frozen v5 remains unchanged. These holds are not rejections of event existence. There is no active reviewer and no recursive review is authorized. See `v5/ATLAS_AUTHORING_BATCH_21_INDEPENDENT_SOURCE_REVIEW.md`, `v5/ATLAS_AUTHORING_BATCH_21_ADJUDICATION.md`, `v5/BATCH_21_AUTHORING_HOLDS.json`, and `v5/BATCH_21_RELEASE_CORRECTIONS.json`. - Atlas authoring Batch 20 has completed its sole independent source review and one adjudication/repair pass. The broad Indo-Aryan row is now an explicit public-authoring hold and next-release event-unit correction: movement and admixture are supported, but the active frozen wrapper is not a bounded qualifying atlas event. Thutmose III Year 23 and Amenhotep II Year 7 pass as separate state capture-and-removal events. Their public cases preserve the distinct Thutmose lists, the contradictory Amenhotep totals, broad rather than invented routes, unresolved later outcomes, and fail-closed traits. Local build `20260828y` has 307 cases / 317 stops and passes 831/831 endpoint checks plus all 317 desktop and mobile frames with zero issues. Local, staged, installed, and live HTTPS files are byte-identical, and the complete live desktop/mobile audits also pass with zero issues. Frozen v5 is unchanged; the Indo-Aryan event-unit decision and Egyptian date/trait corrections are queued in `v5/BATCH_20_RELEASE_CORRECTIONS.json` for the next immutable release and model rerun. Batch 20 is closed; no recursive review is authorized. Governing records are `v5/ATLAS_AUTHORING_BATCH_20_INDEPENDENT_SOURCE_REVIEW.md`, `v5/ATLAS_AUTHORING_BATCH_20_ADJUDICATION.md`, `v5/INDO_ARYAN_AUTHORING_HOLD.json`, `v5/BATCH_20_PUBLIC_TRAIT_DISPLAY_OVERRIDES.json`, and `v5/deployment_20260828y_atlas_authoring_batch_20.json`. - Atlas authoring Batch 19 has passed its sole bounded findings-only source review and its one authorized adjudication/repair pass. All three events are authored locally in build `20260828x`: the directly supported Scandinavian footprint is centered on Denmark; the Southeast Asian map uses only the reviewed early anchors in Vietnam, Laos, and Peninsular Malaysia; and the Iberian case preserves substantial local autosomal ancestry, coexistence, admixture, and the distinction between Y-lineage turnover and physical elimination. Broad movement lines are explicitly schematic demographic connections, not observed or forced routes. All three use source-first F8 zero overrides; Iberia uses a public F7 not-applicable/unknown correction that is queued for the next immutable release. Frozen v5 is unchanged. The live roster is 305 cases / 315 stops, leaving 371 active events unrendered (369 family-classified plus two family/evidence holds). All 827 endpoints and all 315 local/live desktop/mobile frames pass. Local, staged, installed, and live HTTPS files share SHA-256 `5d1053ea95b71aecc8194dd4713495717caca4e728dc848ca6e050cb65332788`. Batch 19 is closed; the governing records are `v5/ATLAS_AUTHORING_BATCH_19_REVIEW.md`, `v5/ATLAS_AUTHORING_BATCH_19_INDEPENDENT_SOURCE_REVIEW.md`, `v5/ATLAS_AUTHORING_BATCH_19_ADJUDICATION.md`, and `v5/BATCH_19_PUBLIC_TRAIT_DISPLAY_OVERRIDES.json`, and `v5/deployment_20260828x_atlas_authoring_batch_19.json`. - Atlas authoring Batch 18 is independently reviewed once, repaired and adjudicated once, fully validated, deployed, and checksum/browser verified as build `20260828w`. The two old public parents (`canary-islands` and `israeli-settlements-opt`) now open neutral indexes of seven active v5 successors and no longer occupy playback. La Gomera is publicly bounded to 1488–1489 and does not expose the obsolete 1449 proxy. West Bank–East Jerusalem and Golan are separate systems beginning in 1967; Gaza is excluded. The frozen v5 event units, weights, families, traits, and publication data remain unchanged. Two source-supported metadata corrections are queued in `v5/BATCH_18_RELEASE_METADATA_CORRECTIONS.json` for the next immutable release and model/boundary-sensitivity rerun. The live atlas has 302 cases / 312 stops, all corresponding to active v5 events, and leaves 374 active events unrendered (372 family-classified plus two family/evidence holds). All 819 endpoints and all 312 local/live desktop/mobile frames pass. Local, staged, installed, and live files share SHA-256 `b0c12ffc4a5cfc59b407ebb350e880992dc83cf26c11b28edc11ed8e205104f6`. See `v5/ATLAS_AUTHORING_BATCH_18_REVIEW.md`, `v5/ATLAS_AUTHORING_BATCH_18_INDEPENDENT_REVIEW.md`, `v5/ATLAS_AUTHORING_BATCH_18_ADJUDICATION.md`, and `v5/deployment_20260828w_atlas_authoring_batch_18.json`. Batch 18 is closed; no agent is active. - Atlas authoring Batch 17 is independently reviewed once, repaired once, fully validated, deployed, and checksum/browser verified as build `20260828v`. It renders the final fourteen active Roman-Italy leaves plus independently retained Narbo Martius. The broad Italy wrapper has left playback and its stable URL now opens a neutral fifty-seven-successor index; Narbo is not counted in that Italian roster. The sole review found no reason for a model rerun, deletion, family change, or trait recode. Its bounded repair removed Narbo from the parent lineage, withdrew the categorical 118 BCE Atacine-citizenship claim, exposed the disputed identity and incorporation timing, tightened direct-source joins and locators, and kept authority distinct from demonstrated human origin. The frozen event units, families, traits, factors, and weights remain unchanged. The live atlas has 297 cases / 307 stops, renders 295 active v5 events, and leaves 381 unrendered (379 family-cleared plus two family/evidence holds); two visible cases remain outside active v5. All 819 endpoints and all 307 local and live desktop/mobile frames pass. Local, staged, installed, and live files share SHA-256 `1204a0acd6d59f3c4ae63ebb208356995bbff751da42cb09e108e16d601cff0b`. See `v5/ATLAS_AUTHORING_BATCH_17_REVIEW.md`, `v5/ATLAS_AUTHORING_BATCH_17_INDEPENDENT_REVIEW.md`, and `v5/deployment_20260828v_atlas_authoring_batch_17.json`. Batch 17 and the Roman-wrapper authoring loop are closed. - Atlas authoring Batch 16 is independently reviewed once, repaired once, fully validated, deployed, and checksum/browser verified as build `20260828u`. It renders the next fifteen chronological active Roman-Italy leaves from 290–194 BCE and raises the still-visible wrapper from 28/58 to 43/58 authored, leaving fifteen. The review found no statistical-release contradiction. Its one Sipontum join blocker and ten authoring/citation findings were applied once: the 194 BCE successor cannot inherit the retired 186 BCE phase; Brundisium retains disputed two-date ranges; Hadria, Aesernia, Curius, Spoletium, and the Campanian count are calibrated to their sources; and thirteen unsupported Rome-origin routes were removed. The frozen event units, families, traits, factors, and weights remain unchanged. The live atlas has 283 cases / 293 stops, renders 280 active v5 events, and leaves 396 unrendered (394 family-cleared plus two family holds); three visible cases remain outside active v5. All 806 endpoints and all 293 local and live desktop/mobile frames pass. Local, staged, installed, and live files share SHA-256 `45fd9eef088128ac55e09ab33244244d62ce94d19566169ce3c15fd7d9e57aed`. See `v5/ATLAS_AUTHORING_BATCH_16_REVIEW.md`, `v5/ATLAS_AUTHORING_BATCH_16_INDEPENDENT_REVIEW.md`, and `v5/deployment_20260828u_atlas_authoring_batch_16.json`. Batch 16 is closed. The final Italy authoring loop is one fixed fifteen-event tranche; once it passes the same bounded gate, the broad Roman wrapper can leave playback. - Atlas authoring Batch 15 is independently reviewed once, repaired once, fully validated, deployed, and checksum/browser verified as build `20260828t`. It renders fifteen active frozen-v5 early Republican Italian leaves from 340–291 BCE and raises the still-visible Italy wrapper from 13/58 to 28/58 authored, leaving 30. The map distinguishes incoming Roman settlement or allotment from demonstrated removal and does not invent survivor destinations. The bounded review repairs added direct Livy locators for Pontiae and Saticula, removed a Cales continuity overclaim, exposed exact Auruncian locators, and removed internal workflow wording. The corrected marker gate now tests the actual final public roster rather than the older retained-case subset; all 785 endpoints pass. The live atlas has 268 cases / 278 stops, renders 265 active v5 events, and leaves 411 unrendered (409 family-cleared plus two family holds); three visible cases remain outside active v5. All local and live desktop and mobile frames pass with zero issues. Local, staged, installed, and live files share SHA-256 `e0721a1c3f8e3f50a8499ac27ec5cccf18176a12adff7ae3864b315d559eb145`. See `v5/ATLAS_AUTHORING_BATCH_15_REVIEW.md`, `v5/ATLAS_AUTHORING_BATCH_15_INDEPENDENT_REVIEW.md`, `v5/ATLAS_MARKER_LAND_GATE_REPAIR_20260828.md`, and `v5/deployment_20260828t_atlas_authoring_batch_15.json`. This batch is closed and does not reopen the frozen statistical release. - Atlas authoring Batch 14 is independently reviewed once, corrected once, fully validated, deployed, and checksum/browser verified as build `20260828s`. It retires `med-roman-po-valley-colonization` from playback and renders all fifteen active frozen-v5 successors once each. Thirteen also overlap the 58-event active Italy descendant roster; the validator now derives and requires that exact active-leaf intersection, and the still- visible Italy wrapper reports 13 authored and 45 remaining. The independent review's 57-denominator conclusion was retracted after the next-batch roster derivation exposed the pre-v5-versus-active-release distinction. The review also corrected a wrong JRS author attribution, an inland camera shortcut, a generic Apuan uncertainty fallback, and a fail-open progress test. None of those repairs changes the frozen statistical release, families, traits, or event weights. The live atlas has 253 cases / 263 stops, renders 250 active v5 events, and leaves 426 unrendered (424 family-cleared plus two family holds); three visible cases remain outside active v5. The substantive desktop and mobile audits pass all 263 frames with zero issues, all 666 endpoints touch land, and the stamped live file is byte-identical to the local audited file at SHA-256 `c925560d502d57e9c4cd3659f35f1b8993bfb3ba75b540a07fc51366a885873c`. See `v5/ATLAS_AUTHORING_BATCH_14_REVIEW.md`, `v5/ATLAS_AUTHORING_BATCH_14_INDEPENDENT_REVIEW.md`, `v5/ATLAS_AUTHORING_BATCH_14_POSTDEPLOYMENT_CORRECTION.md`, and `v5/deployment_20260828s_italy_denominator_correction.json`. This batch is closed, including the correction. - Atlas authoring Batch 13 is independently reviewed once, corrected once, fully validated, deployed, and checksum/browser verified as build `20260828q`. It replaces inactive `manchukuo` with the two active frozen-v5 successors for the 1905–1931 South Manchuria Railway Zone and the 1932–1945 Manchukuo agricultural-colonization program. The adjacent one-stop events remain separate because the sources support different institutions, land settings, affected populations, coercion evidence, families, and trait profiles. The first maps unequal colonial incorporation and urban migration without inventing a removal campaign; the second maps state-directed rural colonization and documents coerced land transfer and collective-village removal without inventing a precise removal destination. The one review required Jiamusi to be labeled as an arrival/ staging point and moved the Railway Zone source to western Japan/Kyushu; both findings were applied once. The parent-swap and leave-program-out sensitivity tests retain nine factors, so the split is historically rather than model-result driven. The corrected local atlas has 239 cases / 249 stops, renders 235 active v5 events, and leaves 441 unrendered (439 family- cleared plus two family holds); four visible cases remain outside active v5. All local and live desktop and mobile frames pass with zero issues and all 666 markers touch land. Local, staged, installed, and live files share SHA-256 `8958171b4326de951d6333342539708f10c1ea120419e971963924a4972f99d7`. See `v5/ATLAS_AUTHORING_BATCH_13_REVIEW.md`, `v5/ATLAS_AUTHORING_BATCH_13_INDEPENDENT_REVIEW.md`, and `v5/deployment_20260828q_atlas_authoring_batch_13.json`. This batch is closed. - Atlas authoring Batch 12 is independently reviewed once, corrected once, fully validated, deployed, and checksum/browser verified as build `20260828p`. It replaces inactive `cahokia-colonies` with the two active frozen-v5 successors at Aztalan and Trempealeau. Both retain Low-state durable expansion as their frozen nearest analytical family, but neither claims demonstrated forced removal; Trempealeau's short duration is explicitly distinguished from the family name. Aztalan also preserves the source's qualified evidence of possible conflict without converting it into conquest or removal. The old URL gives a neutral two-successor notice. The parent alias is now bound to frozen `atlas-cahokia-colonies` and fails closed on retirement status, target count, active-release absence, and exact child roster. The live atlas has 238 cases / 248 stops; it renders 233 active v5 events and leaves 443 unrendered (441 family-cleared plus two family holds). Five visible cases remain outside active v5. All 248 local and live frames pass at desktop and mobile sizes with zero issues. Local, staged, installed, and live files share SHA-256 `cb0cfda24c0cf66d93c72a76d6a3edfa102d66f3b0dc4e8b09c52838dfd7bd22`. See `v5/ATLAS_AUTHORING_BATCH_12_REVIEW.md` and `v5/ATLAS_AUTHORING_BATCH_12_INDEPENDENT_REVIEW.md`; the deployment ledger is `v5/deployment_20260828p_atlas_authoring_batch_12.json`. This batch is closed. - Presentation cleanup v1 is independently reviewed, corrected once, deployed, and checksum/browser verified as atlas build `20260828o`. It closes the two queued presentation questions without changing statistical content: the unreliable legacy free-text Mechanism chip is removed from public analytical prominence while the primary family and evidence-aware traits remain; and the obsolete “Worldwide audit” process label/boilerplate is removed from reader-facing cards while structured review status, confidence, substantive corrections, sources, and provenance remain. The one review found an unqualified provisional-family tooltip and a fail-open idempotence gate; both were corrected once. All 247 local and live frames pass at desktop and mobile sizes, and all deployment copies share checksum `017e3afd510dda05c05aaab9b7b7afc758f837d81cca7c62f996635cccb4c21c`. See `v5/ATLAS_PRESENTATION_CLEANUP_V1_REVIEW.md`, `v5/ATLAS_PRESENTATION_CLEANUP_V1_INDEPENDENT_REVIEW.md`, and `v5/deployment_20260828o_presentation_cleanup_v1.json`. This presentation task is closed for the current release. - Atlas authoring Batch 11 is independently reviewed, corrected once under the bounded stop rule, deployed, and checksum/browser verified as build `20260828n`. It replaces the final visible retired parent from the frozen v5 disposition table—the broad South African Land Act/apartheid removals wrapper—with the 1913 Land Act tenure regime, District Six removal under the Group Areas system, and Bantustan consolidation removals. The independent review removed Sophiatown from the Group Areas unit because that removal used the separate Natives Resettlement Act, and replaced a fail-open prose-path check with the machine-readable thirteen-wrapper authority register `v5/ATLAS_AUTHORING_AUTHORITY_REGISTER.json`. The old URL preserves a neutral three-link notice. All 247 local and live frames pass at desktop and mobile sizes; local, staged, installed, and live files share checksum `684640c60e26e44a680003c595840e626d9975bc18f6e7b275b3785a6609baf7`. See `v5/ATLAS_AUTHORING_BATCH_11_REVIEW.md`, `v5/ATLAS_AUTHORING_BATCH_11_INDEPENDENT_REVIEW.md`, and `v5/deployment_20260828n_atlas_authoring_batch_11.json`. This batch is closed. - Atlas authoring Batch 10 is independently reviewed, corrected, deployed, and checksum/browser verified as build `20260828m`. It retires the defective broad Beylik/Ottoman Anatolia wrapper without rendering