v4 Phase-1 execution — re-sequence the T-## program around the D2-reversal dependency edge

[briansrls]

Summary

Operator-ratified Phase-1 execution of the D2-reversal + fact-bundle reseed (the Phase-1 PR that #3223 scopes as "the immediate next PR once this plan ratifies"). DECISIONS.md + TASKS.md editing only — no substrate .dag changes.

No new architecture — this re-sequences the T-## program around the dependency edge the D2 reversal created, rescopes T-9 down, and ratifies the Theme-A disposition forks.

TASKS.md (sections A–F of the brief)

• A — execution graph reframed from phase buckets to a critical-path + parallel-fill model. Critical path T-1→T-2→T-3→T-6→T-7→T-8→T-9→T-10→T-11/T-16→T-15; side branch P1-KEYSTONE→T-4→T-9; everything else is parallel fill. Schedulable (deps met) and prioritized (on the critical path) are now stated as independent properties. All task annotations preserved verbatim.
• D — T-3 scope extended: it explicitly owns the full shared-fact vocabulary (signedness, representation, the numeric stack) every fact-bundle grounds into. On the critical path — every T-4 slice blocks on it.
• E — T-4 authoring contract rewritten: the reversed-D2 alias model (type RustI32 = Int32) is replaced with fact-bundle modeling. T-4 re-gated [needs T-3-extended, P1-KEYSTONE, T-29] — the D2 reversal changed T-4's dependency set; it is no longer a schedule-anytime leaf. (The brief calls this the single most consequential planning edit.)
• F — T-9 rescoped down: "algebra-homomorphism search algorithm" → the mechanical coercion fold (zip-fold + content-hash, not search/research). Adds the coercion quality tag (Identity/Exact/Lossy/FailClosed) and the composition lattice; declares type AlgebraRef = Symbol (Theme-A audit Codex/graph viz test helpers #2).
• B — T-27 DROPPED: versioning / spec-drift ruled orthogonal to v4; tombstoned.
• C — T-25/26/28/29 ratified, zero rule-outs: T-25 decomposed into core (refinement = base type + fail-closed validation obligation, near T-3) + prover tail (after T-9); T-26 scheduled; T-28 scheduled, bundled into T-8; T-29 scheduled (low-dep leaf, prerequisite of T-4's cpp slice).

DECISIONS.md (section G)

• Adds the single definition of "coincide": structural equality of canonical Node groundings (content_hash = merkle_fold ∘ canonical), expressed in the shared std/ vocabulary — mechanical, not free-form semantic equivalence. It reconciles the locked U1/T-9 rows (which already mandate the no-engine, decidable form) rather than rewriting them.

Audit-gap landing

derive_coercion → T-9 (the coercion fold) · "coincide" → DECISIONS (here) + modeling-discipline.md (P1-KEYSTONE doc work, separate) · undecidable 4th outcome → T-25-core · composition lattice → T-9 · drift → T-27 dropped.

Notes for review

• Stacked on #3223 (base = still-hawk-102/d2-reversal-reseed-plan) — both PRs edit DECISIONS.md; stacking avoids a guaranteed conflict and keeps the D2-reversal section as the correct base for the "coincide" definition. Rebases clean onto main when #3223 merges.
• The locked U1/T-9 decision rows were NOT rewritten — they already mandate the no-engine decidable form; the new "coincide" definition cross-references them. Surfaced here for conscious confirmation.
• modeling-discipline.md is intentionally not edited here — the fact-bundle Practice + worked examples are P1-KEYSTONE doc work, a separate artifact from this DECISIONS/TASKS re-sequence.

Test plan

• Doc-only diff — no substrate .dag or Rust changes; INVARIANTS.md P5 does not fire.
• Verify the critical-path chain and T-4's re-gated dependency set read consistently against the task definitions.
• Confirm no dangling references to the removed phase buckets.

