README.md presents self-development/ as the methodology applied to its own
development. That is true and it is partial, and until 2026-08-20 the partiality was nowhere stated —
so a brownfield adopter reading this directory as the worked reference could copy an omission and think it
was the convention. That is audit finding F-09.
Partial adoption is legitimate. Undeclared partial adoption is not. This file is the declaration.
Read the omissions and adaptations first. A profile that opens with what went well is marketing.
| Component | State | Why |
|---|---|---|
| Committed integrity checker / CI beyond a secret scan | omitted by decision (2026-08-20) | This repo ships no runnable elements. The cost is real: every published count is hand-maintained, and the same tag-count error shipped three releases running. RELEASE_EVIDENCE.md writes the commands down instead — weaker than a control, and stated as such |
Two-layer memory (08) |
omitted | This repo has no memory/ directory. The methodology’s memory discipline is exercised only by the fictional examples/example-project, not here. The memory admission test has therefore never run on real content |
House verbosity setting (08) |
omitted, kept published | Nowhere to set it until today. The sweep put it forward for retirement; maintainer decision 2026-08-20: keep it, marked unexercised — a MAJOR bump is too steep for one paragraph. Revisit at the 2026-11-25 pass |
UI verification loop, blast radius, most of 10 |
not applicable | Docs-only. There is no UI to verify and no consumer graph to enumerate. This is the single largest thing this instance does not exercise, and it is why STATUS.md says the self-application “exercises the planning, review, and memory disciplines far harder than it exercises the UI-verification and testing disciplines” |
| Cross-AI validation as a gate | adapted, then dropped for E10 | Run for E06/E07/E08 (E06’s returned 16 PASS / 2 FAIL, both real). Waived for E09 by decision, and removed from E10’s exit criteria by decision on 2026-08-20. The convention remains published for adopters; this repo no longer gates on it |
| Component | Adapted how | Why |
|---|---|---|
Epic file shape (03) |
Five files required; nine epics ran on four until 2026-08-20 | The rule was not weakened to match. TEST.md was backfilled into all nine as empty-but-present with a pointer to the real verification record in each ARCHIVE.md. No acceptance rows were reconstructed — back-filling scenarios that never ran would be fabricated evidence in a repo audited for asserting unperformed checks |
WIP cap (03) |
2, not the methodology default of 3 | Raised from 1 after E02. Never contended — nine epics, never two active at once — so there is no evidence for raising it further, and raising a limit that has never bound would be a change with no information behind it |
| Epic lifecycle | Five of nine closures ran charter → close inside one maintainer-directed session |
Real, and worth naming against ourselves: same-day charter-to-close is the norm here, which means the WIP cap is not the mechanism doing any work — maintainer attention is |
| Versioning | Self-development work ships as minor or patch; methodology/ versions independently |
Recorded in backlog/README.md |
| Autonomous loop | Tiered: T0/T1 loop-enabled on authoritative docs, T2/T3 maintainer-authored | Stricter than the published default. Every E10 item was T2 or T3 |
Strategy → pillars → epics → items; the BL-#### item format with its eight required fields; the
Status/Test enums and the done hard rule; frozen intent on approved goals and exit criteria; the
Needs clarification marker; Code Maps at Effort M+; the six DoD gates; the file lock protocol; the git
workflow including PR-only trunk; the AI-safety rules; semi-annual currency passes; milestone evaluation
with its rubric.
Plus, adopted the same day it first became possible: the context-integrity canary, now live in root
AGENTS.md. It could not have been adopted before — it needs a project instruction file to
demand a response marker, and this repo had none. It cost six lines, which raised that file’s
self-imposed budget from 50 to 60 rather than trimming the rule; a number invented the day before should
not veto a safety rule.
Three of these were first exercised on 2026-08-20 — the Needs clarification marker, the canary, and,
under real pressure, frozen intent. The sweep records that 7 of 16 conventions added v1.25.0 → v1.31.0
have ever been used at all, four of them for the first time that day.
E11 explicitly adopts A2 for this material operating-rule change. The human repository maintainer owns intent, capability coverage, and acceptance; the AI author performs local checks. Independent review and human acceptance remain pending, so the items are not done. The new profiles do not authorize their own acceptance or change the authority of the reviewed base rules.
This session has one sequential implementer, so parallel-squad numeric item limits are not applicable here. The existing two-active-epic cap still binds: E10 and E11 now occupy both slots. The earlier “never two active” observation in the adaptation table describes the pre-E11 history. Software-squad walkthroughs are illustrative; A0–A3, item queue limits, rotation readiness, and asset graduation remain unvalidated in external adoption.
examples/example-project
is a worked sketch, not evidence.STATUS.md: one production project, one
docs-only self-application, no external adopters, and a closed-beta verdict of NOT READY.