any of its three active frozen successors. Aydın is held for an operational-date defect, a source-inverted certain-zero F8 draft, and unsupported route semantics; Menteşe is an out-of-scope mixed event unit; and Ottoman Bithynia is mixed and phase-compressed. The old URL preserves a neutral three-hold notice with no successor link or redirect. The validator now proves the exact Menteşe and Bithynia CSV event rows, dispositions, and rules. All 245 local and live frames pass at desktop and mobile sizes; local, staged, installed, and live files share checksum `33690982c2a06b0b2cc731a38ddadbb8c020bd736a2c9558e022d845de83b7fd`. See `v5/ATLAS_AUTHORING_BATCH_10_REVIEW.md`, `v5/ATLAS_AUTHORING_BATCH_10_INDEPENDENT_REVIEW.md`, and `v5/deployment_20260828m_atlas_authoring_batch_10.json`. This batch is closed. - Atlas authoring Batch 09 is independently reviewed, corrected, deployed, and checksum/browser verified as build `20260828l`. It retires the broad Greek Sicily/southern Italy wrapper and authors four active v5 successors: Sybaris, the disputed literary tradition of Siris/Polieium, Himera, and Metapontion. Croton remains active but unrendered as one of the two family evidence holds. The one review narrowed Siris and Himera, added explicit post-freeze atlas-authoring authorities for every rendered wrapper roster, made held-successor exclusion fail closed, and attached machine-readable Croton source locators. The old URL preserves a neutral four-link notice plus the Croton hold explanation. All 246 local and live frames pass at desktop and mobile sizes; local, staged, installed, and live files share checksum `1ce147865fb89bf46d13771bd639358702672b1de788f4d9e4bf662100f6b5b8`. See `v5/ATLAS_AUTHORING_BATCH_09_REVIEW.md`, `v5/ATLAS_AUTHORING_BATCH_09_INDEPENDENT_REVIEW.md`, and `v5/deployment_20260828l_atlas_authoring_batch_09.json`. This batch is closed. - Atlas authoring Batch 08 is independently reviewed, corrected, deployed, and checksum/browser verified as build `20260828k`. It retires the broad Han Hexi/Western Regions wrapper and authors the two active v5 successors separately. Hexi uses one 121 BCE–220 CE stop and a public source-first family override to Conquest and incorporation because durable colonization is established but resident replacement is not. The Western Regions case begins at the documented 101 BCE forced-migration order and leaves the program endpoint unresolved. Literal Chinese-heartland/North China Plain origins and full routes fit their reviewed regional frames. The old URL preserves a neutral two-link notice. Frozen family, provenance, and role corrections are queued in `v5/HAN_RELEASE_METADATA_CORRECTIONS.json` without mutating v5. All 243 local and live frames pass at desktop and mobile sizes; local, staged, installed, and live files share checksum `e1497ca50c358e6c2c26154971ab1af071f46ad59e93e5791249d1405ae6e9d8`. See `v5/ATLAS_AUTHORING_BATCH_08_REVIEW.md`, `v5/ATLAS_AUTHORING_BATCH_08_INDEPENDENT_REVIEW.md`, and `v5/deployment_20260828k_atlas_authoring_batch_08.json`. This batch is closed. - Atlas authoring Batch 07 is independently reviewed, deployed, and checksum/ browser verified as build `20260828j`. It retires the broad Cape Khoikhoi/San wrapper and authors the two active v5 successors separately: Khoikhoi land-and-labor dispossession, 1652–1806, and San commando clearance, c. 1700–1806. The old URL receives a neutral two-link notice. The independent review found no blocker and corrected one material sibling-source leak plus three minor citation, chronology, and reader-language issues. Public F8 is zero for both successors; the Khoikhoi referent defect is corrected only in public display and queued for the next statistical release. Frozen v5, weights, factor results, and the 63-file public data package are unchanged. All 242 local and live frames pass at desktop and mobile sizes. Local, staged, installed, and live files are byte-identical; the prior build is preserved for rollback. See `v5/ATLAS_AUTHORING_BATCH_07_REVIEW.md` and `v5/ATLAS_AUTHORING_BATCH_07_INDEPENDENT_REVIEW.md`; the authoritative live record is `v5/deployment_20260828j_atlas_authoring_batch_07.json`. This batch is closed. - Atlas authoring Batch 06 is deployed and checksum/browser verified as build `20260828i`. It replaces the retired broad Ngoni Malawi/Zambia wrapper one- for-one with the active Mbelwa northern-Malawi successor. Source review supports one public stop at `c. 1855` and a disclosed interval of `c. 1855–early 1870s`; the frozen release's inherited `1858` model coordinate remains immutable and is not printed as historical fact. The mapped sequence is bounded to Ng’onga/Henga Valley → Rukuru–Kasitu–Ruzi confluence → Choma; the Mpherembe junction remains a distinct semi-independent faction in the narrative without an invented route. One implementation review produced no critical or high issue and its five bounded corrections were applied once. All 241 frames pass locally and live at desktop and mobile sizes. See `v5/ATLAS_AUTHORING_BATCH_06_REVIEW.md`, `v5/NGONI_PUBLIC_CHRONOLOGY_RULING.json`, and `v5/deployment_20260828i_atlas_authoring_batch_06.json`. This batch is closed; it changes no statistical row, weight, factor result, or public data file. - Atlas authoring Batch 05 is deployed and checksum/browser verified as build `20260827h`. It replaces the broad Marshall Islands nuclear-displacement wrapper with four active community-bounded chains for Bikini, Enewetak, Enjebi, and Rongelap/Ailinginae. Community counts, destinations, returns, blocked returns, and land outcomes remain separate; Utirik and the temporary 1946 Rongelap move were reviewed but not invented as successors. See `v5/ATLAS_AUTHORING_BATCH_05_REVIEW.md` and `v5/deployment_20260827h_atlas_authoring_batch_05.json`. This batch is closed. - Atlas authoring Batch 04 is deployed and checksum/browser verified as build `20260827g`. It replaces the broad Ottoman Circassian settlement wrapper with four active source-bounded successors: the Danube/Dobruja agricultural tract, Vidin–Niš–Sofia frontier tract, Uzunyayla, and Wadi al-Sir. One historical review passed with required narrowing and one implementation audit passed with no findings. Vilayet-wide figures are not assigned to local branches; documented local burdens are not generalized into resident removal; refugee origins are separated from Russian/Cossack perpetrators; and Wadi al-Sir's founding groups, immediate route, predecessor population, and state-command language remain explicitly bounded. All 238 frames pass locally and live at desktop and mobile sizes. See `v5/ATLAS_AUTHORING_BATCH_04_REVIEW.md` and `v5/deployment_20260827g_atlas_authoring_batch_04.json`. This batch is closed; its next-release metadata corrections remain queued without changing the frozen release. - Atlas authoring Batch 03 is deployed and checksum/browser verified as build `20260827f`. It retires the broad Saint-Pierre and Miquelon deportation wrapper in favor of the active 1778–1783 and 1793–1815 successors. One independent review's five concrete findings are resolved without changing the frozen statistical release: the 1783 language is limited to restoration and return permission, actor and affected-population roles are separated, sources are normalized, removal routes use documented receiving regions, and the already-adopted public F8 rule displays N/A and zero rather than the frozen face-invalid labor values. All 235 frames pass locally and live at desktop and mobile sizes. See `v5/ATLAS_AUTHORING_BATCH_03_REVIEW.md` and `v5/deployment_20260827f_atlas_authoring_batch_03.json`. This batch is closed. - Atlas authoring Batch 02 is deployed, source-reviewed, and checksum/browser- verified as build `20260827e`. It replaces `mena-jewish-departure-egypt` with the bounded 1956–1957 and 1967–1971 successors and adds a one-to-many retirement notice linking to both. The historical boundaries and reader-facing qualifications are settled. The finite F7 source review and corpus-wide F8 face-validity review are complete. See `v5/ATLAS_AUTHORING_BATCH_02_REVIEW.md` and `v5/eight_trait_card_f8_review/F8_IMPORTED_UNFREE_REVIEW_v2.md`. Neither review reopens event inclusion, dates, or the statistical factor model. The deployment record is `v5/deployment_20260827e_atlas_authoring_batch_02.json`. - The F8 checkpoint was a public-card issue, not an event-roster loop. Public F8 is now the applicability-gated **Imported unfree labor** construct. Its reproducible comparison changes 541 of 676 cards by at least five points; all eight coding conflicts are adjudicated and none remains unresolved. One implementation audit was the stop rule: all 242 frames pass at desktop and mobile sizes with zero issues. No recursive redesign is authorized. - Atlas authoring batch 01 is deployed and live as build `20260827d`. It replaces the broad Libya, Turkey, and post-1967 Palestinian wrappers with three active bounded v5 successors while preserving 223 cases and 233 stops. The exact crosswalk is now 207 active events rendered and 469 not rendered; 467 of the latter have cleared primary families and two retain family evidence holds. Sixteen visible atlas cases remain outside active v5: ten parent wrappers, two Roman wrappers, and the four separately listed legacy/inactive cases. See `v5/ATLAS_AUTHORING_BATCH_01.md`. - The qualified immutable release `second_pass_v5_676` is active and publicly available in publication build `20260827c`. It contains 676 active events, 748 corpus rows including 72 inactive audit rows, 60 dimensions, and 40,560 score cells. The bounded fairness claim and final Editorial Suite are closed for this release. No candidate-inclusion or model rerun is waiting on atlas authoring. - The current live release-to-atlas crosswalk has 265 active events rendered and 411 not rendered. Of those 411, 409 have cleared primary families and two retain explicit family evidence holds: Croton and the partial Paeonian return around 498 BCE. Those two may receive evidence-hold pages or remain unrendered, but must not receive invented family labels. Three of the 409 family-cleared rows are separately blocked at the authoring layer: Aydın, Menteşe, and Ottoman Bithynia. They remain in the statistical release but cannot enter playback until their recorded date, event-unit, F8, and map-evidence defects are resolved. - The 411-event authoring backlog is not a new research census. Its machine-readable roster is the release `events_enriched.json`, where `on_atlas` is false. Each case still needs bounded reader-facing prose, source-supported route and land-outcome geometry, chronology, current family and trait-card attachment, overlap checks, and map/browser QA. These are content-production obligations, not another corpus-wide statistical gate. - Before Batch 01, nineteen of the atlas's 223 visible cases were not active v5 events. The current live count is three: the temporary Roman Italy authoring wrapper `med-roman-republican-colonies-italy` and the two legacy/inactive cases `canary-islands` and `israeli-settlements-opt`. None may be counted as an active event; each needs an explicit redirect, retirement notice, non-counted context treatment, or removal decision. - The remaining Italy wrapper notice remains correct until its current active descendants are authored and QA'd. It resolves to 58 active v5 leaf events, of which 28 are now public and 30 remain. The original 57-unit Italy roster contains two units that were themselves split during the v5 event-unit pass, which is why the current active-leaf count is 58 rather than 57. - The next public-content work should proceed in bounded authoring batches. No frozen-v5 retired parent remains visible; the Roman Italy broad-wrapper notice is a special temporary authoring case and remains until all 58 active bounded descendants are authored and QA'd. The Po Valley wrapper is already retired from playback and survives only as a neutral stable-URL index. The mechanism/audit presentation cleanup is complete since build `20260828o` and remains present in `20260828t`. The older map-origin, zoom, context-label, phase-condensation, caption-alignment, and `#endstate` queue was already implemented in build `20260823a` and remains present in `20260828p`; it has no residual items. See `../../audit/visual_qa/VISUAL_QA_RESIDUAL_RECONCILIATION_2026-08-28.md`. Future map work belongs to ordinary case authoring or a newly observed, case-specific defect; neither changes the active statistical release by itself. ## Historical pre-promotion finite checkpoint — 2026-08-27 This section records the immediately preceding gate and is superseded by the current post-promotion checkpoint above. - The corrected 580-parent/676-successor event-unit model battery is complete, reproducible, and robust to every single-event deletion. No single event determines the ninth retained direction or moves the nine-dimensional subspace beyond five degrees. Whole-program, date-proxy, weighting, imputation, documentation, and source-resolution sensitivities remain material, so the result does not prove every boundary or factor count uniquely correct. - The mechanism/family lane has reached its terminal stop rule. The audit-corrected pre-repair model at `f41311a` found the two levels related but not redundant, but the public layer depended on independent coding reliability. After one source-first repair and one blinded retest, micro kappa was 0.732, positive-label Dice 0.790, and tag Jaccard 0.653, below the predeclared 0.80 floor on all three. The binding disposition is `FAIL_RELIABILITY_GATE_MECHANISMS_INTERNAL`. - No second mechanism repair, retest, or repaired-tag model rerun is authorized. Normalized mechanism tags remain internal. The public analytical direction for this release is one primary outcome family plus evidence-aware trait bars; legacy free-text Mechanism copy should lose analytical prominence in the next UI cleanup while its prose and ledgers remain as provenance. - The remaining statistical/content decision path is now finite: synchronize the monograph and result records; run one final Editorial Suite and statistical claim/reproduction check; record an explicit candidate promote-or-hold decision; and, only if promoted, build a new immutable release and begin downstream atlas authoring, wrapper-notice retirement, public-data refresh, and deployment. - The current public state remains the immutable `second_pass_v4_583` release, atlas build `20260824a`, and publication build `20260827b`. The 676-event candidate is complete diagnostic work, not an active release. ## Superseding content/statistical gate — 2026-08-26 10:17 MDT - The 689-row/245-relationship event-unit register, full source verification, 147-finding main adjudication, and non-destructive verified register are complete. They are no longer open review loops. - All 88 previously named prospective endpoints have completed one bounded source review: 79 retain their proposed boundary and nine have revised identities. None was activated by review. - All 58 immediate descriptions produced from the remaining target triage have completed one bounded review: 48 are source-supported as stated and ten have revised identities. The combined named-plus-defined register now contains 146 unique reviewed targets—127 supported and nineteen revised—and reproduces at `7ab07e1`; none was activated by source review. The eleven parents that lacked finite child rosters completed their three one-pass research packets at `4b1a819`: eleven parent records yield 87 bounded target records, with 72 residual branches explicitly held or excluded rather than converted into automatic follow-on searches. The frozen normalized admission input now contains 233 unique targets: the 146 reviewed targets plus those 87 roster-derived targets. - Measurement compatibility is resolved for the thirty changed active events: seventeen complete profiles are exact reuses, while thirteen events require 364 freshly coded cells. The v1 pilot is preserved but ineligible. The v2 source-complete packets contain thirteen events, 364 cells, and 43 source records. Their one source-sufficiency