docs(specs): sprint-3 Tier-1 implementation specs (11 PR-X-1 + sequencing + smoke test + consumer template)

[AdaWorldAPI]

Summary

Sprint-3: convert the 7 DESIGN-PHASE patterns (A, B, C, D, E, F, J) and the 3 trivia closures from sprint-2's TD entries into PR-ready implementation specs. Engineer can now execute Tier-1 in ~6 working days.

Predecessor

PR #358 (sprint-2 architecture synthesis) named the 15 patterns A-O and recognized ~80% as already shipped. PR #359 corrected the tier-0 canonical pattern letter assignment. This PR (sprint-3) converts the remaining ~20% into engineer-ready specs.

What this PR does NOT change

• Zero crate source code touched (specs only — implementation comes in sprint-4+ PRs)
• No CI / build / test changes
• Append-only governance preserved on all coordination files

What this PR adds

11 PR-X-1 specs (.claude/specs/)

File	Pattern	LOC est.	Effort
sprint-3-execution-plan.md	(master)	n/a	n/a
pr-a-1-spo-g-u32-slot.md	A	~300	medium ~2 days
pr-b-1-context-bundle.md	B	~200	small ~1 day
pr-c-1-generic-bridge.md	C	~200	medium ~1-2 days
pr-d-1-fma-owl-hydrator.md	D	~600	medium ~3-4 days
pr-e-1-manifest-modules.md	E	~330	medium ~2 days
pr-f-1-ractor-supervisor.md	F	~400	large ~3 days
pr-j-1-int4-32d-atoms.md	J	~120	small ~1 day
trivia-prs-bundle.md	(3 reframes)	~60	trivial
consumer-crate-template.md	(proof)	n/a	n/a
ogit-g-smoke-test.md	(validation)	~200	small
sprint-3-pr-graph.md	(sequencing)	n/a	n/a

Total LOC for sprint-4+ implementation: ~2,470 LOC across ~10 PRs.

Coordination (.claude/board/sprint-log-3/)

• SPRINT_LOG.md (main-thread scaffolding pre-written before agent spawn)
• agents/agent-W{1..12}.md (12 per-agent append-only logs)
• meta-1-review.md (brutally honest review)
• sprint-summary.md (closure)

Critical path (sequenced PR order, per W10 sequencing analysis)

Foundation (parallelizable):
  PR-B-1 (W3) — ContextBundle              [foundation]
  Trivia bundle (W12) — 3 quick wins       [parallel; <1 day total]

Tier 1 wiring (Week 1-2):
  PR-B-1 ──→ PR-A-1 (W2) — SPO-G u32 slot
  PR-A-1 + PR-B-1 ──→ PR-C-1 (W4) — GenericBridge
  PR-B-1 ──→ PR-J-1 (W7) — INT4-32D atoms
  PR-A-1 + PR-B-1 ──→ PR-D-1 (W9) — FMA OWL hydrator (also PR-ANATOMY-1)

Tier 2 supervised mesh (Week 2-3):
  PR-B-1 ──→ PR-E-1 (W5) — manifest.yaml + build script
  PR-C-1 + PR-E-1 ──→ PR-F-1 (W6) — ractor supervisor

Validation (Week 3-4):
  All Tier-1 + Tier-2 ──→ Smoke test (W11)
  All Tier-1 + Tier-2 ──→ Consumer template dry-run (W8; hubspo-rs in <150 LOC)

Critical path: ~11 working days serial / ~6 working days parallelized.

Architectural validation milestones (in execution order)

1. Tier-1 lands — ContextBundle queryable; SPO-G u32 threads; GenericBridge dispatches
2. Tier-2 lands — manifest.yaml drives compile-time consumer binding; ractor supervisor mounts per-G actors
3. Smoke test green — Healthcare consumer dispatch end-to-end through 5 patterns
4. Consumer template dry-run — hubspo-rs in <150 LOC validates the ~12-18× per-consumer reduction (vs medcare-rs's 1865 LOC)

Corrections applied during sprint

• W9-rev2: W9 first attempt pushed to wrong repo (AdaWorldAPI/ada-consciousness instead of AdaWorldAPI/lance-graph; deferred to GITHUB_REPO env var). Main thread recovered the spec verbatim and pushed to lance-graph via pygithub. Provenance note prepended for archaeology. Same wrong-repo failure mode as W7 in sprint-2. Future sprints should add explicit guardrail.