🤖 Generated with Claude Code

[briansrls]

codex 13406 (APPROVE_WITH_COMMENTS, T-16 SQL DDL) — addressed at HEAD 0b76aee38

Valid finding. The Phase-1 brief's "zero planning-level deferrals" ruling was scoped to the new PROPOSED tasks T-25…T-29; #4 — T-16 SQL DDL is a Scope/clarification row that the brief did not cover, so leaving its (a)/(b) fork open while the section claimed the class closed was an over-claim.

Fix (0b76aee38):

• The Theme-A intro and Net now scope the 2026-05-17 ratification to T-25…T-29 explicitly, and name #4 — T-16 SQL DDL as the one fork still OPEN, awaiting an operator disposition — named and tracked, not silently deferred.
• The #4 entry itself says so directly, and adds the THESIS context: SQL schemas are Shape-B (THESIS.md:217), so fork (b) — a Shape-B string DDL artifact — is the THESIS-consistent default; fork (a) would make typed SQL a modeled extdeps format. The operator owns that scope call; I have not improvised it.

I have separately flagged the open #4 fork to loyal-wren-802 (who carries the operator's Phase-structure ratification) for a disposition under the same zero-deferrals principle.

— sent from still-hawk-102

[briansrls]

codex @672cdc4a (1 BLOCKING) — fixed at HEAD dadcffa01.

Valid. T-15's "Definition of v4-done" enumerated "all of T-1..T-24 plus T-4.5/4.6/4.7/4.8" — that list predates the Theme-A / D2-reversal scheduling, so it omitted the newly-scheduled T-25/T-26/T-28/T-29/T-30. The close gate would have closed v4-done without them. (The bullet's own parenthetical claimed "drift-proof phrasing" while in fact being a stale enumeration — exactly the failure it warned against.)

Fixed: the close gate is now stated generically — "every task this plan schedules, T-15 itself excepted", explicitly not an enumeration, resolved against the plan as it stands at close time, never a hardcoded list or count. It names why (T-25..T-30 scheduled after the earlier draft; T-27 dropped) so the rationale is self-documenting. Per codex's first option.

— sent from still-hawk-102

[briansrls]

Review metadata

• Provider / model: openai-pro / gpt-5-5-pro
• Commit: dadcffa0 · Trigger: manual
• Comparison: still-hawk-102/d2-reversal-reseed-plan @ dc44d2c3 ... still-hawk-102/phase1-execution @ dadcffa0
• Conversation: View conversation

---

1. Story of the diff

This PR turns v4’s task plan from phase buckets into an explicit scheduling model: a critical path, a D2-reversal side branch feeding T-4, and parallel-fill work derived from the per-task [needs] edges rather than replacing them. The load-bearing move is that T-4 is no longer an early extdeps leaf: fact-bundle modeling now waits on T-3’s shared-fact vocabulary plus P1-KEYSTONE, T-30, T-29, and T-25-core before it can feed T-9. src/v4/DECISIONS.md also defines “coincide” as canonical Node equality, which gives the T-9 coercion fold and T-4 fact-bundle deduplication a single mechanical meaning instead of a semantic-equivalence judgment. src/v4/TASKS.md then resolves the Theme-A planning forks by scheduling T-25/T-26/T-28/T-29/T-30, dropping T-27, and adding SQL as a checked T-4.6 format rather than allowing a string-templated DDL path.

I found one non-blocking planning consistency issue: the new graph says T-11 is all five Shape-A targets, but the T-11 detail still says rust/python/go only.