review found two concrete defects and no broader failure: one demographic-tempo mapping is now an explicit gap, and one intermittently inaccessible official parliamentary PDF now has a hashed provenance-preserving repository record. The original fail and bounded repair are preserved at `e88e32b`; the same reviewer passed the repair check at `00c7c7b` with all 43 source records openable and zero remaining findings. The ten frozen blind-coder launches are now unblocked. Seven returns are main-validated through `12d0f44`: complete pairs for packets 01–03, covering 220 unique cells, plus packet 04 coder A. Packet 04 coder B and both packet 05 coders are active; all ten launches have occurred. The final three outputs still require normal and optimized main validation and disagreement-only reconciliation before any score is applied. - The target admission/scoring rule is explicit at `a718931`, and its one 233-row governed ledger is reproducible at `196ee95`. Main completed and validated all 233 dispositions through `dcef738`: 158 `ELIGIBLE_PENDING_SCORE`, 38 `EVIDENCE_HOLD`, 28 `SCOPE_EXCLUSION`, and nine `UNRESOLVED_OVERLAP`; zero remain pending. Time is not a universal split threshold, source support alone does not activate a target, and all publication, model, atlas, release, and deployment authority remains false. - The monograph and its fairness scaffold are synchronized at `7941ada`. The final event-by-event fairness claim remains pending completion of the target and scoring lanes, candidate assembly, and the predeclared factor, family, missingness, weighting, grouped-program, parent/successor, era, region, mechanism, documentation, duration, and individual-influence sensitivities. - No current result authorizes target activation, score application, public data refresh, wrapper-notice removal, atlas content change, release, or deployment. ## Superseding live gate — 2026-08-26 03:35 MDT - The five source/provenance lanes are closed under the proportionate assurance standard and their certified evidence has been normalized. The 5,820-cell corrected execution roster is complete and committed; no earlier raw-score row is being silently overwritten. - The 334-cell missingness-only review is complete: 309 ordinary unknowns and 25 structural non-applicabilities. All 500 fresh score-blind cells are complete from independent coder pairs and main-validated. There are 374 exact structured agreements and 126 disagreements. Exact agreements are locked; all 126 disagreements received one evidence-grounded, no-averaging adjudication. The resulting 5,820-cell corrected ledger passes independent validation with 374 changed cells. - The isolated 675/671-event candidates and immediate Hellenistic boundary/model sensitivities are complete, reproducible, and independently validated. Equal-event, successor-only, and grouped-program scenarios retain nine factors; collapsing the entire tranche to one median restores eight; weighting and imputation choices span seven to nine. No single corrected event triggers the influence threshold, while the corrected 92 have median 34.5 missing cells of 60. These are diagnostic results, not promotion. - The corrected Hellenistic run is not the complete fairness claim. The frozen v4 event-unit denominator contains 592 review units and 98 known relationships. The final candidate register must append all 97 Hellenistic review units—92 model-eligible and five audit-only—for an expected 689 rows, without duplicating the four v4 parents used in overlap sensitivity tests. - Event-unit review now follows `event_unit_fairness_v1/STREAMLINED_EXECUTION_PLAN.md`: two independent reviewers on the frozen 147-row calibration sample (all 83 known boundary rows plus 64 controls), one reconciliation, one full 689-row application, and one independent check of every exceptional decision plus a stratified retain sample. Reviewer-level nested manifests and Editorial Suites are retired; the merged register, results, and monograph claims receive one final Suite. - Both calibration reviewers are complete and independently validated. They agree on 119/147 primary event dispositions (kappa 0.563), all 24 matched- pair bias and symmetry judgments, and all eleven relationship gaps. The one proportionate reconciliation resolved 33 rows: 28 disposition conflicts and five exceptional related-ID conflicts. The operational rule and validation are frozen at `67080bb`; diagnostic-grade differences remain visible but do not create recursive row audits. - The reproducible 689-row denominator is now frozen: 675 active, 14 audit- only, and 140 recorded relationships. Four cases with no defensible complete numeric bound retain blank evidentiary dates. - The one model-blind application to the remaining 542 rows is complete and validated. The merged 689-row register—675 active and fourteen audit-only rows—plus 245 recorded relationships is frozen and reproducible at `a7f33dd`; every downstream authority remains false. - The one bounded independent verification packet contains 333 event checks and all 245 relationships. Its qualified source-level lane has completed all 333 event rows and no relationships at checkpoint `b62e9b5`: 225 `CONFIRM`, eighty `REVISE`, twenty-eight fail-closed `HOLD_FOR_MAIN`, and 112 recorded findings. Three hundred three rows use direct source passages, three locator-bearing detailed briefs, and twenty-seven insufficient-evidence holds. Relationship rows 1–12, 13–24, 25–36, and 37–48 are the current disjoint read-only batches; row 49 onward remains untouched. - The latest decisions retain Las Humanas with a `by 1660–1672` date repair, hold the combined Quarai/Chililí-to-Tajique identity pending direct support for Chililí participation, and confirm the inactive Salinas wrapper and initial-colonization West Greenland unit. Nine Greek cases confirm their current boundaries, including the already-pending Samos and Aegina splits. Megara Hyblaea requires only an approximate-date correction; Sybaris requires a city/site scope limit and visible partial-return aftermath. Hestiaea confirms its hold, and Scione confirms its pending split. Gaza through Sambus confirm as bounded campaign or city transitions while conflicting ancient scales and outcomes remain attributed and constrained. Ptolemy VIII's Alexandria fails closed; Astakos/Nikomedeia has an unsupported split finding; and the Troad split requires only a coercion-wording repair. Durine/Charax and Hasmonean Gaza need approximate-date repairs; Pella needs a main-mound footprint limit and explicit Josephus source conflict. Kassandreia and early Lysimacheia fail closed for unproven movement; Marion/Paphos retains only its securely attested core, and later Arsinoe is uncertain non-counting aftermath. The final Macedonian rows add no finding; Thebes confirms its pending split. No proposed child has been activated and no relationship, model output, or atlas state has been changed. - The bounded-search amendment is committed and manifest-bound. The reusable relationship checkpoint recorder's one read-only review found five concrete defects; all five are repaired at `189f688`. Its saved finite regression suite passes in normal and optimized Python across positive, idempotent, and all identified mutation paths. Recorder review is therefore closed rather than recursively repeated. - Independent source verification is complete at `a13cf58`: all 333 event rows and all 245 relationship rows are source-checked and pass the complete package validator in normal and optimized modes. The finite return has 147 findings: 108 event, 35 relationship, and four sample-coverage findings. Nothing in this lane yet authorizes a count, score, release, atlas, or deployment change. - The one main adjudication is complete at `12bb6c2`: 99 verifier revisions and two relationship removals are accepted, 42 evidence holds remain explicit, and four inaccessible scale observations remain `UNKNOWN` sample limitations. No prospective successor has been activated and all downstream authority remains false. - The verified register is frozen and independently validated at `9b22d64`. Its effective decisions are 518 active retains, 85 holds, 65 pending splits, nine audit-only rows, eight inactive parents, and four exclusions. All 245 original relationship rows remain audit-visible, including two effective removals and fourteen non-counting holds. - The target-readiness census is complete: 88 named prospective endpoints and 38 split/retired parents with no bounded target, for 126 obligations total. The named endpoints are divided among twelve immutable bounded source-review packets. All twelve batches are complete and validated through `f51459b`: 79 boundaries are supported and nine identities are revised; none is held, rejected, or deduplicated. Source review can support, revise, hold, reject, or deduplicate a target but cannot activate it or authorize a score or model membership. - The 38 unnamed obligations are now triaged once at `6468321`: nineteen parents have complete packet-ready descriptions, eight have packet-ready units plus explicitly held residual branches, and eleven require a bounded roster search. The first two lanes yield 58 immediate source-review descriptions. None is yet an event ID, active row, score, or model member. - The 58 descriptions are frozen into eight parent-preserving source-review packets at `cf336a7` (8/8/8/8/6/8/6/6). Their provisional IDs do not collide with the earlier 88 named endpoints, and the ordinary review may revise, hold, reject, or deduplicate them. Across completed batches 01 through 05, 33 boundaries are supported and five identities are revised across 38 rows. These include St Helena's 1868 operational close, a commission-only Sipontum 186 identity, a 642 Maghreb start, and Egypt's 1967–1971 detention/departure sequence. Batches 06, 07, and 08 are active; no defined batch remains unlaunched. None grants activation or downstream authority. - The eleven incomplete rosters are frozen into three bounded research packets at `a84688f` (4/4/3). Each receives one cited-source pass plus no more than four targeted authoritative additions, then stops with a finite roster, partial roster plus named holds, complete hold, or contradiction returned to main. These packets are queued and grant no downstream authority. - Thirty accepted source/adjudication findings retain active model status and received an explicit score-compatibility decision at `d41452f`: 17 exact profile reuses, eight targeted recodes totaling 64 named dimensions, five full recodes totaling 300 dimensions, and zero score holds. The 364 recode cells are the complete immediate score-work universe; untouched cells remain locked and materially rescoped events receive full profiles. - The five v1 score-blind packets are frozen at `2666521`, but packet 01 exposed a substantive packet-design defect: they include corrected boundary summaries and named locators without enough independently openable source bodies to score the traits. The 66/74 exact agreements and eight disagreements at `bad8873` are preserved as an evidence-insufficient pilot, not eligible measurements. Packet 02's coders were stopped before either wrote an output. The versioned v2 lane has now completed and main-reviewed source inventories for all thirteen events. Its five-event full-recode dossier fragment passes the exact schema and preserves sixteen source bodies, all eight domains per event, explicit gaps, and false score and downstream-authority flags. The eight-event targeted fragment also passes, covering 27 source records, all 64 target dimensions, and 24 explicit gaps. The reproducible merged set contains thirteen events, 364 cells, and 43 source records; normal and optimized mechanical validation find no inherited-result key and no downstream authority. A pre-coding check found the old output validator could not consume the v2 packet shape; the first sufficiency attempt was stopped before output, and the packets were rebuilt against a v2-aware, source-ID-enforcing validator at `6099598`. Its one fresh independent source-sufficiency review is active. It then receives one coder pair and one no-averaging adjudication—no recursive audit. - Exact remaining statistical/content sequence: apply the adjudication once to a versioned verified register (complete); finish the defined-target and bounded-roster source reviews; finish the source-complete v2 repair, independently code, and reconcile the 364 authorized recode cells; freeze explicit model memberships; rerun the preregistered models and sensitivities; write the measured fairness result and residual uncertainty into monograph v2 (whose working fairness section was updated at `c13bb18`); run one final Editorial Suite; then decide promotion and public atlas authoring. - Roman and Po Valley wrapper notices remain a public-authoring issue after statistical promotion: their bounded successors still need complete maps, phases, overlap QA, and release QA before the notices can be removed. ## Current substantive gate — 2026-08-25 - Craig approved the proportionate assurance standard. Packaging-only defects no longer trigger automatic historical/source replays; substantive defects still receive bounded repair, one independent verification, and main adjudication. - Provenance and Batch D now have final main adjudications preserved at `0a87f51fa7e9cf0e835d3d74788c626a5e5bdb98`. Provenance passes with `CAL-010` and `CAL-022` still mandatory and uncleared. Batch D passes its historical/source burden while retaining two process-only incidents. - Batch B has one fresh independent verification in progress. Batch C has a two-finding focused repair in progress, after which one independent focused verification is required. Batch A is undergoing semantic-preserving shared- contract assessment instead of an automatic full substantive replay. - Three main adjudications remain (A, B, C), and all five legacy machine-gate disposition files remain absent until the shared contract is validated. No certified evidence has entered the recode or models. - Once all five lanes pass the shared gate, the next substantive work is still: integrate only certified evidence; apply and independently verify the recodes; rerun factor/family, missingness, and declared split/merge sensitivities; complete the multidimensional event-unit fairness result; and only then promote statistical and atlas content. - The Roman and Po Valley wrapper notices remain an authoring/release issue, not an unresolved statistical split. Their 57 and 15 bounded successors must be mapped, phased, overlap-checked, and QAed before the notices can disappear. ## Current certification boundary — 21:03 MDT - Batch B v10 and main's exact-preservation intake are immutable at `ba1a5c2…`. Main independently reproduced both bounded assurance repairs, the full four-run control matrix, all eight `_permitted` authority controls, all nineteen Calibration claims and 71 locators, target identity, and zero package or target writes. A genuinely fresh B review is active under the commission committed at `5a69974…`; preservation is not a B disposition. - Four exclusive lanes remain active: the bounded Batch D v9 integrating review, the genuinely fresh Batch B v10 review, the provenance v8 bounded-repair integrating review, and the bounded Batch C v12 citation/manifestation repair. Their partial files remain unstaged and confer no authority. - Batch C v11's original handoff remains exactly preserved at `5ec77ee…`. Its genuinely fresh `FAIL_REPAIR_REQUIRED` review and main intake are now exactly preserved at `5773379…` after independent normal/optimized replay and direct source reopening. The bounded defects are two wrong locators, two inexact short quotations, the wrong Loeb volume/year for four Pausanias rows, the resulting unsupported rights rationale, and mechanical dependencies. The v12 commission is immutable at `54c5969…`; repair agent `01a036de-b79c-7a31-9ca6-eacd1c768c20` owns only `batch_c_v12/`. No C disposition or content/statistical authority follows. - Provenance v11's fresh review and main intake are now exactly preserved at `af1b945…`. Main independently reproduced the narrow assurance pass but also confirmed five open provenance findings, two re-citation flags, nine no-action boundaries, and no historical-truth certification. The first bounded integrating reviewer stopped on an automated safety screen before substantive output; its ten temporary Suite files are non-evidence. Clean retry1 is active under agent `01a036e1-9a06-77d3-9b4a-81c959920de1` and commission `7bce9d9…`, with the original `9dfb63e…` scope still binding. No provenance disposition or downstream authority follows. - Batch A v15 retry1 has returned a 60-file bounded-repair handoff that self-reports `PASS_BOUNDED_PACKAGE_ASSURANCE` and repairs only `A14-MR-001`–`003`. It is untracked and unauthenticated; no A pass or fresh audit commission may be inferred before main's exact preservation intake. - The clean Batch D v12 retry1 review and main intake are immutable at `59835ad…`. They preserve a sound review pass but keep target finding `BDV9-F-001` open. The active bounded-repair integrating review may define only a repair contract; it may not repair, integrate, or create a D disposition. - The complete root 27-gate last passed after `8e90993…` under normal and optimized Python with byte-identical log SHA-256 `fcf1b1d1…`: the certified set was empty, exactly five dispositions were absent, and all downstream authority was false. It passed again after state commit `70c16479…` with the same full-output hash and no tracked-worktree change, preserving that exact fail-closed boundary after the Batch C and provenance lane transitions. - The complete restart state, exact identities, protected paths, and next action are in `AUDIT_ACTIVE_STATE.md`. Five final dispositions remain absent; no event, score, family, roster, model, fairness, release, atlas, publication, or deployment conclusion can advance yet. ## Superseded certification boundary — 18:12 MDT - The remaining five-lane work is chiefly source/process certification, not a fresh vote on hundreds of already admitted events. Batch C v11, provenance v11, the fresh Batch B v9 main review, and bounded Batch A v15 are active. Batch D v12 has returned a formal handoff and is undergoing its required genuinely fresh main review; neither the handoff nor its self-reported tests count as a pass before that review closes. - Batch B v8's post-commit replay defect and provenance v10's scratch-name recursion are real but bounded assurance findings. Neither changes an event, source conclusion, trait score, family assignment, candidate, or model row. Batch B v9's 76-file handoff is preserved at `3b6bc810…`; main independently reproduced its normal/optimized builder, validator, invariant, and external- authentication passes without changing a byte. The owner did not return a formal terminal message and was closed while still marked `running`, so this is preservation—not a pass. The fresh-review commission is saved under `main_reviews/batch_b_v9_commission/`. - Batch A v14 plus its fresh review is preserved at `d0a3380…` with `FAIL_REPAIR_REQUIRED`. The three accepted findings are assurance defects: incomplete authority semantics, self-attesting/incomplete Editorial Suite grounding, and a final-input/canonical-byte reconstruction mismatch. Main independently reproduced all three authority bypasses in both interpreter modes and directly confirmed that `NEGATIVE_CONTROLS.json` captured an older restart-ledger identity than the final distributed ledger. The bounded v15 commission changes no historical/source/event/statistical judgment. - Batch C v10's fresh source audit and main review are preserved at `c2d860b`. They accept a real retained-evidence gap: five of 49 propositions are fully licensed by the displayed exact passage, thirteen need adjacent evidence, thirty-one retain no exact passage, and 107 of 115 relation judgments inherit unresolved proposition grounding. This does not mean the 44 propositions are false. It means the Hellenistic lane cannot be certified until a bounded v11 successor supplies or calibrates the missing evidence and is freshly audited. - Batch D v11 and the canonical main review are preserved at `dbe1b376…`. The prior recursive scratch-copy defect is repaired, but unequal-depth repository-local builds independently reproduce a new bounded packaging failure: eight portable-command receipts encode different relative `--repo` chains, changing the receipt, manifest, and seal. The v12 commission at `0411fe28…` repairs only that path-depth dependence. Its returned package has 44 files / 364,888 bytes and framed SHA-256 `ef509bfa…`; a fresh fourth-depth main review is active. It does not change the open `BDV9-F-001` target finding or any content/statistical judgment. - All five final A/B/C/D/provenance dispositions remain absent. Therefore no pending addition, retirement, split, merge, wrapper replacement, score change, or public-data refresh is statistically final. The certified-package set must remain empty and all downstream authority must remain false. - The next content/statistical authority can arise only after the five final dispositions, certified-evidence integration, independent audit of the 1,700-cell recode workset, factor/family and missingness reruns, all declared event-boundary sensitivities, and the event-unit fairness analysis. - Roman and Po Valley wrapper notices are not waiting on more statistical splitting: their 57 and 15 bounded successors are already defined. The notices remain until those successors are individually mapped, phased, authored, overlap-checked, and release-QAed. Statistical completion alone is not enough to deprecate them. - The complete root 27-gate passed without writes under normal and optimized Python after current restart-state commit `88a85e7`. The normal and optimized logs are byte-identical at SHA-256 `fcf1b1d1…`. It reports an empty certified set, exactly five missing dispositions, and all downstream authority false. ## Current denominators - Active v4 release: 583 active events, 591 total corpus rows, eight inactive audit rows, sixty dimensions, and 34,980 typed cells. - Atlas build `20260824a`: 223 cases and 233 playback stops. The stop reduction consolidated continuous narrated phases and changed no statistical event, family, trait value, score, or weight. It is deployed and checksum-verified; the public working monograph and current immutable-v4 data snapshot are also linked from the atlas. - The v4 roster change is settled: four active rows retire and four bounded rows enter. The matching counts are not a target. ## Closed in v4 - Medieval Norse Greenland, the broad Athabaskan migration, uninhabited Bermuda/Cayman settlement, and the Spanish northern-borderlands wrapper are inactive audit rows. - Sobaipuri is a partial Spanish-ordered relocation, not total San Pedro clearance. - The district-wide Salinas case is retired. Las Humanas and Quarai/Chililí-to-Tajique are separate active events; Abó is held. - West Greenland is admitted as the bounded 1721–1782 mission-and-trade colonization, not a medieval continuity claim or a 1721–1953 super-event. - The complete v4 statistical and atlas sensitivity suites have run. The final parallel-analysis result is eight factors, with the ninth just below its Horn threshold. ## Immediate event-level holds 1. **Abó, 1673.** Vacancy and a probable movement toward Piro-speaking kin are documented. The missing link is evidence that an outside actor coerced the residents' departure. It remains out of the model until that link is found. 2. ~~**Tibetan residential-school separation.**~~ **Resolved 2026-08-21:** retire from the next model while preserving the row and sources. Forced institutional child separation does not meet the territorial-removal rule. 3. ~~**Svalbard wartime evacuation.**~~ **Resolved 2026-08-21:** retire from the next model while preserving it as a boundary example. The temporary, coordinated protective evacuation is outside scope. 4. ~~**Falklands/Malvinas.**~~ **Resolved 2026-08-21:** retain because political land-taking and removal of the garrison and officials qualify, while stating prominently that most civilians remained. Re-audit affected scores before the next release. 5. **Nauru phosphate land destruction.** **Held 2026-08-25:** severe colonial mining damage and the island/population scale are source-supported, but the German, Australian-led, Japanese-occupation, and postwar regime sequence and the labor-program scope are not yet passage-verified. The active row is suspended pending that bounded source repair. ## Active wrappers that still need bounded review These rows retain statistical weight for now but have not earned a public event-unit presumption merely by appearing in v4: - Indo-Aryan migration; - Cabo Verde colonial population formation; - São Tomé plantation colonization and slavery; - Mascarene/Seychelles forced population formation; - Asian indenture to Suriname; - British Malaya plantation labor migration; - St Helena colonial population formation; - Norfolk penal transportation and Pitcairn resettlement; and - Cocos/Christmas Island colonial labor settlement. The island wrappers are especially likely to split or retire because destination settlement on uninhabited land is outside scope even where a bounded source-zone slave, penal, or labor removal may qualify. ## Qualifying cores that need splitting or reframing - ~~Fiji plantation settlement, blackbirding, and Indian indenture~~ — **mixed-parent decision resolved 2026-08-25:** retire the wrapper. Advance only the bounded October 1864–10 October 1874 coerced/deceptive Pacific recruitment child through its remaining blind-coding gates. Keep the post-Cession Pacific stream, Indian indenture, and pre-Cession settler-land proposals on separate evidence holds; - Persian Gulf slavery, debt bondage, escape, and manumission; - Hong Kong/Macau wartime influx, occupation, removals, and repatriation; - Channel Islands civilian evacuation and Nazi punitive deportations; and - Spanish conquest, reducciones, land transfer, and coerced labor in the Andes. The 2026-08-21 first adversarial pass resolves these into a strict ten-unit successor set while retaining evidence holds around additional possible units. The exact decisions and IDs are in `v5/ROSTER_AND_SPLIT_DECISIONS.md`. Splitting can change the denominator. No child is admitted automatically, and no parent's sixty ordinal values may be arithmetically divided among children. ## Family or wording review without a current removal vote - Sri Lankan irrigation settlements; - Cahokian colonies; - Wari and Tiwanaku colonies in Moquegua; - Bioko plantation colonization; - Iraq Marsh Arab drainage and displacement; and - Zimbabwe's Operation Murambatsvina. These remain content or family-label reviews on the present record, not reasons to remove otherwise qualifying events. ## Uncounted candidate pools ### Original gap scan **Re-review completed 2026-08-21; first source-level correction recorded 2026-08-22.** Twelve proposal records represented eleven ideas because two rows described the same Kwajalein/Ebeye history. Seven ideas advanced to source briefs, Chuang Guandong remains held, and three prehistoric proposals are rejected under the frozen evidence bar. The Kwajalein idea survives, but its 1944-present wrapper does not: source review splits it into four separately ordered removal candidates and retracts the claim that 1,470 people were directly cleared. None of the advancing ideas or successor units is active yet. The full positive, negative, and corrected findings are preserved in `gapscan_rereview/` and `v5/gapscan_source_briefs/`. ### Southwest successor search - Antelope Creek remains a hold: succession during drought is not demonstrated displacement. - Jemez captivity is a supporting lead without a bounded community removal and land outcome. - Manso/Suma, Toboso/Lagunero/Concho, named frontier abandonments, and later Lipan/Mescalero movement under Comanche pressure remain search branches. - The Spanish northern-borderlands residual queue includes Acoma, Pueblo revolt and reconquest, California mission corridors, and other bounded successors. ### Ancient and premodern expansion - Ninety-seven prospective Greek/Macedonian/Hellenistic units passed the source-frozen event-unit and overlap gates. Their reproducible, hash-bound briefs, two independent sixty-dimension codings, complete reconciliation, nine family-boundary reviews, sensitivity analyses, correction overlay, and fail-closed 1,700-cell recode workset now live under `v5/`. They remain inactive: the targeted source-certification cycle and recode-application gate have not cleared, and no Hellenistic unit has entered the active model. - RAMÛ access, the wider early Neo-Assyrian sweep, New Kingdom/Nubia splitting, additional Mongol and Sasanian cases, and region-by-region ancient and medieval searches remain open. These are coherent future tranches. They do not reopen the v4 correction. ## Statistical-weight policy 1. One bounded event remains one row. Population size, death toll, moral weight, and editorial prominence do not change row weight. Calendar duration alone neither splits nor merges a row; the event-unit decision follows causal, territorial, actor, mechanism, and outcome continuity at the resolving power of the evidence. 2. Source density is the real weighting hazard. Every major tranche should keep equal-event results beside era-balanced, parent-series, documentation, and missing-data sensitivities. 3. Factor counts and automatic clusters never vote an event in or out. 4. The eight-trait card remains the public analytical lens. The v4 eight-factor result does not prove eight natural kinds, and the six families remain an editorial taxonomy tested against—but not dictated by—the model. 5. Rerun the full sensitivity suite after each coherent large addition tranche, not after isolated wording or map changes. 6. Before the next promoted release and monograph v2, complete the release-wide event-unit fairness audit and alternative-resolution gate in `EVENT_UNIT_FAIRNESS_AND_RESOLUTION_PROTOCOL.md`. Equal event weights are not a fairness result unless the denominator's split/merge decisions survive independent and cross-era review. ## Mechanism/family information and legacy audit presentation The post-model editorial analysis in `MECHANISM_FAMILY_AND_AUDIT_UI_TODO.md` is required before the next public promotion. It must establish how mechanism labels and primary analytical families are separately derived, then test whether they carry meaningfully different information using association, conditional entropy, predictive redundancy, within-label diversity, cross-era stability, and manual review of discordant cases. The outcome determines whether one visible label is removed, both remain, or normalized mechanism tags move onto the trait card. This is not answerable from visual similarity alone and must not be decided before the promoted roster and profiles exist. The same pass should remove repetitive “Worldwide audit” process language from ordinary public case prose once the new release's evidence/status and wrapper history are authoritative. Audit status, confidence, correction text, source locators, and supersession history remain preserved data; only the stale reader-facing wording is a deprecation candidate. ## Next content sequence 1. Finish the qualified source verification of all 333 event checks and 245 relationships, preserving every inaccessible or unsupported claim as a hold rather than filling it from corpus prose. 2. Main-adjudicate the resulting findings once, including the remaining four explicit main-queue questions, and construct only the split/merge alternatives that the evidence identifies. 3. Freeze the corrected 689-row register and rerun equal-event, successor-only, parent/group-balanced, era, region, mechanism, evidence-resolution, missingness, and declared boundary sensitivities. 4. Write the release-wide event-unit fairness result and limitations into the monograph, then run the one final Editorial Suite over the merged register, statistical claims, public data, and prose. 