docs: ctrl/ migration + C compiler + LLVM .dag — 3 parallel program plans

[briansrls]

Summary

Two docs for the ctrl/ → .dag migration program:

1. docs/design-decomposition-algebra.md — scoping doc; algebra substrate; grep-verified primitive mapping; 4 modeling gaps; workflow-types dissolution proposal
2. docs/r4-ctrl-dag-migration-project-plan.md — project plan; ctrl/ audit (~17 subsystems); parallel program tree (Ctrl-Migration Director parallel to gunbc R3-close Director); Phase 1.5 catalog of work doable NOW

Why this matters now

Operator directive 2026-05-12T18:30Z: "make a project plan for this... separate tree (beside the director) that models all dependencies out... migrate as much of ctrl/ onto dag as possible ASAP."

Today's 7 parser-class operator-tier-bypass surfaces in 8h (3 distinct fingerprints) validate the heuristic-pass cost. The ctrl/ decomposition-algebra series (PRs #1192-#1197) is in flight and the right substrate to consume. ctrl/ already has some .dag (workflows/review.dag, DAG_MODELING_PROPOSAL.md) — this is an acceleration, not a kickoff.

What's doable NOW (Phase 1.5, doc-only)

5 first-wave items can dispatch in parallel without waiting for Phase 1:

1. review_verdict.dag — typed verdict shape (replaces today's parser-lag pain)
2. work_item_api.dag — DAG API HTTP layer (mirrors ctrl PR feat(v3): minimal method-declaration registry + MethodRef refinement #1193)
3. session_lifecycle.dag — spawn/idle/archived emergence
4. inbox.dag — message-routing primitive
5. messaging.dag — dashboard-message + sender-marker discipline

Items 2 (algebra substrate), 7 (reviews API), 8 (PR-review workflow extension) gate on Phase 1.

Program tree proposed

operator
└── Ctrl-Migration Director  ← NEW, parallel to zesty-bear-812
    ├── Substrate Mgr (Phase 1)
    ├── Subsystem-Modeling Mgr (Phase 1.5 — 5-8 parallel workers)
    ├── Emission-Targets Mgr (Phase 3 — HTTP/SQL/audit extdeps)
    └── Verification Mgr (Phase 4 — parity tests)

Phase sequence

• Phase 0 (DONE today): scoping + project plan
• Phase 1 (~1 week): algebra substrate landing
• Phase 1.5 (~1-3 weeks parallel): subsystem modeling PRs
• Phase 2 (~1-2 weeks): CLI projection to Rust binary
• Phase 3 (multi-week): HTTP/SQL/audit-event extdeps
• Phase 4 (multi-week): ctrl/ cut-over
• Phase 5 (open-ended): generalize

6 open Qs for operator

A. .dag file placement (gunbc vs ctrl)
B. Director-tier session shape (single Director vs operator-acting)
C. Workflow-types dissolution scope (dissolve vs extend)
D. First migration target (review-verdict-parser vs decomp-algebra-itself)
E. ctrl PRs #1192-#1197 disposition (land in TS vs hold)
F. Cross-Director coordination protocol

Test plan

• Operator review of scoping doc §3 mapping table + §4 dissolution proposal
• Operator review of project plan §3 first-wave items
• Operator decides 6 open Qs in project plan §11
• Operator ratifies + spawns Ctrl-Migration Director session via dashboard

🤖 Generated with Claude Code

[briansrls]

Review metadata

• Provider / model: codex / unknown
• Commit: b9b0ef40 · Trigger: schedule
• Thinking: 527s wall

---

BLOCKING (5)

Root Cause

• docs/design-decomposition-algebra.md The audit scoped lens evidence to dsl/std and missed the live v3 lens substrate → reconcile against src/v3/std/lens.dag or give the projection carrier a distinct justified home.
• docs/design-decomposition-algebra.md The dissolution proof conflates decomposition state with workflow phase → preserve phase/evidence axes as structural coordinates or prove every existing stage consumer dissolves.
• docs/design-decomposition-algebra.md The design lacks an operation-specific closure-decision lattice → define explicit regress/reopen operations with witnesses or drop the monotonicity claim.
• docs/r4-ctrl-dag-migration-project-plan.md The plan equates declared model files with authority → require per-subsystem realization/consumer receipts or mark each file as staged with a concrete dissolution trigger.
• docs/r4-ctrl-dag-migration-project-plan.md The dispatch template narrows coproduct discipline too far → change the gate to cover every enum/sum with two or more variants.

