DEED conversion campaign — tracking & acceptance criteria (.a2ml → .deed)

[hyperpolymath]

Status: tracking issue, opened 2026-09-17 after the owner ruling that the a2ml project is officially retired and that .a2ml artifacts are to be translated using the DEED grammar. No conversion PRs are attached yet — this issue is the campaign's acceptance surface. (The DEED README's (#64) reference for "the conversion campaign" resolves to an unrelated closed Hypatia issue; this issue replaces that dangling pointer.)

What DEED requires (implications for every translation)

• S-expression only. No key = value, no [section] productions — "a file using = is not a deed." Translation is a rewrite into target forms, not a surface transform.
• Four document forms, dispatched by FILENAME STEM before parsing: estate_chora.deed → estate-deed; ATLAS.deed → estate-atlas-deed; <stem>_praxis.deed → praxis-deed; <stem>_chora.deed → repo-deed. Normative side condition (not ABNF-expressible): in repo-file, stem ≠ estate, and estate_chora.deed also matches repo-file — implementations MUST test estate-file first (spec §chora-dispatch-exclusion; ruled on standards#752).
• Required on every form: :schema-version. Header = ;; SPDX-… lines.
• Every .a2ml artifact must therefore first be classified into one of the four forms; per-family mapping specs are the prerequisite of any mass translation.

Grammar/surface hazards to fix BEFORE any translator is written

1. 1-formats/deed/spec/abnf/deed.anbf and deed.abnf_v1.0 diverge — one must be ruled normative and the other deleted/aliased.
2. The spec links ../abnf/deed.abnf — that file does not exist.
3. The campaign had no real tracking issue until this one (README points at the closed, unrelated Hypatia dogfooding job red estate-wide — unresolvable setup-beam pins (companion to hypatia-side fix) #64).
4. The filename-dispatch ordering side condition must be hand-encoded in every validator.

Semantics that MUST survive translation (acceptance criteria)

These live only in about-to-be-translated surfaces and have no deed-era home yet — the campaign is not done until each is greppable in the deed era:

Semantic	Current home	Must land in deed era as
licence per-category default (MPL-2.0 code / CC-BY-SA-4.0 docs); PMPL reserved for palimpsest-license/plasma/consent-aware-web	1-formats/templates/STATE.a2ml.template:46 comment (fixed by #645)	template comment in the deed STATE/journal generator + licensing-policy successor
CLADE [lineage] type incl. hub/satellite (satellite parent stays "")	rsr-template-repo build/templates/CLADE.a2ml.in (fixed by rsr#122)	the repo-deed lineage clause — CLADE has no v2 surface: highest-loss-risk item
STATE status incl. planned (≡ CLADE phase reserved) + AUTHORITY rule (CLADE phase wins on disagreement; STATE is re-derived)	rsr-template-repo .machine_readable/descriptiles/STATE.a2ml:14	same repo-deed lineage/status clauses
STATE v2 doctrine: discard the derivable; keep resume-state only	1-formats/templates/STATE.a2ml.v2.spec.adoc	unchanged — deed design embodies it
Enum values have no canonical spec home (templates are the source of truth)	— proven during #726 close-out	deed spec chapter + conformance samples

Existing tooling honesty

state-migrate-v1-to-v2.sh / state-scm-to-v2.jl (1-formats/templates/) are STATE-only and target the pre-DEED @state thin journal — they are not .a2ml → .deed translators. No general translator exists. Recommend a single canonical campaign translator (avoid the pin-generator trap of N writers).

Recommended campaign order

1. Grammar hygiene: one normative ABNF; fix broken abnf/deed.abnf links; point the README at THIS issue. (Prepared but held for owner review.)
2. Mapping specs, one PR per family: CLADE/META/ECOSYSTEM/AGENTIC/NEUROSYM/PLAYBOOK → repo-deed clause mapping; STATE → journal/praxis-deed decision; scorecards → repo-deed-clause vs separate form.
3. Canonical translator + conformance lane: CI parses every translated file against the normative ABNF, including the estate_chora dispatch-ordering test.
4. Per-district conversion: standards .machine_readable/ (dogfood) → rsr-template mint sources → estate wave; flip assess.just RSR gates from .a2ml presence to .deed presence alongside.
5. Acceptance: the §above table verified greppable; enum vocab byte-preserved (hub|satellite|planned|reserved, PMPL carve-out wording).

Related

• PR fix(ci): echidna-verify — evict idris2-a2ml job (a2ml project retired; no ruleset pins its context) #836 — first artifact of the retirement ruling: idris2-a2ml CI job evicted with anti-deadlock ruleset verification.
• Orphan pin class: a reusable pin can be a real commit GitHub still refuses to run — 61 startup-dead rows, 0 alive; check-action-pins-resolve.sh tests the wrong predicate #782 (ancestry probe pattern) is the model for "detector before campaign" — the conformance lane in step 3 plays the same role here.
• Volume anchor: 70+ *.scorecard.a2ml in standards .machine_readable/, 6 template families, 435 estate repos with descriptiles (assess.just gates).

