Acceptance scope

The executable acceptance command builds fresh studies, runs conformance and retains a machine-readable report. It marks a milestone passing only after the associated scenarios execute successfully. This document maps the criteria to the implementation. The current rc2 receipt records 113 passing tests and all I1–I5 scenarios; the v0.2 audited receipt in docs/audit/evidence/acceptance.json remains historical evidence.

python3 -m allagma acceptance --output build/acceptance

Criterion	Concrete passing evidence
I1 — portable foundation	Generic, Codex and Claude Code entrypoint fixtures resolve one identical bundle; three complete walkthroughs produce the same scientific summary without native hooks/subagents
I2 — replaceable composition	Independent context-only, reviewer-only and recipe-only replacements complete the workflow; shared module revisions are unchanged; controlled comparisons retain adoption and rejection decisions
I3 — contributor path	A fixture adds one module and runs its targeted example without an account; conformance, governance, ownership, issue/PR templates and authoring guidance are present
I4 — versioned studies	Central/profile isolation, update-stage receipts, generated-file conflicts, contract mismatch, coexistence, retirement, historical resume, complete rollback and separate scaffold migration/inverse
I5 — research walkthrough	26 successful runs (2 pilots + 24 confirmation), 1 failed and 1 interrupted attempt; raw manifest, recomputed tables, sensitivity, figure, supported/contradicted claims, manuscript and exact-revision audit; a separate limited-budget partial manuscript

The conformance receipt additionally includes controller crashes, worker timeouts after controller death, stale evidence, duplicate/overlapping inputs, late intent edits, local variants and recovery from interrupted adoption and multi-file migration/inverse operations. No skipped test is counted as passing.

Source inventories include canonical module files and bundled helpers. The acceptance runner checks that source and test files did not change during the run. Optional independent validation uses the published JSON Schema documents, a separate JSON Schema implementation and a YAML parser/skill validator.

Retained and reproducible evidence

The repository retains the current compressed acceptance tree and JSON receipts under docs/v0.3/evidence/rc2/acceptance/. The v0.2 archives in docs/audit/evidence/ and docs/evidence/ remain historical evidence. The acceptance archive includes every toy raw observation and attempt, exact bundles, protocols, comparison candidates, update histories, migration examples, manuscripts and review traces. Its inventory and SHA-256 are recorded in archive-manifest.json.

mkdir -p work/reproduction
tar -xzf docs/v0.3/evidence/rc2/acceptance/acceptance.tar.gz -C work/reproduction
python3 -m allagma campaign audit \
  --study work/reproduction/allagma-acceptance/generic --campaign toy-v1

Reanalysis dispatches to the archived campaign’s pinned helper. Historical absolute command paths describe the original execution; scientific artifact references are study-relative and survive relocation. Byte-level recomputation is checked in the recorded Python environment; record any environment change when interpreting a difference.

Scope of the result

The generic local path is executed. Codex and Claude Code are format/contract checked, with installed version probes when available. Native skill activation, hosted-model tool use and research quality are not claimed. The toy result tests a known analytical expectation, not Noemetric or a novel scientific hypothesis. The review is deterministic with explicit coverage, not independent peer review.

The complete machine-readable receipts and retained evidence are the basis for the milestone status. The presence of a manuscript alone is not acceptance.

Canonical source · Generated from the maintained repository document at this build’s commit.