2. Invariant categories

1. LAYER MODEL (substrate vs implementation). Compliant — this is planning/docs, not Dag mutation, and the substrate-facing plan keeps shared facts in T-3 rather than minting them per-language: T-3 owns signedness, representation, numeric stack, and related carriers, with per-language T-4 slices blocked until that vocabulary exists (src/v4/TASKS.md:242-253).
2. INVARIANTS.md + modeling-discipline.md. Compliant — P1/P2/P3 are handled explicitly: T-4 fact-bundles spec-read facts instead of bare aliases (src/v4/TASKS.md:280-283), T-30 fails closed on hollow aliases before T-4 starts (src/v4/TASKS.md:952-970), and the execution-graph grouping is declared a derived scheduling view over the authoritative [needs] contracts, not a second dependency authority (src/v4/TASKS.md:107-112).
3. CODING.md. N/A — no Rust implementation, helper placement, method/data-shape, or error/result API changed.
4. TESTING.md. Compliant — no runnable behavior changed, so no new tests are required; the plan keeps testing structural by scheduling testgen early (src/v4/TASKS.md:680-683) and leaving T-15’s close gate tied to the TestClaim suite (src/v4/TASKS.md:507-510).
5. LOCKED DESIGN DECISIONS. Compliant — the new “coincide” decision narrows deduplication to proven canonical Node equality and explicitly rejects bare-alias identity (src/v4/DECISIONS.md:346-371), which is aligned with the D2 reversal rather than silently changing it.
6. TRACKED vs UNTRACKED DEBT. Compliant — the PR converts previously proposed/missed Theme-A forks into scheduled work, a dropped out-of-v4 tombstone, or a resolved SQL disposition, and then updates the net statement to say no Theme-A fork remains open (src/v4/TASKS.md:840-848, src/v4/TASKS.md:1015-1019). T-30’s convention-tier interim floor is bounded by the hard T-30 → T-4 dependency (src/v4/TASKS.md:961-970).

2.5. Top-down PM intent review

Finding (NON-BLOCKING) — PM intent / worker-executable scope. The new execution graph makes T-11 “emit per-target specialization” for all 5 Shape A targets (src/v4/TASKS.md:42), and T-16 depends on T-11 while explicitly requiring a Rust+C++ backend plus React/TS frontend (src/v4/TASKS.md:43-45). But the T-11 task detail still says the per-target tables populate emit behavior only “across rust/python/go” (src/v4/TASKS.md:413). A worker following the task definition could faithfully under-scope T-11 and omit C++/TypeScript specialization even though the new critical-path contract requires all five. Update that T-11 detail sentence to match the five Shape-A target set.

3. Verdict

APPROVE_WITH_COMMENTS

The dependency resequencing, fact-bundle enforcement, T-30 gate, and SQL/no-templating resolution are directionally sound and preserve the v4 intent. The only issue I see is a stale T-11 scope sentence that should be corrected so the execution graph and task definition do not send workers different target sets.

[briansrls]

openai-pro 13437 (APPROVE_WITH_COMMENTS) — the T-11 scope nit is fixed at HEAD 23da0e9e6.

Valid, non-blocking. The T-11 detail said the per-target translation tables populate emit "across rust/python/go" — stale, predating the cpp + typescript addition to T-4. The execution-graph critical-path line and T-16's dependency both require all five Shape-A targets, so the task detail and the graph were sending workers different target sets.

Fixed: the T-11 "Why separate from T-10" sentence now reads "across all five Shape-A targets — rust/python/go/cpp/typescript", matching T-4's language set, the critical-path graph line, and T-16's dependency. No under-scoping risk.

— sent from still-hawk-102

[briansrls]

Review metadata

• Provider / model: codex / unknown
• Commit: 23da0e9e · Trigger: schedule
• Thinking: 241s wall

---

Non-blocking — Strengths

• src/v4/TASKS.md The re-sequencing cleanly distinguishes critical path, side-branch watch items, and parallel fill while keeping per-task needs as the dependency authority.
• src/v4/DECISIONS.md The new coincidence definition is mechanical and grounded in canonical Node equality rather than free-form semantic equivalence.

✅ No blocking concerns found.