[hyperpolymath]

Redirect layer landed: PR #838 (open) fixes every broken pointer this issue's hazard table named — #64 → here in both spec copies, all abnf/deed.abnf links resolved per-doc-depth to the file holding the cited content, K9-spec links repointed, three extra dead links in a2ml/anchor/README.adoc repaired, and both grammar files now carry comment-only banners stating the two-file situation. No normativity verdict was stamped (no renames): the surviving decision is the canonical-file ruling (likely git mv deed.anbf deed.abnf + archive rename to deed.abnf_v0.1), which the PR body proposes for your ruling.

Also today: echidna a2ml-job eviction = PR #836 (open). Both PRs await merge; nothing in this campaign issue's order §6 starts before them.

[hyperpolymath]

Merge trail: PR #838 (pointer redirects) MERGED 2026-09-17 (f5ba975e). Hazards §4 items 2-3 are resolved by it; item 1 (canonical grammar file + rename) remains the open ruling. PR #836 (idris2-a2ml eviction) also MERGED (5574d3c4) — the deferred-corpus list in echidna-verify is now: lol + AVOW guarded-dormant, A2ML gone for good.

[hyperpolymath]

Progress: first campaign deliverable is up as PR #840 — the mapping-spec frame (1-formats/deed/mappings/README.adoc) plus family 1: CLADE.a2ml → repo-deed.

• Full field table; enums + closed taxonomies carried verbatim as symbols (incl. the harden(ci): estate-wide concurrency-cancel guard + scope affinescript-verify push #122 hub/satellite lineage values, else they lose their other home).
• P-1..P-5 provenance rules: uuid re-derive fail-closed (never copy); source must parse as a2ml first; every emitted deed parses against the normative grammar; estate stem dispatch fail-closed; CLADE-003/004/005/006 gate semantics re-expressed before the checker retires.
• :beholding-chora is resolved from the registry with refuse-to-emit absent — no invented identifiers.
• Worked translation of the real rsr-template instance, machine-checked structurally before commit.
• Three open questions recorded (§8), NOT decided in the PR.

Still gated on owner alone: the canonical-grammar-file ruling (rename/archival from #838's body). Everything else in §6 unblocked once #840 lands.

[hyperpolymath]

Documentary grounding for family 3: the v2 thin-journal doctrine that #843 cites now has its exact citation pinned — 1-formats/templates/STATE.a2ml.v2.spec.adoc (v2.0.0-draft, 2026-04-05), with live siblings STATE.a2ml.v2.template, state-migrate-v1-to-v2.sh, state-scm-to-v2.jl.

Two facts worth the record:

1. The v2 spec's "Removed v1 field → where it lives now" table independently corroborates docs(deed): mapping spec family 3 — STATE v1 decision spec (options + recommendation) #843's option B: session history → git log, completion % → VeriSimDB, milestones → ROADMAP, metadata → META, related projects → ECOSYSTEM. The doctrine predates the campaign; the campaign is its faithful continuation.

2. The v2 model (phase / blockers / last_action / next_action / session_marker / updated) survives the a2ml deprecation — its consumers (VeriSimDB STATE-journal octad, PanLL session-resume, coordination.k9 reflex guards) want the fields, not the container. The container's grammar (a @state … @end directive dialect) is deed-era dead, and its k9-init path was never finished. New open residual for the owner, alongside docs(deed): mapping spec family 3 — STATE v1 decision spec (options + recommendation) #843's maturity question: where does the live v2 journal live in the deed era — a (state-journal …) clause or consumer-side ingest elsewhere? Placement ruling, not mapping fact.

[hyperpolymath]

Conformance lane up: PR #849. The campaign now has its tooling half alongside the spec half (#840–#848 all merged). Highlights:

• deed_lint.py: full validator for deed.anbf v1.0.0 (14-case self-test + fixture dir green, incl. a REAL translated deed)
• a2ml_to_deed.py: implements exactly the mapping specs, nothing more. Full six-instance translation of rsr-template-repo passes end-to-end (uuid oracle verified: uuid5(URL, github.com/hyperpolymath/rsr-template-repo) == a5ea1382-…)
• Fail-closed caught three live un-tabled surfaces → PRs docs(deed): family-2 table extension — axis-3-audit-rules populated, ADR array format #846/docs(deed): family-4 extension — [methodology] tabled; D4-1 comment-canon doctrine #847/docs(deed): family-4 completion — PLAYBOOK [rsr-repo-skeleton] tabled #848 extended the tables exactly as the frame prescribes
• CI: deed-conformance.yml gates self-test → fixtures → every committed *.deed

Findings delivered, NOT silently fixed:

1. standards' own .machine_readable/CLADE.a2ml is missing primary-name (CLADE-006 territory) — the translator correctly refuses to translate it. Needs an instance fix by the owner (or a rule that template-era instances get repaired at mint-time).
2. Scorecard census (this repo's corpus): 28 files, 48 absolute-path leak lines (/home/user/standards/…). Full TSV rides with the residue artefacts for the owner-side clean per docs(deed): mapping spec family 5 — scorecard corpus decision + path-leak remediation #845's remediation.

Everything now executable without owner input is executed. Remaining gated on you: the rulings bundled earlier (canonical grammar filename; #843's STATE option + :maturity extension; #845's scorecard option B + journal placement; #840's three open questions), the actual conversion wave (needs those rulings + a repo where deeds may land), and #645's re-propagation run.