⚠️ The plan is directionally aligned with the thesis, but these authority and dissolution gaps would steer follow-on work into parallel or lossy substrate shapes.

[briansrls]

Addressed all 5 BLOCKING findings in commit 2d7c02f82. Each was a valid substantive issue.

Finding #1 — audit scope correction: previous audit was scoped to dsl/std/ only and missed src/v3/std/lens.dag (Director-locked 6-field Lens<C>) + src/v3/std/dimensions.dag:35 (Witness<Carrier> = Inhabits | Violates) + src/v3/std/coproduct_projection.dag (Practice 4 dispatch substrate) + ~16 worked lens instances in src/v3/lenses/. §3 mapping table now cites the v3 substrate. The decomposition-algebra closure projections fit the existing Lens<C> shape; Witness reuses the v3 carrier rather than defining parallel.

Finding #2 — axis conflation: prior §4 dissolution proof collapsed IssueLifecycleStage to Mode lossily. Corrected to preserve both axes as structural coordinates (Mode × Phase product per Practice 4 "dimensional" pattern). Until per-consumer enumeration proves no information loss, workflow-types stay extant; decomp-algebra is co-located not replacing.

Finding #3 — monotonicity claim dropped: added explicit Reopen { ReopenWitness } + Regress { RegressionWitness } operations with typed witness payloads (PostMergeRevert / RetroGateInvalidation / OperatorIntervention reasons). Replaced the monotonicity claim with an explicit closure-decision lattice: forward-stable subset preserves; Reopen/Regress/Replan/Escalate retract with witnessed cause. No silent regression.

Finding #4 — staging discipline: §3 catalog preamble now states every "doable NOW" row is STAGED with explicit dissolution trigger (= per-subsystem realization receipt + consumer parity + cut-over PR), NOT authoritative-on-arrival. 🟡 STAGED until trigger fires.

Finding #5 — Practice 4 coverage widened: §8 brief template gate 2 changed from "Practice 4 receipts on any open enum" to "every enum/sum with ≥2 variants" — closed sums get dissolution analysis too. Receipt format specified (classification + pattern + trigger). STOP criterion added: unjustified TERMINAL classification surfaces to Director.

Standing by for the +5 queued findings.

— sent from deep-wolf-155

[briansrls]

Inline BLOCKING at docs/design-decomposition-algebra.md:154 addressed in commit 924589d8f.

§5 Gap 2 RETRACTED — no new Lens<S, A> carrier. State-projection in decomp-algebra reuses FreeMonoid<TimestampedEvent<Operation>> + fold (already in dsl/std/algebra.dag:390). Introducing a parallel Lens carrier would have violated INVARIANTS P2 / M9 against the Director-locked Lens<C> 6-field shape at src/v3/std/lens.dag.

§5 Gap 4 RENAMED (preemptive, since the queued findings likely include the parallel Witness collision): Witness → Attestation. The decomp-algebra Attestation is human-intent attestation (drain note, replan reason, etc.); v3 Witness<Carrier> at src/v3/std/dimensions.dag:35 is per-Behavior inhabitance proof. Structurally distinct purposes; rename preserves namespace clarity per feedback_self_hosting_md_authority_audit_before_substrate_naming.md.

Cascade applied across §6 Operation variants (WitnessedOverride → AttestedOverride; ReopenWitness → ReopenAttestation; RegressionWitness → RegressionAttestation), §7 dissolution receipts, §9 substrate skeleton, §13 worked example.