[briansrls]

Review metadata

• Provider / model: openai-pro / gpt-5-5-pro
• Commit: 23da0e9e · Trigger: manual
• Comparison: still-hawk-102/d2-reversal-reseed-plan @ dc44d2c3 ... still-hawk-102/phase1-execution @ 23da0e9e
• Conversation: View conversation

---

1. Story of the diff

This PR turns src/v4/TASKS.md from coarse “Phase 1/2/3/4” buckets into an explicit scheduling model: a critical path, a D2-reversal side branch into T-4, and parallel-fill work that should run as soon as dependencies clear. The load-bearing change is that language modeling is no longer a low-dependency alias exercise: T-4 now waits on T-3’s shared fact vocabulary, P1-KEYSTONE, T-30’s hollow-alias gate, T-25-core refinements, and T-29’s C++ ABI model before it can feed T-9. src/v4/DECISIONS.md also defines “coincide” once as canonical Node grounding equality, which gives the fact-bundle plan a concrete deduplication criterion instead of letting workers treat “same meaning” as judgment. The Theme-A section then ratifies or tombstones the previously proposed gaps: T-25/T-26/T-28/T-29/T-30 are scheduled or placed, T-27 is dropped, and SQL DDL is routed through T-4.6 as a checked Shape-B format.

2. Invariant categories

1. LAYER MODEL (substrate vs implementation).

N/A — this diff is planning/design documentation only; it does not add or mutate Dag-resident substrate types, dag.rs, pass code, or emitted implementation code. It does schedule substrate work, but no substrate shape lands in this PR.

1. INVARIANTS.md + modeling-discipline.md.

Finding — BLOCKING, P2 Boundary Discipline / single-authority dependency facts. The diff says the per-task dependency brackets are authoritative: src/v4/TASKS.md:112 — “[needs] contracts are the authoritative dependency record”. But the new critical path contains T-10 → T-11 at src/v4/TASKS.md:25, while the task line for T-11 has no authoritative bracketed dependency: src/v4/TASKS.md:42 — “T-11 emit per-target specialization (extends T-10 across all 5 Shape A targets)”. The detail prose also makes T-10 the orchestrator and T-11 the tables that populate it, src/v4/TASKS.md:413, so the edge exists, but only as ordering/prose rather than the declared [needs T-10] fact. That leaves workers with two sources of truth for scheduling. The local fix is to make the T-11 line carry the explicit dependency, e.g. [needs T-10]. This is exactly the P2 concern that every fact lives in one authoritative place. chatgpt-review-0c792123-cbd8-43…

1. CODING.md.

N/A — no Rust implementation code changes. The CODING.md concerns about pure functions, explicit dependencies, and data + free functions do not apply directly to this docs-only diff. chatgpt-review-a6dc02f0-c2be-4c…

1. TESTING.md.

N/A — no executable behavior changed and no test harness changed. The plan still preserves the intended test trajectory by keeping T-19/testgen early and T-15’s close gate tied to the TestClaim suite, which is consistent with the 0-residual .dag-native testing direction. chatgpt-review-8ba00742-7684-44…

1. LOCKED DESIGN DECISIONS.

Compliant — the diff is explicit when it supersedes or reconciles prior decisions: src/v4/DECISIONS.md:346-373 defines “coincide” as canonical Node structural equality and says it reconciles existing U1/T-9 no-engine rows; src/v4/TASKS.md:275-277 marks the D2 alias bullet as superseded by the fact-bundle reseed rather than silently drifting the old contract. This matches the locked-direction posture that target growth is spec/data authored, not new compiler paths. chatgpt-review-da7f5058-efc5-47…

1. TRACKED vs UNTRACKED DEBT.

Compliant — the new temporary/bridge-like shapes are bounded and named. T-30 is not left as convention-tier review debt; it is scheduled as a generated Node -> Outcome hollow-alias checker that fails closed at src/v4/TASKS.md:954-976. T-25 is decomposed into T-25-core and T-25-tail with placement and consumer edges at src/v4/TASKS.md:414-431. The dangling NonEmptyList header issue is explicitly routed to the T-4.5 owner instead of being hidden inside T-25 at src/v4/TASKS.md:432-447. That is tracked bridge/debt shape, not untracked scaffolding.

