# Main construction decisions

2026-09-07. Craig authorized continuation. These are prospective implementation
decisions for a private analytical candidate, not results, global admissions,
or permission to promote or publish a release. No model results were consulted.
The source/coding decisions remain in their accepted packages.

## Roster and overlap

Replace `ancient-qin-nanyang-clearance-convict-colony` with the three accepted
Rang281, Nanyang280 and Yan-Deng279 definitions in this separate candidate.
Qualification, full independent coding/adjudication, metadata and source/cohort
limits are closed. The parent remains in inactive audit metadata. No parent
and child coactivation is allowed. Linked-program and historical-parent
comparisons must disclose possible cohort overlap and the shared received
annalistic tradition; small model effects cannot validate these boundaries.

Combine the accepted retained checkpoint with all14 accepted new definitions
by exact ID, never array position. The resulting expected roster is676:
675 baseline minus13 retirements plus14 replacements/additions. This count is
a consequence, not a target. Family holds remain active statistical rows;
held prospective events remain outside the active matrix. Public/frozen v6
and the14-row inactive source sidecar are unchanged.

Append the nine identity-only absent-parent audit records already proposed in
JOIN_PREPARATION. They carry no dates, scores, or reconstructed history and do
not enter any factor matrix. Retain all existing inactive parents as well.

## Comparable macroregions

Use WEST_ASIA for both Sidon–Babylon–Susa and Shapur's260 campaign program in
both derived comparison joins. Keep each original corpus and legacy normalized
assignment unchanged and expose the override in analysis metadata.

- Sidon's bounded source and the two documented receiving places are all
  within the established West Asian category. “Eastern Mediterranean” does
  not create an additional macroregion here.
- Shapur's accepted source program concerns Syria, Cilicia/Cappadocia and
  aggregate placement in named Sasanian regions. The accepted pre/post summary
  and B49-02 retain Persis, Parthia, Khuzistan and Asoristan as regional context,
  not a North African event or an allocation of each city to each destination.
  This mapping must not substitute the ruler's empire for the bounded program.

Accept the remaining exact-ID/approved-alias crosswalk inheritance. Main read
the ten accepted affected-population changes: Matiatu, Unqi, Miletus, Xianyang,
Faesulae, Norba, Volaterrae, Nola, Suessula and Shapur. They refine which people
are measured, or the evidence for their outcomes, without placing the bounded
event outside its existing macroregion. Preserve each changed referent verbatim.
The14 new mappings follow MAIN_AUTHORING_DECISIONS without further inference.

## Dates, measured duration, and observation horizons

Accept the single legacy-date normalization proposed in INPUT_GAPS §4 and
implemented by `propose_legacy_date`. Preserve explicit flags and numeric
coordinates; absent flags mean no affirmative proxy designation, not exact-date
certification. Exclude current-null/proxy events from both continuous-date
comparison subsets, including the two explicit old ordering-only exceptions.
Approximate nominal starts can enter this expressly labeled diagnostic.
All such events still enter the ordinary factor analysis.

No separately bound, measured transport/influx-duration series is supplied for
the inherited rows. Mark this additional continuous-duration diagnostic
unavailable for both datasets, with that input limitation as the reason.
Do not infer that historical tempo is unknowable, recode any existing tempo
indicator, or replace missing duration with end minus start. The accepted three
tempo indicators retain their original, distinct measured constructs. Any
previously prescribed calendar-envelope weighting is a separate illustrative
stress test and must say it does not measure movement duration.

For inherited rows, bind the unchanged60-dimension codebook as the default
observation-horizon rule, rather than assigning one common elapsed period to
all constructs. Its operative, generational, consolidation and long-run
constructs keep their respective definitions and anchors. Preserve the accepted
per-event metadata_date_semantics plus coding-output and instruction locators
as overrides where applicable. No new date, descendant cohort or geography can
be borrowed to fill a score. For the14 new units, retain their explicit
operation/follow-up objects and source-use overrides. In particular IB06 may
use linked aftermath for the original cohort where its construct requires it,
without backdating later legal instruments or importing new troop cohorts.

This records what the accepted instrument and corrections require. It does
not claim that every inherited score received a new horizon audit.

## Execution still pending

Main is building core inputs. Boole's separate program_projection directory
is translating existing finite instructions, not authoring new historical
alternatives. Numerical alternatives require independently coded profiles;
unavailable ones stay explicitly unestimable. Family-only comparisons stay
editorial. Main must review the actual program projection and complete inputs
before creating execution acceptance. No fits have run.