Plan of record is docs/V6.md; backlog is docs/project-todo-v6.md.
This page yields to both. Every number below comes from the data island in this file's source,
counted by the rules in Methodology.
One chart carries the whole argument for v6. Each bar is a day's commits on
GOTT.Sibyla, split by what the commit did: feature and schema code, or
governance — records, reviews, rejections, amendments and approvals.
In July, 87% of commits were code (71 of 82). From 3 August it inverted to governance-heavy — 52% governance (150 of 290) — and the share keeps falling: 14–19 August ran 41 code commits against 24 governance ones, and those governance commits are reconciliation records, not gates. The single heaviest day in the project's history is 6 August at 61 commits; 5 August and 10 August tie behind it at 35 each, then 11 August at 33 and 7 August at 22. The early-August output — 5, 6 and 7 August, 118 commits between them — produced no user-visible feature. The 10 August bar is v6 itself: the plan, the revocation, the archive and the V6-O1 decision. Governance too — but the kind that ends governance. The two bars after it are what the decision bought.
The early-August code bars are honest but need a footnote: they are the P1-1a and P1-1b
migrations and their tests, not application features. That footnote stops applying on
11 August: the Layer 2 seeder, its parsers and the seeder CLI are application code, and
11–12 August carry 46 code commits — the reference-data seed for EXCRAT, DOCTYP and DOCEFL,
the effective-dating migration, and the party-collapse schema contract. Before that, measured
on application code rather than schema DDL, the last change was fix: authorize strict
WhatsApp LID senders on 24 July.
Elapsed project execution time per day — project sessions, subagents and autonomous
processes, parallel intervals merged globally so nothing is double-counted. Window:
2026-08-07 10:48:48 UTC (end of the earlier timing report) to
2026-08-18 15:58:56 UTC (last completion before the 2026-08-18 revision; not
re-measured for the 2026-08-19 revision), measured in two
increments — 89h 30m 07s to 2026-08-17 17:45:10 UTC, plus 14h 42m 26s from
hermes/state.db for 17–18 Aug. Total 104h 12m (exact: 104h 12m 33s)
across 12 days — ~8.7h/day. This is elapsed automation time, not billable human hours.
The heavy days line up with the heavy merges, not with commit counts. 15 Aug (16h 16m)
is S1.5 completion plus the item-master consolidation on 9 commits; 11 Aug (15h 35m) is
the Layer 2 seeding day; 13 Aug (13h 52m) is phase B — the destructive party-collapse
migration — on just 7 commits. Long execution against few commits is what a stop-list
migration being built and proven looks like; the inverse (10 Aug: 35 commits, 5h 24m) is what
an archiving/re-planning day looks like. 18 Aug (10h 08m) repeats the pattern: the
DOCTYP routing cutover, the stop-list DocumentTypeSetting drop and the Entity
360 shell — three merges of application code on long, test-heavy runs.
Five decisions taken on 2026-08-10. Their combined effect on the size of Phase 1:
The last row is the one that mattered most. v5.0 placed all ERP integration in Phase 2, which meant the stated core loop — document arrives, is catalogued, reaches the ERP — could not complete in Phase 1 by construction.
Estimates, not commitments. They exist so that a phase running at triple its estimate is visible as a signal rather than absorbed silently. All of them assume the light review method (V6-O1); if the current gate structure stays, multiply by roughly three.
Detail is capped at the current phase plus the next. S0–S2 are itemised; S3–S7 are sketched and get itemised when the phase before them closes. v5.0 itemised fourteen phases and delivered one.
Gates, reviews and decisions are not backlog items and never move the numerator. A phase closes on its exit criterion being demonstrated to a person, not on its boxes being ticked.
Nothing here is called worthless. Each carries the condition that would justify reinstating it, so it can be revisited on evidence rather than re-argued from scratch.
The condensed chronicle. Full narratives live in git history and the dated records, now
moved to docs/archive/ — this table points, it does not re-narrate.
This page renders its numbers from the embedded data island with JavaScript. The data is
readable in the EVO-DATA block in the page source. Regenerate the commit series
with git log --format='%ad|%s' --date=short and the classifier in the
Methodology table.