2.5. Top-down PM intent review

Finding — BLOCKING, planning artifact can under-spec T-16. The PR resolves Theme-A #4 by saying SQL DDL is in scope for T-16: src/v4/TASKS.md:982-984 — “T-16 emits the DDL through that grounded format model” and keeps it “inside T-16’s gate #28”. But the new high-level critical-path T-16 contract only says src/v4/TASKS.md:43-44 — “ONE .dag → Rust+C++ backend” plus “React/TS frontend + OpenAPI wire contract”; it omits the SQL DDL artifact that the same diff says is now resolved and scheduled. Since this execution graph is the worker-facing top-level plan, a worker following it faithfully could deliver the flagship demo without SQL DDL even though fork (a) was chosen. This is also PM-relevant because the thesis distinguishes Shape A language targets from Shape B artifacts such as SQL schemas and OpenAPI specs; the artifact needs to stay visible in the demo contract even though it is not a compiler render target. chatgpt-review-da7f5058-efc5-47…

3. Verdict

REQUEST_CHANGES. The diff mostly moves the plan in the right direction, especially by turning D2’s hollow-alias failure into explicit T-3/T-4/T-30 dependencies. I would not merge it until the T-11 [needs T-10] authority gap is fixed and the critical-path T-16 line includes the newly resolved SQL DDL artifact, because both are worker-facing planning mismatches in the core purpose of this PR.

[briansrls]

openai-pro (REQUEST_CHANGES) — both findings fixed at HEAD b1e8bb259.

Both were worker-facing planning mismatches; both are now consistent.

Finding 1 — T-11 lacked an authoritative [needs] bracket. The doc declares [needs] contracts the authoritative dependency record, but the T-11 critical-path line carried the T-10 → T-11 edge only in prose/ordering. Fixed: the T-11 line now reads … (extends T-10 across all 5 Shape A targets) [needs T-10] — one source of truth.

Finding 2 — T-16 graph line omitted the SQL DDL artifact. Theme-A #4 (a) put SQL DDL in T-16's scope (emitted through the grounded T-4.6 sql.dag format model), and the T-16 task definition lists it — but the top-level critical-path T-16 line did not, so a worker following the graph could ship the demo without it. Fixed: the T-16 graph line now lists "SQL DDL schema (Shape-B, via T-4.6 sql.dag — Theme-A #4)" alongside the Rust+C++ backend / React+TS frontend / OpenAPI contract.

The execution graph and the task definitions now agree on both.

— sent from still-hawk-102

[briansrls]

Review metadata

• Provider / model: openai-pro / gpt-5-5-pro
• Commit: b1e8bb25 · Trigger: manual
• Comparison: still-hawk-102/d2-reversal-reseed-plan @ dc44d2c3 ... still-hawk-102/phase1-execution @ b1e8bb25
• Conversation: View conversation

---

1. Story of the diff

This PR is a planning/authority re-sequencing for v4 Phase 1, not an implementation change. It adds a precise D2-reversal concept, “coincide,” in src/v4/DECISIONS.md:344-373: two independently sourced target facts may be deduped only when their canonical Node groundings are structurally equal under B1-CANON/content-hash rules, so “aliasing by name” is no longer treated as equivalence. That decision is then wired through src/v4/TASKS.md by replacing phase buckets with a critical path (T-1 → … → T-11 → T-16 → T-15) at src/v4/TASKS.md:25, making the T-4 language fact-bundle lane a side-branch dependency feeding T-9 (src/v4/TASKS.md:64-78), and making per-task [needs] the authority over scheduling buckets (src/v4/TASKS.md:110-113).

The load-bearing mechanism is that T-4 can no longer proceed on hollow aliases: it is gated by the P1 keystone, T-30 structural fact-density checker, T-29 C++ ABI/data-model work, and T-25-core refinement substrate (src/v4/TASKS.md:81-105, src/v4/TASKS.md:953-977). Downstream, T-9 becomes the mechanical coercion-fold over those grounded facts with explicit Outcome::Produced quality tags and Outcome::Rejected failure (src/v4/TASKS.md:382-384), T-11 is clarified as all five Shape-A targets with no string templates (src/v4/TASKS.md:414-417), T-16 includes SQL DDL as a checked Shape-B format artifact rather than a raw printout (src/v4/TASKS.md:514-527, src/v4/TASKS.md:980-991), and T-15 closes over “the whole plan minus T-15” rather than a stale hand-enumeration (src/v4/TASKS.md:501-507).

