# Bounded publication-text check

Scope: `SUCCESSOR_RESULTS.md` only, plus later public status/link wiring.
Venue: Craig Talbert's public working monograph and release notes, not an
encyclopedia or peer-reviewed final article. AI-assisted drafting/revision is
authorized. The author's Chicago author–date preference remains; this results
appendix uses named project-artifact locators, not invented references.

The attached Editorial Suite governs grounding, tells and calibration. The
approved streamlining standard limits this to changed claims, not a new
historical audit. Main performs this editorial check, not an independent
verification of its own numerical work. Peirce's separate numerical review
is closed and its limits are disclosed. No whole-manuscript pass is claimed.

## 1. Fact check — PASS AFTER ONE WORDING CORRECTION

Main read the complete Fact-Check and Revise reference (v2.7 draft) from the
ZIP. The first output was truncated; the missing middle and attribution
section were subsequently read explicitly. Local sources consulted on
2026-09-07: complete core summary, relevant actual diagnostic JSON fields and
scenario records, complete verifier report/findings, main diagnostic/family
decisions, family implementation and current deployment/state records.
These are the project's own bound outputs. No external historical paper or
funding assertion supplies a new ground for this results text. Prior source
reviews are decisions being reported, not claimed re-read primary sources.

| Claim group | Ground / locator | Verdict |
| --- | --- | --- |
| Candidate, not active/public release | Active pointer, deployment receipt, main disposition | Correct; status changes only after actual promotion/deployment. |
| 675−13+14=676; 771 metadata /95 inactive | core/SUMMARY.json roster fields | Correct. |
| 17 packages /2,716; 840 new; 927 changed /949 confirmed | Accepted complete input, main disposition, verifier input report | Correct; coding, private integration and publication distinguished. |
| 37,844 unchanged, including100 fractions | Core summary and exact-string verification | Correct: 662×60−1,876=37,844, not a whole-corpus unchanged claim. |
| 672 classified /four family holds | Expected family CSV, main family disposition | Correct; not all workflow holds. |
| Nine factors; angles18.69°/11.78°; named Tucker0.078 | DIAGNOSTIC_SUMMARY canonical/successor; verifier report | Correct; named-axis congruence is not rematched space agreement. |
| 17.98% legacy reassignment /662; ARI0.632 | Actual comparison, verifier report, main disposition | Correct; explicitly not editorial families. |
| 565/536; retention524/29/12; separate battery6–8 | Diagnostic summary retention_counts/indicator_overlap; verifier | Correct; not independent replications or probabilities. |
| Weights676 real rows /total583 /effective615.78 /angle44.67° | Weighted scenario and verifier report | Correct; illustrative rather than population weights. |
| Dates628/627; duration absent; two alternatives unestimable | Core summary, acceptance, readout limits | Correct; not a claim that all historical tempo is unknown. |
| Approved F8; six raw follow-ups still open | MAIN_F8_INHERITANCE and existing decisions | Correct; implementation not described as deployed. |
| Verification limits | Complete REVIEW.md Role and limits / Actual completed results | Correct; all scenario reconstruction versus selected direct fits, prior work and nonhuman status disclosed. |
| Census/universality limits | Stated methods scope, not a numerical hypothesis result | Correct editorial interpretation. |

Finding: “The sources, coding and this text also involve AI assistance” could
misattribute AI authorship to historical documents. Changed to “Source review,
coding and this text also involve AI assistance.” No project provenance changed.
The first combined file patch failed on line wrapping and wrote nothing;
the actual line was read before applying the correction.

No overstated historical result, unsupported outside statistic, political
spectrum label, promotion or unmarked moral inference was found in this scope.
Order: register → results → limits. Interpretation belongs to the project,
not historical authors. No outside quotation requires web archiving; local
hash bindings and Git preserve the consulted results. Package-link mapping
remains a mechanical check, not an already click-tested public artifact.

### Required self-audit

| Item | Status | Basis |
| --- | --- | --- |
| Grounding | PASS | Consulted outputs, not recalled outside sources. |
| Claims | PASS | Exact findings retained; no universality claim. |
| Deletions | PASS | No substantive claim cut for citation anxiety. |
| Criticisms | PASS | Numeric findings and F8 mismatch checked against files. |
| Internal consistency | PASS | Candidate, frozen release and public atlas distinct. |
| Register discipline | PASS | Interpretation explicitly the project's conclusion. |
| Moral framework | PASS | No moral conclusion disguised as empirical result. |
| Political-spectrum contamination | PASS | No spectrum descriptors. |
| Attribution posture | PASS | Own outputs and model verification identified. |
| Quotations | PASS | No outside direct quotation in scoped text. |
| Causal reasoning | PASS | No historical causal inference from factor changes. |
| Categorical/dimensional | PASS | Fits, traits and editorial families separated. |
| Denotation/connotation | PASS | Weight total is not rows; same count is not same structure. |
| Proper names | PASS | Four names tied to exact family-hold IDs. |
| Engagement | PASS | Unfavorable weighting result kept explicit. |
| Confidence calibration | PASS | Documented alternatives, not fairness certification. |
| Source disclosures | PASS | Own-source/AI provenance and limitations read; no outside funding claim. |
| Citation resolution | PENDING PACKAGE LINKS | Local locators resolve; final public links await assembly. |
| Adopted reviewer text | PASS | Numeric claims checked against actual summary and review scope. |
| SYNTH | PASS | Interpretation owned; no fabricated historical synthesis. |