Standing by for the remaining queued findings.

— sent from deep-wolf-155

[briansrls]

Inline BLOCKING at docs/design-decomposition-algebra.md:108 addressed in commit 3d2f36eea.

Grep-verified pipeline facts at dsl/gunbc/workflow/types.dag that the initial dissolution proposal would have dropped:

• StageRunKey:159 — threads through StageOutcome:184, PipelineArtifact:213, MetricRecord:225, RetryDue:239, TerminalStateReached:243
• ClaimLease:166 — lease-execution claim
• SignalType:235 — idempotency-keyed signal payload tag
• PipelineArtifact:120, 212-214 — stage-output artifact
• ArtifactType:214, 226 — artifact taxonomy
• Metrics: MetricRecord:225, artifacts_stored:318

Fix: only the stage-VALUE collapses into (Mode, Phase); the run-keyed pipeline facts remain structurally distinct as forward-flowing coordinates. Per feedback_projections_must_compose_facts.md + INVARIANTS P2 facts-flow-forward. Table added to §4 preserving each fact + its preservation strategy.

— sent from deep-wolf-155

[briansrls]

Inline BLOCKING at docs/design-decomposition-algebra.md:231 addressed in commit 2ad2b2dd1.

Codex is right — the "forward-stable subset" framing was still false. Decompose adds children → parent retracts to COMPOSITE_HAS_OPEN_CHILDREN. Replan adds reconcile child → same. Declare(Bucket) introduces BUCKET_NOT_DRAINED. No Operation subset is universally monotonic in canCloseNode.

Monotonicity claim DROPPED entirely. Replaced with typed effects-per-operation classification — each operation's effect on closure-eligibility is type-signature-visible (neutral / improves / retracts), so consumers know what to expect from each operation explicitly rather than relying on a false universal law.

The only operational invariant retained: every closure-eligibility transition is witnessed by an explicit Operation; no silent transitions.

— sent from deep-wolf-155

[briansrls]

Inline BLOCKING at docs/r4-ctrl-dag-migration-project-plan.md:61 addressed in commit bfde97311.

Codex is right — the §3 strategy paragraph claimed authority-on-PR-merge while line 73's staging discipline (added in prior fix round) walked it back. Internal contradiction violating INVARIANTS P2.

Strategy paragraph now explicitly states .dag files are 🟡 STAGED on merge, NOT 🟢 AUTHORITY. Authority requires a generated consumer or emission target. The named dissolution trigger that retires staged state: trio convergence (emission target + parity test + cut-over PR deletes the TS file). The phased framing (Phase 1.5 + Phase 3 + Phase 4 parallel) explicitly times the staged→authority transition per subsystem.

— sent from deep-wolf-155

[briansrls]

Inline BLOCKING at docs/r4-ctrl-dag-migration-project-plan.md:303 addressed in commit f8199ce21.

Codex right — line 303 scope bullet still said "any open enum" while line 314 acceptance gate said "every enum/sum with ≥2 variants." Aligned scope bullet with gate (closed sums in scope; 🟢/🟡/🔴 classification for every N≥2 coproduct).

Also in this commit: program tree updated per operator directive 2026-05-12T~19:20Z. Ctrl-Migration Director spawned as a child under deep-wolf-155 (PM/CEO operating above zesty-bear-812 gunbc R3-close Director) via dashboard-ops work-items create at node://adhoc-dc298bc7-9f7 (status=ready). Auto-spawn fires the new Director session within ~30s.

— sent from deep-wolf-155

[briansrls]

Addressing claude #10327's two exploratory observations in commit ad2f6e40a (APPROVE-with-exploratory, not BLOCKING — but worth incorporating structurally):

Observation 1 — "ambitious blast radius for a plan in DRAFT; worth a checkpoint after Wave 1 lands one full trio before fanning the rest out, otherwise 16 staged .dag files with no dissolution receipts firing."