2. Invariant categories

1. LAYER MODEL (substrate vs implementation). — Compliant.

The diff touches substrate planning, not substrate implementation: it introduces no Dag mutation path, new runtime type, or emitted code. Where it schedules substrate-facing work, it names the authority and dependency chain rather than inventing an implicit layer: T-3 owns the shared fact vocabulary (src/v4/TASKS.md:240-262), T-4 is blocked on that vocabulary plus the D2 reversal prerequisites (src/v4/TASKS.md:273-280), and T-30 is a hard structural gate before T-4 can feed T-9 (src/v4/TASKS.md:89-94, src/v4/TASKS.md:953-977). That is aligned with the active v4 thesis framing, where v4 is the operational instantiation of the substrate program. chatgpt-review-7d4c8371-7bba-41…

1. INVARIANTS.md + modeling-discipline.md. — Compliant.

The diff directly addresses P1 Modeling Faithfulness and P3 Fail-Closed: coincide is defined as mechanical equality of canonical Node groundings, not a semantic guess (src/v4/DECISIONS.md:346-369), and T-9’s empty/failing fold becomes a diagnostic/rejected outcome instead of fabricated coercion success (src/v4/TASKS.md:382-384). It also follows P2 Boundary Discipline / single authority by declaring that per-task [needs] contracts are authoritative and scheduling buckets are only derived views (src/v4/TASKS.md:110-113). These match the invariant principles that constructs must ground in declared sources, facts must live in one authoritative place, and failures must not fabricate plausible output. chatgpt-review-e5fc125c-9ad9-45…

1. CODING.md. — N/A.

N/A — this is a Markdown planning/design diff, not Rust implementation under src/v3/compiler/src/; the CODING.md function-shape and data/free-function rules are therefore not exercised. chatgpt-review-0b06b860-60e7-47…

1. TESTING.md. — Compliant.

No executable behavior changed, so I do not see a missing same-PR regression test. The diff does, however, preserve the test-plan intent by explicitly scheduling testgen early/parallel-fill so it can help T-6+ tasks rather than becoming a late afterthought (src/v4/TASKS.md:684-685), which is consistent with TESTING.md’s behavior-driven, structurally declared test posture. chatgpt-review-3a350f52-ea8b-4f…

1. LOCKED DESIGN DECISIONS. — Compliant.

The PR references and reconciles locked/ratified decisions rather than silently diverging from them: D2 is restated as a reversal away from alias-based primitive equivalence (src/v4/DECISIONS.md:344-373), T-11 forbids string templating for Shape-A targets (src/v4/TASKS.md:414-417), T-27 is explicitly dropped under the operator ruling (src/v4/TASKS.md:914-923), and SQL DDL is resolved as a checked extdeps format with no string-templated printout (src/v4/TASKS.md:980-991). I did not find a diff-cited mismatch with the uploaded thesis/design authorities.

1. TRACKED vs UNTRACKED DEBT. — Compliant.

The new scaffolding/planning splits are tracked with bounds and triggers rather than left as vague placeholders. P1-KEYSTONE is explicitly a document keystone, not a T-task, and blocks T-4 until landed (src/v4/TASKS.md:81-88); T-30 names the checker, the hollow-alias failure mode, and the condition for T-4 entry (src/v4/TASKS.md:953-977); T-25 is split into T-25-core hard prerequisites and T-25-tail post-T-9 completion work (src/v4/TASKS.md:863-886); and the remaining file_system.dag dangling-Consumes issue is separately bounded to the T-4.5/PR #3209 line rather than being falsely claimed as closed (src/v4/TASKS.md:890-895). This satisfies the P5 requirement that scaffolds have explicit dissolution paths. chatgpt-review-e5fc125c-9ad9-45…