## 2. LLM Artifact Tells — PASS AFTER TWO STYLE EDITS

The full 245-line v2.7 draft reference was read in four contiguous chunks.
All five tiers were inspected in order. Mechanical searches covered tracking
parameters, internal citation tokens, attribution objects, span markers,
private-use Unicode, collapsed reference chips, raw writing/wikitext blocks,
return arrows, placeholders and malformed dates: zero matches in the scoped
text. A separate DOI/ISBN/access-date/reference-pattern scan also returned no
matches; there are no such identifiers to validate in this text. Project file
names are actual locators, not fabricated titles of outside publications.

| Instance | Tell / tier | Class | Disposition |
| --- | --- | --- | --- |
| Opening section heading's “not a larger claim about history” | Repeated negative framing /4 | Style-only | Shortened to “Register corrections”; substantive scope limits remain later. |
| Final instruction to connect locators when incorporated | Drafting instruction in reader text /3 | Style-only | Moved the pending link-wiring requirement into this companion note. |

The other contrasts distinguish real recurring confusions: 583 total weight
versus676 observations; nine exploratory factors versus eight descriptive
traits; legacy clusters versus editorial families; unavailable movement
duration versus elapsed dates. Removing those contrasts would lose meaning.
Degree signs and arithmetic belong to the statistical subject, not a claim
of AI/human authorship. No detector, humanizer or provenance concealment was
used. The exact numerical and source-claim burden is unchanged by these edits.

| Self-audit item | Status | Basis |
| --- | --- | --- |
| Checklist vs heuristic | PASS | Context determined findings, not a banned-word list. |
| False positives | PASS | Mathematical contrasts and necessary scope caveats retained. |
| Hard artifacts | PASS | Specific searches run, zero hits; no DOI/ISBN/access-date claims. |
| Tier coverage | PASS | Tiers1–5 inspected in order. |
| Decay weighting | PASS | No perishable detector statistic used as proof. |
| Guideline/style sort | PASS | Two discretionary style edits, no claimed venue violation. |
| Introduced tells | PASS | Removed heading/instruction only; no new filler transitions. |
| Writer's voice | PASS | Precise mathematical terms and direct results retained. |
| Register homogenization | PASS | Results and owned interpretation remain distinguishable. |
| Quality tradeoffs | PASS | Reader-facing heading shorter; reproducibility information remains. |
| SYNTH | PASS | No new interpretive connection added. |
| Venue | PASS | Authorized AI-assisted working paper, not restricted-venue laundering. |
| Provenance honesty | PASS | Model drafting/review disclosure retained. |

## 3. Calibration Review — MATCHED IN DECLARED SCOPE

The complete v1.2 draft reference was read. Coverage is every substantive
paragraph of `SUCCESSOR_RESULTS.md`, grouped below by its load-bearing claim.
No full monograph, historical-source corpus, unproduced release package or
future deployment is certified. No background-knowledge plausibility verdict
substitutes for evidence. Passages below are exact strings from the consulted
local JSON/review/decision records, not paraphrases in quotation marks.