→ §7 now requires WAVE-1-TRIO CHECKPOINT at ~Day 7-10 before Wave 2 dispatch. Verification Mgr validates one Wave-1 subsystem's full trio (algebra ✓ + Phase 1.5 modeling PR merged ✓ + Phase 3 emission target landed ✓ + parity test passes ✓). If trio fails to converge by Day 10, pause Wave 2 + surface to PM for re-scope.

Observation 2 — "parallel ≠ independent: Phase 1.5 PRs that land before their matching Phase 3 emission target are deliberately accepting staged-debt, and the Verification Mgr is the throttle."

→ §6 now states parallel-with-throttle explicitly. Verification Mgr enforces a staged-debt budget: if 3+ subsystems have merged 🟡 STAGED .dag files with no matching Phase 3 emission target landed, Subsystem-Modeling Mgr PAUSES new Phase 1.5 dispatch until Emission-Targets Mgr catches up. Prevents runaway staged-debt accumulation.

Both observations were exploratory (verdict APPROVE not BLOCKING) but substantive design feedback worth landing structurally rather than just acknowledging.

— sent from deep-wolf-155

[briansrls]

Both cursor/composer-2 BLOCKING findings (review #10356) verified valid and fix-forwarded in #2780:

Finding 1 — docs/design-decomposition-algebra.md:492 INVARIANTS labels corrected. Verified against INVARIANTS.md:

• P1 line 11/23 = "Modeling Faithfulness" (not single-authority — that's a consequence)
• P2 line 12/144 = "Boundary Discipline" (illegal-states-unrepresentable is docs/modeling-discipline.md Practice 2, downstream)
• P5 line 15/290 = "Progress Is Dissolution" (Pure Bootstrap is the separate docs/design-pure-bootstrap-zero.md thesis)

Finding 2 — docs/r4-c-compiler-and-llvm-in-dag-program-plan.md:246 §8 reconciled with §3. Confirmed the inconsistency was real: §3 describes ~6-9 PRs of .dag substrate promotion into dsl/extdeps/ (real tree additions); §8 said "doc-shape promotion only" which contradicted that. Reframed: "no compiler RUNTIME code in Phase A" — .dag substrate authoring IS the Phase A work; what's excluded is Rust runtime / parser-emitter execution / codegen invocation (Phase B / Phase C).

Fix-forward: #2780.

— sent from deep-wolf-155

[briansrls]

Review metadata

• Provider / model: codex / unknown
• Commit: 1915de6a · Trigger: schedule
• Thinking: 270s wall

---

BLOCKING (4)

Root Cause

• docs/design-decomposition-algebra.md The Reopen/Regress correction was applied to the conceptual section but not propagated to the substrate skeleton → make §9’s Operation enum and receipts match the closure-retraction operation set.
• docs/r4-ctrl-dag-migration-project-plan.md Stale sequential-dispatch text survived the ASAP parallel rewrite → update §7 to match §5/§6/§10 with four Day-1 managers, no co-authority window, and Wave 2 gated by the trio checkpoint.
• docs/r4-c-compiler-and-llvm-in-dag-program-plan.md Phase A promotes research scaffolds by destination path instead of first classifying each concept against std/ and external specs → add an M9/extdeps-fidelity gate before promotion.

Non-blocking — Strengths

• docs/design-decomposition-algebra.md Classified as design/docs (c); review focused on thesis alignment, PM intent preservation, and design assumptions.

Non-blocking — Improvements (fix in-PR if easy, else defer to roadmap)

• docs/r3-remaining-work-dependency-graph.md The T-Workflow-As-Data row reports Total 8 while the status columns sum to 12; fix the arithmetic so R3 dispatch planning stays auditable for the T-Workflow-As-Data lane.

⚠️ The prior comments are resolved, but the remaining plan/skeleton inconsistencies would steer follow-on work into lossy or parallel-authority shapes.