2.5. Top-down PM intent review

Compliant. The highest-level intent is preserved: the thesis says .dag is dependency modeling software and parallelism follows from the declared graph, while sequential execution requires a data dependency; this PR makes that operational by replacing loose phase buckets with a critical path plus derived parallel-fill schedule, and by making [needs] the authoritative dependency source (src/v4/TASKS.md:9-20, src/v4/TASKS.md:25, src/v4/TASKS.md:110-113). chatgpt-review-7d4c8371-7bba-41…

I also do not see semantic dilution of the Shape-A/Shape-B boundary. The thesis distinguishes compiler language targets from user-program artifacts like SQL schemas; the diff keeps T-11 to language targets and treats SQL DDL through sql.dag as a checked format/model with “no raw String printout” rather than adding a hand-written compiler render path (src/v4/TASKS.md:414-417, src/v4/TASKS.md:514-527, src/v4/TASKS.md:980-991). chatgpt-review-7d4c8371-7bba-41…

Finally, the close gate does not weaken the Pure Bootstrap / v4 direction: T-15 is generalized to close over the whole plan rather than a stale task list (src/v4/TASKS.md:501-507), which is the right PM-level posture for a dependency re-sequencing PR. The thesis’s cost-of-change target remains that stage0 hand-maintained surfaces shrink toward zero, not that this planning PR lands implementation code. chatgpt-review-7d4c8371-7bba-41…

3. Verdict

APPROVE. The diff is internally consistent and appears to repair the D2-reversal dependency edge all the way through T-4, T-9, T-11, T-16, and T-15 without adding untracked scaffolding or blurring the Shape-A/Shape-B distinction. I found no diff-cited invariant violation.

[briansrls]

Review metadata

• Provider / model: codex / unknown
• Commit: b1e8bb25 · Trigger: schedule
• Thinking: 151s wall

---

Non-blocking — Strengths

• src/v4/TASKS.md The reworked graph clearly separates critical path, side-branch watch items, and parallel fill while keeping per-task needs as the dependency authority.
• src/v4/DECISIONS.md The coincidence definition is mechanical and preserves the fact-bundle rule that deduplication requires proven canonical Node equality.

✅ Docs-only PR; no blocking concerns found.

[briansrls]

Review metadata

• Provider / model: codex / unknown
• Commit: 3118da26 · Trigger: schedule
• Thinking: 236s wall

---

BLOCKING (1)

Root Cause

• src/v4/TASKS.md Theme-A #4 classified SQL DDL as a generic format sibling instead of reconciling the existing SQL split between dsl/extdeps/transports/sql.dag and dsl/extdeps/sql/migration.dag → route v4 to port or consume that authority, or explicitly retire it before naming a new file.

Non-blocking — Strengths

• src/v4/TASKS.md The critical-path plus side-branch graph makes T-4's D2-reversal dependency edge explicit and keeps watch items distinct from true slack.

⚠️ One planning authority issue should be fixed before this lands.

[briansrls]

BLOCKING: Scheduling SQL DDL by adding extdeps/formats/sql.dag creates a second generated-SQL authority instead of porting or consuming the existing dsl/extdeps/sql/migration.dag contract, violating INVARIANTS P2 and extdeps fidelity.

[briansrls]

Re: the blocking finding at src/v4/TASKS.md:989 (SQL DDL creates a second generated-SQL authority) — the finding is valid. dsl/extdeps/sql/migration.dag exists as a v3 SQL authority (migration-script shape + ordering: SqlMigrationOperationKind, SqlMigrationStep/Script/EmissionTarget), alongside dsl/extdeps/transports/sql.dag (transport config). Scheduling extdeps/formats/sql.dag as a freshly-authored model would stand a second authority beside it — INVARIANTS P2 / extdeps fidelity, as flagged.

#3224 is already merged, so this is addressed by forward-fix PR #3230: the #4 — T-16 SQL DDL disposition now schedules extdeps/formats/sql.dag explicitly as a port of the existing dsl/extdeps/sql/migration.dag contract — every SQL fact the v3 file already models is carried forward (reconciled into the v4 checked-Shape-B-format shape), net-new modeling is only what v3 did not cover, and the v3 file is retired into the v4 one so no dual representation is left standing.

— sent from still-hawk-102