| Claim / location | Load-bearing wording | Source and verbatim grounding | What it licenses | Materiality | Verdict / change |
| --- | --- | --- | --- | --- | --- |
| Opening status | “still a private candidate” | Main disposition: “It does not activate a release, alter frozen v6, or publish data.” | Completed accepted diagnostics, not promotion | Thesis | Matched; refresh after actual stage change only. |
| Register arithmetic | “676 modeled events” | Core summary: `"baseline_active": 675`, `"candidate_active": 676`, `"retired": 13`, `"new": 14` | Exact private roster arithmetic | Section | Matched. |
| Metadata versus public atlas | “separate counts” | Core: `"candidate_metadata_rows": 771`, `"inactive_audit_rows": 95`; current snapshot: “370 cases / 380 stops” | Distinct metadata/active/rendered units | Section | Matched. |
| Coding and untouched fields | “retain their exact stored values” | Verifier: “All37,844 untouched retained cells preserve exact score spelling and U/NA, including100 fractions.” | Literal parity outside reviewed retained cells | Section | Matched. |
| New/reviewed cell accounting | “840”; “927 changed and949…confirmed” | Core: `"reviewed_retained_cells": 1876`, `"changed_retained_cells": 927`, `"confirmed_retained_cells": 949`; verifier: “840 cells.” | Actual counts, not every reviewed cell changed | Section | Matched. |
| Families and holds | “editorial”; “not…clustering” | Family ledger: `"holds": 4`, `"classified": 672`, `"model_or_mechanism_used_to_select_family": false` | Four explicit holds; accepted or inherited labels, not inferred clusters | Thesis | Matched. |
| Retention and angles | “both retain nine”; “18.69°…11.78°” | Summary: `"retained_factors": 9`, `"angle_8_vs_baseline": 18.689066772609834`, `"angle_9_vs_baseline": 11.784825032329376` | Canonical agreement in count with changed spaces | Thesis | Matched. |
| Named congruence distinction | “separate comparison” | Verifier: “minimum named Tucker is0.077522.” Main: “it is not the separate optimally rematched whole-space comparison.” | Named-axis result must not stand in for optimal alignment | Thesis | Matched. |
| Legacy clustering | “17.98%”; “not…public editorial families” | Verifier: “Legacy k6 reassignment is17.975831% across662 shared rows (ARI0.632360), not an editorial-family classification.” | Old automatic partition diagnostic only | Thesis | Matched. |
| Scenario counts and retention range | “reuse”; “neither independent…nor probabilities” | Verifier: “The actual record distribution is524 Horn9,29 Horn8 and12 Horn7; separate battery omissions retain6–8.” Also: “These are not independent-experiment counts.” | Correlated scenario inventory, not replication strength | Thesis | Matched. |
| Program weighting | “676 real rows”; “illustrative” | Verifier: “Literal-program weighting keeps676 real rows with total weight583 and effective sample size615.777, and shifts8-space44.671536° versus canonical successor.” | Influence sensitivity without pseudo-observations | Thesis | Matched. |
| Boundary/fairness interpretation | “For this project”; “no…universal…threshold” | Main: “There is no prespecified universal pass/fail angle, and no rule to choose historical boundaries to improve stability.” | Owned methods interpretation, not proof against all possible taxonomies | Thesis | Matched. |
| Time/horizon distinction | “No separately measured continuous series” | Core: `"measured_duration_eligible": 0`; verifier: “Eligible nominal dates:628 baseline,627 successor; no separately measured duration series is invented.” | Dataset limitation, not absence of historical tempo evidence | Section | Matched. |
| Missing alternatives | “remain explicitly unestimable” | Verifier: “No separately coded642-only Egypt or internal Kitakami phase profiles were manufactured.” | Two specific absent raw profiles; other tests remain available | Section | Matched. |
| Descriptive cards/F8 follow-ups | “already approved”; “do not…close” | Main F8 instruction: “A public F8 disposition is not itself a sixty-dimension raw recode.” | Inherited public definition; six old raw obligations remain | Section | Matched. |
| Verification extent and AI provenance | “selected”; “not…external human” | Verifier: “No source/PDF research, new agents, full-suite refit”; “Model-performed, not human or external audit.” | Bounded AI verification with prior contributions disclosed | Thesis | Matched. |
| Final limits | “not proof” | Main: “They do not prove an unbiased census” and “universal traits” | No census/universality/fairness certificate claimed | Thesis | Matched. |

Concentration: no remaining confidence mismatch found in this bounded text.
The sensitive points were counts carried between stages, what a retained
factor count can establish, related scenarios masquerading as independent
evidence, and model verification versus outside historical review. They are
explicit rather than hidden in a general caution.

Reverse check: actively looked for underclaims. None found. The text states
the completed nine-factor retention, computed changes and accepted correction
counts directly. It does not hedge those results because broader inference is
limited. No unsupported-but-sound or uncheckable source claim was retained;
future publication status/link wiring remains a separately named task.

| Review self-audit | Status | Basis |
| --- | --- | --- |
| Direction balance | PASS | Matched claims retained; no manufactured downward edits. |
| Lane restraint | PASS | Calibration made no prose-style edits. |
| Deletion versus flag | PASS | No claim deleted to hide missing support. |
| Mandatory re-cite flags | PASS | None required; package-link work remains in this artifact. |
| Verbatim grounding | PASS | Each row has exact source fields/passages and named locator. |
| Coverage/skips | PASS | All scoped substantive paragraphs grouped; larger manuscript and future package expressly excluded. |

## Final publication status

CHANGED TEXT CHECK COMPLETE WITH PACKAGE-LINK/STATUS CONDITION. The three
passes cover the results narrative above, not the whole monograph or release.
Before publishing, replace source locators with working package links, recheck
the actual promotion/deployment status, and bind the resulting text bytes.
Those mechanical checks do not call for another historical audit. No public
artifact has been deployed by this text review.