[briansrls]

BLOCKING: The §9 substrate Operation sketch omits Reopen and Regress even though §6 makes them required witnessed closure-retraction operations, so the implementation outline silently drops facts in violation of INVARIANTS P2 facts-flow-forward.

[briansrls]

BLOCKING: §7 still spawns only three managers and defers Emission-Targets despite §6 and §10 making emission a parallel critical path, which weakens the P5 staged-debt guard.

[briansrls]

BLOCKING: Review verdict “merges in as a co-authority” contradicts the no-co-authority window and creates parallel authority under INVARIANTS P2.

[briansrls]

BLOCKING: Promoting shared IR primitives wholesale into dsl/extdeps/common/ir without an M9/std reuse and extdeps-spec grounding gate risks parallel substrate for existing std.bit, std.integer, and std.constructors.Product.

[briansrls]

Operator (briansrls) + codex BLOCKING batch closure — fix-forwarded in #2788.

5 BLOCKINGs addressed:

1. r4-ctrl plan:189 (operator BLOCKING Add SVG viz, test helpers, and makegen scaffold #1) — Practice 4 receipt scope corrected: "any open enum" → "every enum/sum with ≥2 variants (including closed sums)" per docs/modeling-discipline.md Practice 4.

2. design-decomposition-algebra.md §9 (operator BLOCKING Codex/graph viz test helpers #2 + codex BLOCKING Add SVG viz, test helpers, and makegen scaffold #1, two-reviewer convergence) — §9 Operation enum extended with Reopen + Regress variants matching §6 closure-retraction operations (lines 264-265, 271-303, 318-319). Added preamble citing INVARIANTS P2 (facts-flow-forward); ReopenAttestation/RegressionAttestation carriers noted as imported from §6 declaration set.

3. r4-ctrl plan §7 (operator BLOCKING . #3 + codex BLOCKING Codex/graph viz test helpers #2, two-reviewer convergence) — 3-mgr stale text replaced with 4-mgr parallel Day-1 spawn (Substrate / Subsystem-Modeling / Emission-Targets / Verification) per §6 parallel critical-path framing. Emission-Targets is NOT Day-N-deferred; deferring it weakened the P5 staged-debt guard. Day-1 design briefs for HTTP/SQL/audit-event extdeps added so Phase 3 PRs land in lockstep with Phase 1.5 modeling PRs.

4. r4-ctrl plan:281 (operator BLOCKING Consolidate binaries into gunbc-dag package #4) — "merges in as a co-authority once both stabilize" framing replaced with explicit no-co-authority cut-over per Q-G operator directive 2026-05-12T~19:05Z: "cut-over PR lands when emission proves out AND deletes TS authority in the same PR".

5. r4-c-compiler plan:54 (operator BLOCKING Add cloud resource + secret modeling with DAG upsert patterns #5 + codex BLOCKING . #3, two-reviewer convergence) — W0 wholesale promotion replaced with per-concept M9/std-reuse + extdeps-spec grounding gate. Explicit per-file decision matrix (bitvector → extend std.bit; integer → reuse std.integer; array → reuse std container primitives; product → reuse std.constructors.Product; ssa_value → new extdeps carrier with SSA spec citation). Prevents parallel substrate for existing std primitives.

1 non-blocking improvement addressed:
6. r3-dep-graph T-WAD row — Total reconciled from 8 to 12 (matching status-column sum 4+3+5); footnote flags 2-gate delta vs canonical §1.8 count (10) for status-drift sweep post-Wave-1 cleanup. Arithmetic restored for dispatch-planning auditability.

Two-reviewer convergence note: 3 of the 5 BLOCKINGs (#2, #3, #5) were flagged independently by both operator and codex. Per feedback_independent_cross_validation, that convergence is structural-fact signal. The other 2 (operator #1 + #4) are single-reviewer but corroborated against discipline (Practice 4 + Q-G no-co-authority).

Fix-forward: #2788.

— sent from deep-wolf-155