CCA2A protocol upgrades validated

• pygithub-first with quote-stripped GITHUB_TOKEN for main thread (worked); agents fell back to MCP (sandbox proxy doesn't expose REST endpoint)
• Pre-written SPRINT_LOG-3.md scaffolding — agents saw coordination state from turn 0
• Canonical pattern letters embedded in every prompt — no pattern-letter divergence this sprint (vs sprint-2 W2's invented A-G)
• Branch + repo pre-verification before agent spawn

Test plan

• All 12 spec files exist on branch at expected paths (verified via pygithub repo.get_contents)
• All 12 agent logs present (verified)
• PR-D-1 spec on lance-graph (W9-rev2 correction applied; original ada-consciousness push is harmless residue)
• Cross-references between sister specs verified (e.g., PR-B-1 → ContextBundle for PR-A-1/PR-C-1/PR-D-1/PR-E-1/PR-F-1/PR-J-1)
• Critical path graph (W10) reconciles with each PR's stated dependencies
• Append-only governance preserved on coordination files

Notes

• This is specs-only, not code. The 11 PR specs constitute sprint-4's execution backlog.
• The trivia-prs-bundle.md 3 quick wins (PR-CAM-DIST 1-line, PR-ADJ-THINK ~30 LOC, PR-DEEPNSM-NSM ~30 LOC + 5 deletes) can ship in parallel with the critical path; they close TD-CAM-DIST-REGISTRATION-9, TD-ADJ-THINK-EXPOSE-10, TD-DEEPNSM-NSM-COLLAPSE-11 and reframe the corresponding ledger rows.
• Consumer template (W8 spec) is the architectural validation gate: if hubspo-rs requires >300 LOC of glue, the GenericBridge / ConsumerPointer design needs revisiting before more consumers land.

Branch state

• HEAD: 6dab08ac (post W9-rev2; meta + summary commits land on top)
• ~28 commits total

🧠 Engineer can now execute Tier-1 in ~6 parallelized working days. PR-B-1 first; everything else fans out from there.

🤖 Generated by Claude Code

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: b07858a0f7

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Skip inert bundles when spawning actors

When the registry contains the required inert bundles from the manifest spec (for example DOLCE and FMA have consumer_pointer = None), this loop treats the missing pointer as an ActorProcessingErr, so CallcenterSupervisor::pre_start aborts before spawning any real consumer. The smoke-test spec explicitly expects DOLCE to have no pointer while Healthcare still spawns, so the supervisor needs to iterate only bundles with a consumer pointer (or continue on inert Gs) rather than failing on them.

Useful? React with 👍 / 👎.

[chatgpt-codex-connector]

Keep consumer actor types out of the contract crate

Emitting medcare_rs::MedCareActor from lance-graph-contract means the contract crate must depend on the consumer crate whenever module-medcare is enabled, but the consumer template makes each consumer depend back on lance-graph-contract; enabling the feature therefore creates a Cargo dependency cycle and the required feature-enabled build cannot compile. The concrete actor registration needs to live in an outer crate/binary or use a callback/registry mechanism instead of generated code inside the contract crate.

Useful? React with 👍 / 👎.

[AdaWorldAPI]

Spec corrections appended (post-review, pre-merge)

Two real architectural defects flagged during review have been corrected via append-only sections at the end of the affected specs:

Defect 1 — PR-F-1 supervisor inert-bundle handling

pre_start would have unwrap-panicked or aborted on inert bundles (DOLCE, FMA have consumer_pointer = None by design per W11 smoke test). Correction at end of pr-f-1-ractor-supervisor.md: explicit match that continues on None, plus new test supervisor_skips_inert_bundles_and_spawns_consumers.

Defect 2 — PR-E-1 Cargo dependency cycle

Build script in lance-graph-contract was specified to emit use medcare_rs::MedCareActor references — but contract is consumed by every consumer crate, creating a cycle when module-medcare feature is enabled. Correction at end of pr-e-1-manifest-modules.md: build script emits data only (OGIT::* u32 constants + MANIFEST_METADATA phf::Map of strings/values, no Rust refs). Actor registration moves to consumer-side inventory::submit! macro (Option A, recommended). No cycle.

Why append-only

Both PR-F-1 and PR-E-1 specs preserve their original content unchanged; corrections are appended sections at the end. The engineer picking up implementation reads the correction sections as authoritative for the contested sub-areas. This matches the APPEND-ONLY governance pattern used across .claude/board/ files.

Affected sister specs

• W8 consumer template (consumer-crate-template.md) — references PR-E-1's manifest pattern; its ~50 LOC sample should grow ~5 LOC for the inventory::submit! macro line. No spec rewrite needed; engineer applies during implementation.
• W11 smoke test — explicitly expects DOLCE inert + Healthcare active; the PR-F-1 correction is exactly what makes that smoke test pass.

Both correction commits land on the sprint-3 branch this PR tracks; no separate fix-PR needed.