5. After statistical promotion, complete the Abó search, bounded successor briefs/coding, gap-scan admissions, ancient-coverage work, and atlas maps in their recorded queues without restoring retired umbrellas. ## Exact source-certification state at this checkpoint - **Batch A:** v10 audit, integrating review, bounded v7 repair, and genuinely fresh v11 audit are independently preserved through `0ab49cf`. V11 finds four repair-required process/prose defects and one note, with no historical, relation, score, event-boundary, candidate, or family change. It needs a separate integrating review, any authorized bounded repair, another fresh audit and review as required, and a final disposition. - **Batch B:** bounded v6 repair and genuinely fresh v7 audit are independently preserved through `ad42ee4`. The audit has no rejected historical proposition; its high-severity findings concern process self-authentication and honest-reseal acceptance. Its separate integrating review is active in `independent_audits/integrating_reviews/batch_b_v7/`; a final disposition or any review-authorized repair/audit cycle remains downstream of that review. - **Batch C:** bounded v5 replay repair is preserved at `1417348`; genuinely fresh v9 audit reached a sealed on-disk package but is **not complete**. Its agent did not return a terminal handoff, and main's first zero-write final builder check rejected an `AUDIT_INCIDENTS.json` byte mismatch. The package is uncommitted and must return to the same auditor for a bounded finalization/control repair and all four normal/optimized build/validator checks before it can be preserved. It then needs a separate review and a final disposition. The older interrupted C package remains forbidden as input, evidence, staging material, or cleanup target. - **Batch D:** v7 findings-only audit and its independently verified integrating review are preserved through `2b455cb`. The review accepts seven bounded target findings and no score/event consequence; its separately preserved post-seal runner incident is process-only. Bounded RR-001–RR-007 successor repair v4 is active in `batch_d_repair_v4/`; a genuinely fresh audit, another separate review, and a final disposition then remain. - **Provenance:** independently verified bounded v5 repair is preserved at `08cb952`. The fresh v7 auditor returned a 109-file **FAIL_REPAIR_REQUIRED** package with six findings. It independently derives 30 v4-to-v5 scalar changes rather than the target-declared 27 and identifies a machine-ledger claim omitted from the declared Calibration universe. Main reproduced its roster, manifest, content hash, target binding, and all-false authority state, but its validator freezes the entire concurrent worktree's status and no longer passes after later review lanes and commits changed that unrelated status. The package therefore remains uncommitted and uncertified pending a narrowly scoped auditor correction/reseal, followed by a separate review and final disposition. All five disposition files remain absent. The root 27-gate suite passed in both execution modes after commit `9c97fd9`. Its integration preflight remains blocked on all five missing dispositions, with no certified package or downstream authority. Once those five files exist and independently validate, the remaining critical path is: integrate only certified evidence; apply the auditable recode register; rerun the models and the complete sensitivity suite; complete the release-wide event-unit fairness review; promote a versioned statistical release; then author and QA the approved atlas cases. The hundreds of already admitted but unrendered events are an authoring backlog, not unresolved inclusion votes. The detailed closed decisions and statistics are in `SECOND_PASS_V4_DECISIONS.md`; machine-readable authority lives under `v4/`. ## Superseding source-certification checkpoint — 2026-08-24 The earlier per-lane paragraph above is retained as process history. Current state is: - **Batch A:** the v11 audit and separate integrating review are preserved through `5cc0aed`. Four bounded process/prose repairs are confirmed; all 128 relation rows and all historical, hold, score, event, family, candidate, and release conclusions remain unchanged. The bounded successor repair, another genuinely fresh audit/review as required, and a final disposition remain. - **Batch B:** the v7 audit is preserved; its separate integrating review has completed Fact-Check and LLM Artifact Tells and has reached Calibration under agent `01a034ab-6320-74b0-950d-df7181baa43a`. No final disposition exists. - **Batch C:** finalized v9 is preserved at `91f01ce` after main independently reran the normal/optimized builder and validator with zero writes. Its five findings, 24 incidents—including the premature closure and failed main reproduction—and 21 validation records remain explicit. Its separate v9 integrating review is active under agent `01a034e4-ddf6-7252-9755-84fe8446f0fc`; any accepted repair/audit cycle and final disposition remain. - **Batch D:** bounded repair v4 is now preserved at `fcc7b9b`; its normal and optimized builder/validator checks all pass with zero writes. Fresh v8 audit agent `01a034dc-065a-71a0-ad6c-d0cf07fc1f33` is active against that immutable commit in `independent_audits/batch_d_v8/`. A separate review and final disposition remain downstream. - **Provenance:** corrected v7 audit is preserved at `ebd910b`. It is a valid `FAIL_REPAIR_REQUIRED` package with six findings and all authority false. A separate integrating review is active under agent `01a034e1-8a63-7332-9113-abc34e01b67e`; any bounded successor repair, another fresh audit/review as required, and a final disposition remain. After `5cc0aed`, the full 27-gate suite passes in both normal and optimized execution. This is a fail-closed process pass, not source certification: all five disposition files remain absent, the certified set is empty, Gate 26 reports exactly those five blockers, and every downstream authority is false. The content/statistical sequence is unchanged: finish and disposition all five source lanes; integrate only certified evidence; apply the 1,700-cell recode workset; rerun the models and all predeclared sensitivities; complete the release-wide event-unit fairness and alternative-resolution audit; only then promote the next statistical release and author the approved atlas additions. Calendar duration is never the sole split/merge rule, and no automatic factor or cluster result may admit, remove, split, or merge an event. ## Superseding open-loop snapshot — 2026-08-24 12:12 MDT - **Batch B:** its v7 integrating review is now preserved at `9a6f922` and remains `FAIL_REPAIR_REQUIRED`. Historical checking found 59 bounded-accepted and four `Uncheckable` propositions, zero rejections, and all 22 holds still open. Five bounded process/assurance repairs remain before another fresh audit: successor-hash repair, eight-invariant validator coverage, external Git/executable bindings, truthful Editorial Suite scope, and complete adversarial-roster coverage. These do not currently alter any event, score, or family. - **Batch A:** v8 bounded repair is active and may implement only the four requirements accepted in its v11 review. A fresh external audit/review and a final disposition remain afterward. - **Batch C:** the separate v9 integrating review remains active; any accepted bounded repair, fresh audit/review, and final disposition remain afterward. - **Batch D:** the genuinely fresh v8 audit remains active against committed repair `fcc7b9b`; a separate review and final disposition remain afterward. - **Provenance:** the separate v7 integrating review remains active; any accepted repair, fresh audit/review, and final disposition remain afterward. After `9a6f922`, all 27 root checks pass in normal and optimized execution, but this is deliberately not certification: the five final dispositions are still absent, the certified-package set is empty, and all downstream authority is false. The next content/statistical work therefore remains gated in this order: finish the five source lanes; integrate only certified evidence; apply the 1,700-cell recode workset; rerun models and sensitivities; complete the event-unit fairness and alternative-boundary audit; promote a statistical release; then update and deploy the atlas. Already admitted but unrendered events remain an authoring backlog rather than unresolved statistical votes. The protected Potosí brief, legacy A/D v2 directories, interrupted C directory, and typed-metadata pilot files remain outside this work lane and must not be staged, cleaned, or used as evidence. ## Superseding open-loop snapshot — 2026-08-24 12:54 MDT - **Provenance:** v7's separate review is preserved at `0f2120b` and remains `FAIL_REPAIR_REQUIRED`. The historical/event state is unchanged. The open issue is exact process accounting: 30 scalar changes rather than 27, with four omitted changed selectors and one wrongly listed unchanged selector, plus missing Calibration coverage for that load-bearing machine claim. Bounded repair v6 is active and must satisfy eight mechanics/Calibration requirements before a fresh audit/review and final disposition. - **Batch A/C/D:** their v8 repair, v9 review, and v8 audit respectively remain active. **Batch B:** its five-requirement bounded successor repair is queued for the next available agent slot. After `0f2120b`, all 27 root checks pass in both execution modes while remaining correctly fail-closed on the same five absent dispositions. No new evidence, score, event, family, candidate, release, atlas, or deployment authority exists. The statistical/content critical path remains: complete the five lane chains; integrate only certified evidence; apply the 1,700-cell recode workset; rerun models and sensitivities; complete alternative event-boundary fairness tests; then promote the release and author/deploy atlas additions. ## Superseding open-loop snapshot — 2026-08-24 13:06 MDT - **Batch D:** fresh v8 external audit is preserved at `4dbd1e9` and remains `FAIL_REPAIR_REQUIRED`. Its historical/source checks support the 35 source findings and 84 relation labels at bounded scope, while seven process defects remain: execution-evidence authentication, write-path containment, complete seal semantics, broader date and ISBN candidate enumeration, 20 physical Markdown patch coordinates, and four intermediate editorial-hash claims. Separate integrating review is active under agent `01a0352a-f0de-7c70-b863-e042c94a63dd`; any accepted repair, another fresh audit/review as required, and Batch D's final disposition remain downstream. - **Batch A:** bounded repair v8 remains active; a fresh external audit/review and final disposition follow. **Batch C:** v9 separate integrating review remains active; any accepted bounded repair and fresh audit/review follow. **Provenance:** bounded repair v6 remains active; a fresh audit/review and final disposition follow. **Batch B:** its five-requirement successor repair remains queued for the next free slot. After `4dbd1e9`, all 27 root checks pass in normal and optimized modes, but the certified set remains empty and all five dispositions remain absent. Therefore no evidence may yet be integrated, no cell recoded, no model or sensitivity promoted, and no atlas content or deployment may claim these results. Once all five lanes clear, the remaining sequence is: integrate certified evidence; apply the 1,700-cell recode workset; rerun models and predeclared sensitivities; run the release-wide event-unit/alternative-boundary fairness audit; finish the monograph account; promote a versioned statistical release; then author, QA, and deploy approved atlas additions. Protected dirty/untracked material remains outside this lane and untouched. ## Superseding open-loop snapshot — 2026-08-24 13:13 MDT - **Batch C:** its separate v9 integrating review is preserved at `6e51788` and remains `FAIL_REPAIR_REQUIRED`, with three accepted and two narrowed process findings. Full review found no historical/source defect and no current content or statistical consequence. Six bounded successor requirements cover complete authority/UTF-8 scanning, Unicode separator grammar, exact package hygiene and mode/type binding, rights-occurrence metadata scope, package-local positive controls, and portable audit-stage controls. Replay repair v6 is active under agent `01a03530-aa24-7d11-8fb2-0925db5cf7dd`; another genuinely fresh audit, separate review, and final Batch C disposition remain after it. - **Batch A, Batch D, and provenance** remain active in their previously recorded bounded repair, separate-review, and bounded-repair lanes. **Batch B** remains queued for its five-requirement bounded repair. After `6e51788`, all 27 root checks pass in both modes while the five final dispositions remain absent and the certified set remains empty. The open statistical/content sequence is unchanged: finish every lane's repair/audit/ review chain and final disposition; integrate only certified evidence; apply the 1,700-cell recode workset; rerun models and predeclared sensitivities; complete event-unit and alternative-boundary fairness testing; finish the monograph account; promote a versioned release; then author, QA, and deploy approved atlas additions. Protected paths remain untouched and cannot serve as evidence. ## Batch D review-process incident — 2026-08-24 13:19 MDT The first Batch D v8 review process was interrupted by an automated platform filter and returned no formal result. Its partial `integrating_reviews/batch_d_v8/` directory is retained only as a non-evidentiary process record and must not be staged, cleaned, or used. A clean replacement is active under agent `01a03536-7965-7322-9d3e-a9c657253374` in `integrating_reviews/batch_d_v8_retry1/`, bound only to the committed repair and external-audit packages. This changes none of Batch D's seven open review questions and has no content or statistical consequence. Batch D's separate review, any authorized repair/fresh audit, and final disposition all remain open; the overall fail-closed sequence is unchanged. ## Superseding open-loop snapshot — 2026-08-24 13:31 MDT - **Batch A:** bounded repair v8 is preserved at `25b6219` (package subtree `c04332c42f4c95fe45934beebaa2a976c84ae2fe`). Main reproduced its complete normal/optimized and isolated checks with zero writes and verified all 6,374 successor authority values false. The package changes only the permitted archive-local-retention consultation field; all historical/source and relation rows, six `Uncheckable` findings, event boundaries, scores, and public conclusions remain unchanged. Fresh v12 review is active under agent `01a03541-cc85-7df3-8c8f-e8033688f2b7` in `independent_audits/batch_a_v12/`; its review, any genuinely required successor, and Batch A's final disposition remain open. - **Batch C:** six-requirement replay repair v6 remains active under `01a03530-aa24-7d11-8fb2-0925db5cf7dd`; a fresh audit, separate review, and final disposition follow. - **Batch D:** the clean retry1 review remains active under `01a03536-7965-7322-9d3e-a9c657253374`; only its preserved findings may authorize a repair/fresh-audit successor, and its final disposition remains open. - **Provenance:** eight-requirement repair v6 remains active under `01a0351e-33f9-7f12-8014-7571a1aea9a8`; fresh audit/review and disposition follow. - **Batch B:** its five-requirement bounded repair is the next general queued lane; fresh audit/review and disposition remain after it. After `25b6219`, the root suite passes all 27 checks in normal and optimized modes. It still reports an empty certified set, exactly five absent final dispositions, and no downstream authority. The content/statistical critical path is therefore: finish all five complete lane chains; integrate only certified evidence; apply the auditable 1,700-cell workset; rerun models and predeclared sensitivities; test alternative plausible event boundaries and release-wide event-unit fairness; finish the monograph account; promote a versioned release; then author, QA, and deploy the approved atlas backlog. Nothing in the active repair/review work yet adds or removes a public event. Protected dirty, legacy, interrupted, and typed-metadata paths remain excluded from evidence and staging, including the interrupted Batch D v8 review. Active agent directories remain uncommitted until formal handoff and independent main validation. Exact-path staging only; never `git add .`. ## Batch B queue advanced — 2026-08-24 13:38 MDT Batch B's bounded successor is no longer merely queued. Main initialized local repair v7 and wrote its complete bindings, five blocking requirements, no-action boundary, protected paths, and restart procedure to `batch_b_repair_v7/PROGRESS.md` before implementation. It may repair assurance mechanics only; it cannot alter the 63 historical propositions, 117 relations, 22 open holds, critical `Uncheckable` limits, scores, event units, families, candidates, releases, or atlas. Its active directory remains uncommitted until formal handoff and independent validation, after which a different fresh audit and separate review still precede Batch B's final disposition. The five-lane critical path is now fully occupied: Batch A fresh review, Batch C repair, Batch D retry review, provenance repair, and Batch B local repair. The same five final dispositions remain absent, so this scheduling change has no content, statistical, release, or deployment consequence. ## Provenance handoff preserved; no content gate opened — 2026-08-24 Main independently reproduced the complete provenance v6 handoff and preserved its exact 60 files at `d6dd795` (package subtree `5df181a0...`). All eight normal/optimized builder, validator, isolated-replay, and adversarial-control commands passed without canonical writes; the manifest, seal, flat roster, no-action record, 30-selector delta, two machine claims, and six byte-identical domain-state carriers also checked independently. The package remains only `ready_for_fresh_external_audit`: it is not certified and authorizes no event, score, family, model, release, or atlas change. A new-owner fresh provenance v8 audit is active under `01a03558-d68f-7fd2-9c07-8cc888450703` in `independent_audits/provenance_v8/`, bound to exact target commit `d6dd795` and subtree `5df181a0...`. Its separate review and final provenance disposition still follow. Batch B implementation has also begun, but only assurance mechanics are in scope; the 63 propositions, 117 relations, 22 holds, and all content/statistical outcomes remain unchanged. The real content sequence is therefore unchanged: complete all five disposition chains, integrate only certified evidence, apply and audit the recodes, rerun models and plausible event-boundary sensitivities, finish the monograph fairness account, and only then promote and deploy atlas content. ## Batch D review preserved; Batch C at main-validation gate — 2026-08-24 - **Batch D:** the clean retry1 integrating review is preserved in its complete 50-file form at `befc895` (package subtree `672504dd...`). It accepts six findings and narrows one without changing any historical adjudication. Seven bounded assurance repairs—IRR-001 through IRR-007—remain before another fresh audit, separate review, and final D disposition. The incomplete 46-file preservation commit `266ed22` is retained only as a superseded process incident; the exact four omitted Suite references are present in `befc895`. A separate failed rebuild invocation is likewise recorded and had no canonical or package effect. Fail-fast shell execution is now mandatory for multi-step validation and preservation. - **Batch C:** replay repair v6 has formally stopped `ready_for_fresh_external_audit` and its agent is closed. It reports a sealed 99-file package, unchanged 12-row no-action boundary, all thirteen authority fields false, rights text and conclusions unchanged, six retained failed attempts, and only authorized occurrence-scope metadata changes. These are handoff claims, not yet main-certified facts. Independent reproduction, manifest/hygiene checks, exact-path preservation, and a genuinely fresh C audit remain next. - **Batch A and provenance:** their fresh v12 and v8 reviews remain active. **Batch B:** its assurance-only v7 repair remains in local implementation. The five final dispositions are still absent and the certified-package set is still empty. No event may yet be added or removed on this evidence; no recode, model, fairness conclusion, release, atlas authoring, or deployment gate has opened. The remaining content/statistical sequence is unchanged: finish and independently certify all five source lanes; integrate certified evidence; apply and audit the 1,700-cell workset; rerun the models and all predeclared event-boundary sensitivities; complete the event-unit fairness analysis and monograph account; then promote and deploy the resulting atlas content. Batch D's bounded IRR-001–IRR-007 successor is now active under agent `01a03563-4a5f-7ca0-a625-f5f231626b97` in `batch_d_repair_v5/`. Its sole-write scope, no-action boundary, fail-closed checks, and terminal ceiling are recorded on disk. This is assurance work only and changes none of the open content or statistical conclusions above. ## Batch C preserved and under fresh audit — 2026-08-24 Batch C replay repair v6 is no longer an unvalidated handoff. Main reproduced its complete normal/optimized and exhaustive checks, independently verified the 99-file/3,645,848-byte manifest and terminal rosters, all 47,658 authority occurrences false, its twelve no-action rows, six retained process failures, and exact Git bindings, then preserved it at `a0e7e8f` (package subtree `9c58cdbc...`). The root 27-gate still passes in both modes with an empty certified set and exactly five missing dispositions. Fresh v10 audit is active under `01a0356d-f883-7e80-8f5e-734caf7e4127` in `independent_audits/batch_c_v10/`; a separate review and final C disposition remain downstream even if that audit passes. A command-safety rejection of an initial temporary-file wrapper is recorded in the work-state and orchestration ledgers; it ran nothing and changed no byte. The corrected fail-fast, no-temporary-file validation passed. Therefore this transition has no content or statistical consequence: all event decisions, recode application, model reruns, event-boundary fairness conclusions, release promotion, and atlas deployment remain closed until all five dispositions exist and validate. ## Batch B assurance repair complete; content conclusions still closed — 2026-08-24 Batch B's five blocking assurance defects are now repaired in immutable commit `d60dbb3d9a5d0c5e44cb4fb1d5cbda9979ac06d4` (package subtree `6509a59705208b8bfb69cd1e24a6dc3a75bb32b6`). Normal and optimized independent validation, both complete 16-control adversarial matrices, the full ordered Editorial Suite, exact Git-object comparison, and out-of-band executable authentication all pass. The five incorrect external-binding controls all fail as required. The repaired validator now rejects each of the eight honestly resealed false semantic/process surfaces that its predecessor accepted. This changes no event-content verdict. Batch B still contains 63 propositions (59 bounded-accepted and four Uncheckable), 117 relations (113 bounded-accepted and four Uncheckable), and 22 open holds. Flower-body, Justin source-critical, Vacante body-dependent, and mutable-manifestation limits remain fail-closed. No score, event boundary, family, candidate, release, atlas, or deployment decision changed. A fresh new-owner audit, separate integrating review, and final Batch B disposition still remain. That fresh audit is queued for the next reviewer slot because the four current slots are occupied by A v12, C v10, D v5, and provenance v8. The root 27-gate passes after `d60dbb3` in both interpreter modes but correctly continues to report an empty certified set and exactly five missing final dispositions. The real content/statistical sequence therefore remains: 1. finish the independent A/B/C/D/provenance audit-review-disposition chains; 2. integrate only evidence admitted by those valid final dispositions; 3. apply and independently audit the 1,700-cell recode workset; 4. rerun factor, family, missingness, era/region/mechanism, and all predeclared alternative event-boundary sensitivity analyses; 5. write the event-unit fairness result into the monograph, including why duration is never the sole split/merge rule; and 6. only then promote a release and author, QA, document, and deploy approved atlas additions/removals/merges. Nothing in this checkpoint deprecates the atlas's temporary broad-wrapper notice. That notice can be retired case-by-case only when each bounded successor has passed content admission, score/family coding, map/phase authoring, overlap checks, and release QA; the present Batch B assurance repair is a prerequisite, not that completion event. ## Batch A packaging repair added; Batch D awaits main validation — 2026-08-24 Batch A v12 passed its internal historical/assurance review but failed main's repository-packaging gate. It sealed a protected-copyright full DAI source body and 27 files excluded by the repository's ordinary ignore rules. This does not change any event, relation, score, family, or source-content decision; it means the valid audit cannot yet become the durable distributable audit. Main's exact finding is recorded under `independent_audits/main_reviews/batch_a_v12/`, and a new-owner v13 successor is active to retain only distributable metadata, locators, hashes, access/rights observations, audit reasoning, and fail-closed authority fields. A final Batch A disposition still comes only after that successor passes fresh validation and review. Batch D repair v5 has stopped at `ready_for_fresh_external_audit`, reporting an unchanged historical boundary and a sealed 269-file package. Main has not yet accepted it; its unusually large external-evidence payload must first pass the same source-retention and ordinary-stageability test that caught Batch A. Fresh Batch B v8, Batch C v10, and provenance v8 audits continue independently. Consequently none of the actual content/statistical open loops has closed: five final dispositions are still absent; no evidence has entered the certified integration set; the 1,700-cell recode application and independent recode audit have not run; models and alternative plausible event-unit sensitivities have not been rerun; the monograph cannot yet claim event-by-event fairness; and no temporary wrapper notice, event addition/removal/merge, release, or atlas deployment is authorized by this work. ## 2026-08-24 superseding lane status - **Batch D:** repair v5 is now immutably preserved at exact commit `33bb7df903591ad641eade4f8e2652f81df62754` after main independently validated all 269 package files, all 46 external-evidence files, the normal and optimized runs, the two 77-control matrices, four rebound-evidence attacks, Git preservation, no-action boundaries, and all-false downstream authority. Fresh findings-only v9 audit is running in `independent_audits/batch_d_v9/`; no D disposition exists. - **Provenance:** v8 returned `FAIL_REPAIR_REQUIRED` with five assurance and grounding findings. Its handoff still requires main validation and immutable preservation before a separate review may define any bounded repair. It authorizes no content or statistical change. - **Batch A:** v13 must replace the locally valid but non-distributable v12 package without protected source bodies or ignored working captures. - **Batch B and C:** fresh v8 and v10 audits respectively remain unfinished or awaiting formal handoff review. Presence of files is not certification. The substantive open-loop order is unchanged: finish all five independent audit/review/disposition chains; integrate only certified evidence; apply and independently audit the 1,700-cell recode set; rerun models and all plausible event-boundary sensitivities; write the resulting event-unit fairness analysis into the monograph; then promote a release and update the public atlas. Time span is explicitly not the sole event-unit rule: mechanism, actors, affected population, geography, land outcome, causal sequence, evidence resolution, and double-counting risk must also be tested. Until that sequence is complete, broad-wrapper notices remain case-by-case authoring TODOs rather than globally removable warnings, and no pending event addition, deletion, split, merge, trait recode, family assignment, or deployed map should be represented as statistically final. ### Provenance packaging correction Main independently accepts provenance v8's five repair-requiring findings but does not accept its 1,241-file working directory as a sealed audit package. The superseding review at commit `43bed3a` records the missing audit-package manifest/roster/seal, ignored Suite copies, and a validator that rewrites its own output. Fresh provenance v9 is packaging those same findings into a bounded distributable successor. This is assurance work only: it neither adds nor removes an event and does not advance the recode/model/fairness gate. ## 2026-08-24 current content/statistical boundary after A v13 and D v9 review No event-content or statistical decision has been promoted by the latest audit work. It has instead prevented two inadequately assured packages from becoming inputs to the recode/model lane: - Batch A v13 removed the prohibited source bodies and ignored files, but main proved that its unauthenticated terminal prose, JSON/CSV twin divergence, and transformed source-locator identity all fail open. Its exact package, replay controls, and `FAIL_REPAIR_REQUIRED` review are preserved at `c782672`. Batch A v14 is queued to repair only those four assurance findings. - Batch D v9 correctly found that v5's external-execution evidence cannot be reproduced from its Git object because it binds mutable absolute paths. Its unsealed findings handoff and main review are preserved at `417cc9a`; D v10 is building only a sealed, independently reproducing successor handoff. - Provenance v9's hard-coded negative-control outcomes and honest-reseal failures are preserved at `2ddcee2`; provenance v10 is active. Batch B v8 and Batch C v10 remain active fresh audits. Therefore the real content/statistical open loops remain, in order: 1. finish and independently review the A/B/C/D/provenance audit chains; 2. create all five final dispositions, with the certified set empty until all gates actually clear; 3. integrate only disposition-authorized evidence; 4. apply and independently audit the 1,700-cell recode workset; 5. rerun factor count, family structure, missingness, era/region/mechanism, and every predeclared plausible split/merge/event-boundary sensitivity; 6. write the event-by-event fairness result into the monograph, explaining temporal scale alongside actors, mechanism, affected population, geography, land outcome, causal sequence, evidence resolution, and double-count risk; 7. promote a statistical release; then author and QA the approved atlas/data/ monograph changes and deploy them. Five final dispositions are still absent. No wrapper notice—including the Roman/Po Valley notice—can yet be deprecated on statistical authority alone. Each notice can be removed only when its bounded successors have cleared admission, coding, overlap, map/phase authoring, and release QA. No pending event addition, deletion, split, merge, trait change, family change, or public data refresh is statistically final at this checkpoint. The root suite passed all 27 checks in both interpreter modes after `c782672` and again after the state-ledger commit `14fc5ea`, preserving exactly that boundary: zero certified packages, five missing final dispositions, and all downstream authority false. ## Batch D fresh-review preservation closes an assurance subloop only — 2026-08-24 The Batch D v12 retry1 fresh review is preserved at commit `59835ad3c5412dd790c3d03f07b1e4a06d4103cc`, after main independently replayed its complete normal/optimized review, depth-11 rebuild, 152 adversarial executions, scratch/portable-command controls, Editorial Suite retention, authority checks, and cleanup. The preservation is real, but its scope is deliberately narrow: it validates the v12 review package and finds no new v12 defect; it does not resolve target finding `BDV9-F-001`, certify the underlying source claims, or create the final Batch D disposition. Accordingly the real content/statistical open loops remain: 1. finish D's full pre-disposition chain and the still-active A/B/C/provenance review or bounded-successor chains; for D specifically, first run the queued independent integrating review that adjudicates and bounds `BDV9-F-001`, then repair the accepted defect, obtain a new zero-finding audit, and obtain a later separate zero-objection integrating review before any disposition; 2. validate all five final dispositions before populating the certified set; 3. integrate only disposition-authorized evidence and independently audit the resulting 1,700-cell recode application; 4. rerun factor count, families, missingness, era/region/mechanism checks, and all predeclared plausible event-unit alternatives; 5. write the monograph's event-unit fairness conclusion from those results, treating duration as only one criterion; and 6. promote a release before changing public event membership, wrappers, maps, phases, data, or notices. Five dispositions are still absent and the certified set remains empty. The Roman and Po Valley wrapper notices therefore remain valid temporary notices: their removal still requires bounded-successor admission, coding, overlap, map/phase authoring, and release QA, not merely completion of Batch D's package review. No current result adds, removes, splits, merges, or recodes a public event. The root 27-gate passed again after D preservation/state commit `5c1bb89` and queued-review commit `5af1098` in normal and optimized modes with byte-identical logs, zero certified packages, exactly five missing dispositions, and all downstream authority false. At 20:00 MDT, Batch B v10 returned for main authentication and the freed independent slot launched the queued Batch D v9 integrating review. This is a scheduling advance only: B's claimed assurance repairs are not accepted before reproduction, and D's review may only adjudicate/bound `BDV9-F-001`. The five content/statistical dispositions remain absent. ## Batch C source failure is bounded; no content authority opened — 2026-08-24 21:03 MDT Batch C v11's failed fresh review and independent intake are preserved at `57733799…`; the bounded v12 repair commission is preserved at `54c59699…` and is active in a new sole-write lane. The reproduced defects concern two page locators, two exact short quotations, four rows' work/edition/locus metadata, the associated rights rationale, and dependent evidence/report records. They do not presently add, remove, split, merge, recode, or re-family an event. The first provenance integrating-review attempt stopped before substantive output on an automated platform screen. Its temporary directory is non- evidence. Retry1 is active under `7bce9d9…` with the same bounded five-finding scope and a new empty directory. This likewise changes no content conclusion. The real open-loop order remains: 1. complete and independently preserve the active A/B/C/D/provenance assurance chains; 2. create and validate all five final dispositions while Gate 25 remains empty; 3. integrate only disposition-authorized evidence; 4. apply and independently audit the 1,700-cell recode workset; 5. rerun factor count, family structure, missingness, era/region/mechanism checks, and every predeclared plausible event-boundary alternative; 6. write the event-unit fairness result into the monograph, treating duration as one criterion alongside actors, mechanism, affected population, geography, land outcome, causal sequence, evidence resolution, and double- counting risk; 7. promote a statistical release; and only then author, QA, publish, and deploy approved event, wrapper, map, phase, raw-data, and monograph changes. Roman and Po Valley wrapper notices still require their bounded successors to be mapped, phased, overlap-checked, authored, and release-QAed. Statistical splitting alone does not deprecate them. Five dispositions remain absent and no current result authorizes a public event or notice change. The complete root 27-gate passed after state commit `70c16479…` under normal and optimized Python with byte-identical full-output SHA-256 `fcf1b1d1…`, no tracked write, zero certified packages, exactly five missing dispositions, and all downstream authority false. ## Assurance transitions; substantive content still closed — 2026-08-24 21:56 MDT Batch A v15 retry1, Batch D's bounded integrating review, and Batch B's failed fresh review are now independently preserved. Their effects are deliberately narrow: - A requires one frozen construction/index state; no historical row changed. - D requires portable replay of already retained execution evidence; no source or historical verdict changed. - B requires complete/short Calibration grounding and reproducible wrappers; all 63 proposition, 117 relation, and 22 open-hold outcomes remain unchanged. A v16 and D portable successor v1 are active; C v12 and provenance retry1 continue; B v11 is committed and queued. All five final dispositions are still absent. Consequently the event roster, scores, families, wrapper notices, statistical release, fairness conclusion, public raw data, monograph build, and atlas deployment remain unchanged and unauthorized. The real remaining sequence is still: 1. finish, preserve, and independently review every active/queued assurance successor, including RC-013/RC-014 for D and a fresh post-v11 B review; 2. create and validate A/B/C/D/provenance dispositions with an otherwise empty certified set; 3. integrate only disposition-authorized source evidence; 4. apply and independently audit the 1,700-cell recode workset; 5. rerun factor/family/missingness models and every predeclared plausible event-unit alternative across time, actors, mechanism, population, geography, land outcome, causal sequence, evidence resolution, and double-counting risk; 6. write the resulting event-by-event fairness argument into the monograph; 7. promote the next statistical release; then author and QA the bounded Roman/ Po Valley replacements and every other approved event/map/phase/data change; 8. only then remove obsolete wrapper notices, rebuild public monograph/data artifacts, and deploy. The Batch B intake namespace collision is recorded and corrected without any content effect. The complete 27-gate suite must be rerun after the new state commit; the earlier `70c16479…` execution remains the last recorded pass. ## Gate verification — 2026-08-24 22:01 MDT The complete 27-gate root suite now passes after state commit `ac9052651595` under normal and optimized Python with byte-identical 226-line logs at SHA-256 `fcf1b1d18c78bb3719a6e5ea235027c9515ea4af2a2394bab8db07ae240f9100` and no tracked-tree change. This confirms the open-loop list rather than closing it: there are still no certified packages, exactly five final dispositions remain absent, and every downstream authority is false. The next work is still to preserve and independently review the active A/C/D/provenance returns, launch queued B v11 at the first free slot, and complete all five chains before any evidence integration, recoding, model rerun, fairness conclusion, wrapper notice removal, publication, or deployment. The same normal/optimized suite was rerun after documentation commit `37dca9366dd5` and returned the identical 226-line log and SHA-256 without a tracked-tree change. The saved open-loop ledger therefore matches the live gate state. ## C and provenance assurance advances; content/statistical gates remain closed — 2026-08-24 22:34 MDT Provenance retry1 and Batch C v12 have now been independently authenticated, fully replayed, and preserved at `92eb3d1…` and `ca10eba…`, respectively. Both original returns landed in the wrong legacy tree; main accepted neither on that basis. Exact immutable-object authentication, no-hard-link relocation, normal/optimized replay, mutation testing, byte/mode comparison, and canonical transport all passed before preservation. The legacy tree is non-governing. These are assurance advances, not content or statistical conclusions: - provenance still requires bounded repair v7, a new-owner zero-finding audit, and a later separate zero-objection review; - Batch C still requires the newly committed v13 genuinely fresh review and a later separate integrating review; - Batch A v16, Batch B v11, Batch D portable successor v1, and provenance repair v7 are active; Batch C v13 is queued for the first free slot; and - all five final dispositions remain absent, so no package is certified. The substantive remaining sequence is unchanged: 1. finish and independently preserve/review every A/B/C/D/provenance successor; 2. create and validate all five final dispositions, leaving the certified set empty until the five-lane gate clears as a whole; 3. integrate only disposition-authorized evidence; 4. apply and independently audit the 1,700-cell recode workset; 5. rerun factors, families, missingness, era/region/mechanism checks, and all predeclared plausible event-unit alternatives across duration, actors, mechanism, population, geography, land outcome, causal sequence, evidence resolution, and double-counting risk; 6. write the event-by-event fairness result into the monograph; 7. promote a statistical release; then author and QA approved bounded events, wrapper replacements, maps, phases, public data, and monograph artifacts; 8. only then remove obsolete notices and deploy. Accordingly, the Roman and Po Valley `Broad wrapper retired` notices cannot yet be deprecated. Their statistical successors must also finish admission, coding, overlap checks, map/phase authoring, and release QA. The public atlas must continue to display the temporary wrapper warning until that entire chain is complete. The post-transition root gate passes normally and optimized with identical 226-line logs at SHA-256 `fcf1b1d18c78bb3719a6e5ea235027c9515ea4af2a2394bab8db07ae240f9100`: zero certified packages, exactly five absent dispositions, and all downstream authority false. This verifies the list above; it does not close an item. The post-ledger-commit replay after `4ee27e4` produced the same byte-identical normal/optimized result and no tracked-state change. The open-loop list is therefore both committed and live-gate verified. ## A/B assurance failures preserved; content and statistical conclusions remain closed — 2026-08-24 23:15 MDT Batch A v16 and Batch B v11 have now been independently inspected and preserved rather than promoted: - A v16's historical/no-action payload is internally exact, but its handoff is not yet freshly auditable: the live ledger has two incompatible terminal statuses and the validator demands unrecoverable raw Git-index cache bytes. Bounded v17 repair is active. No event, source judgment, or coded value changed. - B v11's technical grounding, locator, control, relocation, and Editorial Suite work passes, but the owner correctly failed the lane after two empty diagnostic files were briefly created outside the absolute write boundary. A clean v12 replay is commissioned and queued. The 63 proposition, 117 relation, 22 open-hold, eight `Uncheckable`, and all no-action outcomes remain unchanged. Batch C v13 and Batch D RC-013 are now active genuinely fresh audits; provenance repair v7 and Batch A v17 are the other two active lanes. Batch B v12 waits for the next independent slot. Five final dispositions remain absent, Gate 25 must remain empty, and nothing from these assurance transitions may enter the certified evidence set. The real content/statistical sequence is unchanged: 1. finish, preserve, and independently review all A/B/C/D/provenance chains; 2. create and validate all five final dispositions; 3. integrate only disposition-authorized evidence; 4. apply and independently audit all 1,700 recode cells; 5. rerun factor count, families, missingness, era/region/mechanism checks, and every predeclared plausible event-boundary alternative across duration, actors, mechanism, population, geography, land outcome, causal sequence, evidence resolution, and double-counting risk; 6. write the event-by-event fairness result into the monograph; 7. promote a statistical release, then author and QA approved bounded events, maps, phases, wrappers, public data, and monograph artifacts; and 8. only then remove obsolete notices and deploy. The Roman and Po Valley `Broad wrapper retired` notices therefore remain necessary. Statistical splitting alone is not enough: their successors still need certified admission/coding, overlap checks, map and phase authoring, and release QA. The complete post-snapshot root gate passes under normal and optimized Python with byte-identical 226-line logs at SHA-256 `fcf1b1d18c78bb3719a6e5ea235027c9515ea4af2a2394bab8db07ae240f9100`. It confirms zero certified packages, exactly five absent dispositions, and no content/statistical promotion authority. The post-ledger replay after `9292e40b7d02` returned the same byte-identical normal/optimized result and no tracked-state change. ## Provenance repair preserved; substantive pipeline still closed — 2026-08-24 23:42 MDT Provenance repair v7 is now independently authenticated and preserved at `bd3b3ae0729b`. Main reproduced all 24 canonical/portable positive checks and the complete 56-class / 224-execution adversarial matrix, with unchanged inventories, zero validator writes, and normal/optimized parity. The repair preserves the five findings, nine no-action rows, six domain-state carriers, and the 48-row `40/6/2/0/0` Calibration baseline. This closes no content or statistical question. Provenance still requires: 1. the newly commissioned genuinely fresh audit at `2ee1bf9a7bef`; 2. independent CAL-010 component and CAL-022 observation review; 3. a later separately owned zero-objection integrating review; and 4. a validated final provenance disposition alongside the other four lane dispositions. Batch B v12 is now active under `01a0376f-78a3-7bb1-abfb-66e15df3f5fb`; A v17, C v13, and D RC-013 remain the other active lanes. Provenance fresh audit is queued next. The real content/statistical sequence is unchanged: finish all five assurance chains and dispositions; integrate only certified evidence; apply and independently audit the 1,700-cell recode; rerun all models and predeclared event-unit alternatives; write the fairness result into the monograph; promote a release; then author, QA, publish, remove obsolete wrapper notices, and deploy. The Roman and Po Valley notices therefore remain necessary. Rerun the complete normal/optimized 27-gate after this transition is committed. The post-transition execution after `1441c9059eb3` passed all 27 gates in normal and optimized modes with identical 226-line output at SHA-256 `fcf1b1d18c78bb3719a6e5ea235027c9515ea4af2a2394bab8db07ae240f9100`. It confirms zero certified packages, exactly five absent dispositions, and no content/statistical promotion authority. ## Current content/statistical gate — 2026-08-25 00:26 MDT The assurance pipeline has advanced, but no event-level evidence or coded value has entered the certified corpus: - Batch A v17 is independently preserved at `35d7bbe81090`; the genuinely fresh A v18 audit is active under `01a0378e-4a5f-7cb2-8372-1fdf0c184d38`. - Batch B v12 and Batch C v13 remain active under `01a0376f-78a3-7bb1-abfb-66e15df3f5fb` and `01a03740-fc99-7e61-99ad-049dc5b24bbb`. - Batch D's zero-finding RC-013 audit is independently preserved at `43a4bf31aeac`; the separately owned RC-014 integrating-review commission is committed at `6d4d14f` and queued. Its three evidence holds remain open. - Provenance repair v7 is preserved at `bd3b3ae0729b`; its genuinely fresh audit is active under `01a0378c-34d8-7671-9043-961263abd3e3`. The remaining content/statistical sequence is therefore: 1. finish, close, independently authenticate, and preserve the four active returns; 2. launch and complete D RC-014 at the first free independent slot, and run every other still-required separately owned integrating review; 3. create and validate exactly five final A/B/C/D/provenance dispositions; 4. integrate only disposition-authorized evidence; 5. apply and independently audit the complete 1,700-cell recode workset; 6. rerun the factor count, families, missingness, era/region/mechanism checks, influence diagnostics, and every predeclared plausible event-unit alternative across duration, actors, mechanism, population, geography, land outcome, causal sequence, evidence resolution, and double counting; 7. write the event-by-event fairness result and residual uncertainty into the monograph; 8. promote a statistical release, then author and QA approved bounded events, wrappers, maps, phases, public data, and monograph artifacts; and 9. only then deprecate obsolete `Broad wrapper retired` notices and deploy. The Roman and Po Valley notices therefore remain necessary. Five final dispositions are absent, the certified set is empty, and no current result authorizes an event addition/removal/split/merge, score or family change, wrapper retirement, data/monograph publication, atlas change, release, or deployment. The root 27-gate must be rerun after this transition is committed. The post-transition root execution after `b4322e3` passed normally and optimized with byte-identical 226-line output at SHA-256 `fcf1b1d18c78bb3719a6e5ea235027c9515ea4af2a2394bab8db07ae240f9100`. It confirms an empty certified set, exactly five absent dispositions, and no content/statistical promotion authority. ## Current content/statistical gate — 2026-08-25 01:04 MDT - Batch A v18 is immutable at `6fa716a` and awaits a separate integrating review. - Batch B v12 is immutable at `41ba8e9`; its genuinely fresh v13 audit is active under `01a037b8-ea08-7151-8e27-9499b35e20e2`, with the contradictory terminal/progress ledger state made an explicit semantic audit question. - Batch C v13 remains active under `01a03740-fc99-7e61-99ad-049dc5b24bbb`. - Batch D RC-014 is active under `01a037ab-8e27-7dc1-ba06-f626930db708` against the amended commission at `144c3a0`. - The provenance fresh audit found two real assurance defects: exact POSIX modes were not enforced and its Calibration records were not grounded in the pinned Suite's required form. Main then accidentally ran its mutating finalizer during intake. The original owner is active again solely to reconstruct or explicitly supersede that failed terminal package and record the main incident. Its findings remain non-integrable pending preservation and a separately owned bounded repair. This changes no event or statistical conclusion. The exact continuation is: 1. finish, close, independently authenticate, and preserve the four active B/C/D/provenance returns; 2. repair provenance's two findings under a new bounded owner, then fresh-audit and integrate-review that successor; 3. complete A and B's separately owned integrating reviews and every remaining C/D review step; 4. resolve the cross-lane schema-v3 disposition interface through a separately specified and audited implementation if the exact immutable chains cannot be consumed as-is; 5. create and validate exactly five final A/B/C/D/provenance dispositions; 6. integrate only disposition-authorized evidence; 7. apply and independently audit the complete recode workset; 8. rerun all models and predeclared event-boundary alternatives, including time, actors, mechanism, population, geography, land outcome, causal sequence, evidence resolution, and double-counting sensitivity; 9. write the event-by-event fairness result and residual uncertainty into the monograph; then promote a release, author and QA approved cases/wrappers/ maps/phases/public data, remove obsolete notices, and deploy. The Roman and Po Valley `Broad wrapper retired` notices therefore still cannot be deprecated. Five final dispositions are absent and the certified set is empty. No addition, removal, split, merge, recode, model conclusion, public artifact, or deployment is authorized by the assurance work above. ### 01:12 MDT provenance transition The restored failed provenance audit and main intake are immutable at `eb6e1a3`. Its two findings remain assurance-only: exact POSIX mode enforcement and Suite-compliant Calibration grounding/self-audit. The bounded successor commission is immutable at `c9d1003` and active under `01a037c1-c079-7620-a9c8-c604572d1e1b`. It may change no historical claim, event, score, family, factor, candidate, hold, model, or public artifact. The post-transition root gate passes normally and optimized with identical 226-line output at `fcf1b1d…f9100`; the certified set remains empty and exactly five final dispositions remain missing. The content/statistical sequence and wrapper-notice hold immediately above are unchanged. ## Current content/statistical gate — 2026-08-25 01:24 MDT Batch C v13 is no longer an active unknown. Its exact failed audit and main intake are preserved at `90c7fab`. The only accepted substantive defect is a missing accepted dependency for the phrase that some Delians entered Athenian civic categories; the currently accepted passages license remaining or returning and uncertainty, not that additional implication. The v14 bounded repair is active under agent `01a037cd-3c6a-7283-bdf0-59bfb357286a` and commission `a3ffada`. It must either ground and link an exact source passage or narrow the relation and Calibration claim from five licensed relations to four. This is an assurance repair, not an event or statistical result. The current exclusive lanes are B v13, C v14, D RC-014, and provenance v8. Batch A's integrating review is queued. Exactly five final dispositions remain absent, the certified set remains empty, and no addition, removal, split, merge, recode, factor/family conclusion, fairness conclusion, public artifact, wrapper-notice change, or deployment is authorized. The substantive order remains: complete all five assurance chains and final dispositions; integrate only certified evidence; apply and independently audit the full recode; rerun models and all predeclared event-boundary alternatives; write the event-by-event fairness result and uncertainty into the monograph; then promote a release, author and QA approved public cases/data/monograph artifacts, deprecate only genuinely obsolete wrapper notices, and deploy. The last complete root gate passed after `553d356` normally and optimized with identical 226-line output at `fcf1b1d…f9100`. Rerun it after this transition is committed. Post-transition verification after `be29b1d` passed all 27 checks normally and optimized with byte-identical 226-line output at `fcf1b1d…f9100`. The certified set is empty, exactly five dispositions are absent, and all content/statistical promotion authority remains false. ## Current content/statistical gate — 2026-08-25 02:03 MDT Batch D RC-014 found no historical, source, relation, hold, or substantive audit defect. It failed only because the exact immutable RC-013 package does not natively match the schema-v3 disposition interface. That failed review is preserved at `36263c9`; the newly commissioned native-contract successor at `c157287` must rerun the audit and emit the required artifacts without aliases or old-package mutation. It is active under `01a037f1-8eb3-7cf2-a7d9-dd3aa0af698b`. The current content/statistical blockers are therefore still assurance work, not event votes: B v13 fresh audit, C v14 dependency repair, D native-contract successor plus a future new RC-014, provenance v8 successor plus its later reviews, and A's queued integrating review. Exactly five final dispositions remain absent and the certified set is empty. No event addition/removal/split/merge, recode, model conclusion, fairness claim, wrapper-notice removal, public data/monograph promotion, atlas change, release, or deployment is authorized. The substantive order remains unchanged: clear and disposition all five lanes; integrate only certified evidence; apply and independently audit the recodes; rerun models and every predeclared event- boundary alternative; complete the fairness result; then promote, author, QA, deprecate obsolete notices, and deploy. The last complete root gate passed at `a24e51c` normally and optimized with byte-identical 226-line output at `fcf1b1d…f9100`. Rerun it after this transition is committed. Post-transition verification after `0a9c5b0` passed normally and optimized with byte-identical 226-line output at `fcf1b1d…f9100`. The certified set is empty, exactly five dispositions remain absent, and all content/statistical promotion authority remains false. ### Batch A assurance queue correction — 2026-08-25 02:21 MDT Immutable A v18 is substantively zero-finding but has the same native-contract gap D exposed: it lacks the exact top-level audit artifacts the disposition builder requires. The assessment and bounded v19 commission are preserved at `0381097`; v19 must run before A's integrating review. This is an assurance- packaging correction only and changes no event, code, model, or public result. The complete post-queue gate after `326ef77` passed normally and optimized with identical 226-line output at `fcf1b1d…f9100`; the certified set remains empty and exactly five final dispositions remain absent. ## Current content/statistical gate — 2026-08-25 03:02 MDT No newly researched evidence, event boundary, coded value, family, factor, or public atlas row has crossed into the certified corpus. The assurance work has advanced as follows: - A v19 native-contract work is queued after immutable assessment at `0381097`. - B v13's two terminal/interface findings are preserved at `0dcb6bc`; the v14 bounded repair is active under `01a03822-a3a8-7493-a389-7b7eaba1b592`. - C v14's bounded Delian dependency repair is immutable at `fb03404`; its v15 genuinely fresh audit is active under `01a0381b-a6bb-7c61-be2a-9d2c1ec9393b`. - D's new native-contract RC-013 audit is zero-finding and immutable at `5c74399`; its separate RC-014 integrating review is active under `01a03827-68c6-7c50-b423-4540931fdeae`. - Provenance v8 is preserved at `c141073` plus visible identity correction `e84ed64`; its fresh audit is active under `01a03816-9e75-7f63-a0b7-09dd80dcd2b8`. The remaining content/statistical sequence is unchanged: 1. finish and preserve the four active lanes, launch A v19, and complete every still-required separate integrating review; 2. create and validate exactly five final A/B/C/D/provenance dispositions; 3. integrate only disposition-authorized evidence; 4. apply and independently audit the full 1,700-cell recode workset; 5. rerun factor count, families, missingness, era/region/mechanism balance, influence diagnostics, and every predeclared plausible split/merge event- unit alternative—not duration alone; 6. write the event-by-event fairness result and residual uncertainty into the monograph; 7. promote a statistical release and author/QA the approved bounded events, maps, phases, wrappers, public data, and monograph artifacts; and 8. only then remove obsolete `Broad wrapper retired` notices and deploy. The Roman and Po Valley notices therefore remain necessary. Five final dispositions are absent, the certified set is empty, and no event addition, removal, split, merge, score/family change, notice removal, publication, atlas change, release, or deployment is currently authorized. Run the complete root gate normally and optimized after this state commit; require Gate 25 empty, Gate 26's exact five blockers, Gate 27 reproduction, and all downstream authority false. The complete post-checkpoint gate after `73d56f2` passed in normal and optimized modes with byte-identical 226-line output at SHA-256 `fcf1b1d18c78bb3719a6e5ea235027c9515ea4af2a2394bab8db07ae240f9100`. The certified set is empty, exactly five dispositions remain missing, Gate 27 reproduces, and every downstream authority flag is false. ## Streamlined assurance transition — 2026-08-25 09:34 MDT Craig approved the proportionate assurance standard in `ASSURANCE_STREAMLINING_STANDARD.md`. This backlog no longer doubles as an agent roster; current machine state is `PIPELINE_STATE.json` and the narrative authority is `/CURRENT_WORK_STATE.md`. The scholarly sequence remains unchanged, but packaging-only failures no longer trigger complete historical re-audits. Main will now: 1. authenticate and preserve the four returned packages; 2. classify each defect as substantive, mechanical, or editorial; 3. apply at most one bounded repair and one independent verification; 4. issue main dispositions when those burdens pass; 5. integrate only after all five dispositions exist; 6. audit the recodes and rerun models plus event-boundary sensitivities; and 7. carry the required fairness result from `MONOGRAPH_EVENT_UNIT_FAIRNESS_SCAFFOLD.md` into the monograph before publication. Roman and Po Valley wrapper notices remain necessary until that sequence and public authoring/QA are complete. No current return authorizes a content, statistical, publication, atlas, or deployment change. ## Superseding finite open-loop checkpoint — 2026-08-26 The earlier assurance-chain lists in this document are historical provenance, not the current critical path. All five evidence lanes have passed the streamlined root gate and their authorized evidence has already entered the corrected diagnostic workflow. Target admission is complete for 233 targets: 158 eligible pending score, 38 evidence holds, 28 scope exclusions, and nine unresolved overlaps. The active-event v2 lane has also finished research and double coding. All ten returns validate; the 364-cell comparison contains 317 exact agreements and 47 rows requiring one main no-averaging adjudication. Commit `f48e7d3` preserves the raw returns, packet 05's non-substantive contract normalization, validation, and complete comparison. The real remaining content/statistical sequence is therefore: 1. ~~adjudicate and validate the 47 active-score disagreements once;~~ Complete in `c206a2d`: 364 cells resolved, validated, and still unapplied; 2. ~~run the prepared eight-packet source-sufficiency review for the 158 eligible prospective targets and make at most one bounded repair.~~ Complete. The proportional score-reuse inventory in `191c6df` finds 17 one-to-one successors, 140 split children, and one merged-parent target. The completed whole-row review in `15b79d4` finds material boundary changes in all seventeen one-to-one pairs, so all 158 targets default to fresh scoring after clearing the source gate. The eight source reviews and reproduced consolidation are now complete at `0bae011` and `8d5300b`: 131 rows pass with explicit gaps, 21 require access-or-locator repair, and six require one missing-domain source repair. The one permitted two-lane repair is complete: `21203f1` preserves the twenty-one access repairs and main review; `880abcd` freezes the final gate at 157 source-cleared rows and one admission hold. The held Somalia 2000–05 wrapper has a preserved May 2002 north-Mogadishu successor lead. The 157 cleared rows are now frozen into eight blind score packets at `b01c1f5`, totaling 9,420 cells. Complete two blind coders per packet next; 3. reconcile actual prospective-score disagreements once and assemble explicit candidate memberships with parent/child and overlap controls; 4. rerun the preregistered models and all plausible event-boundary, weighting, era, region, mechanism, evidence-resolution, duration, and influence sensitivities; 5. write the measured event-unit fairness result and uncertainty into the monograph; 6. resolve the mechanism-versus-primary-family information test and remove the obsolete public `Worldwide audit` prose while preserving provenance; 7. run the single final Editorial Suite, promote the statistical release, and author/QA/deploy the approved atlas, data, monograph, wrapper, map, phase, and interface changes. Evidence holds, unresolved overlaps, and broader ancient/medieval searches remain explicit research queues; they do not silently enter the candidate and do not automatically block the next release. No recursive audit cycle is authorized absent a newly demonstrated substantive defect.