v4 Theme-A planning-debt closure — schedule unplanned substrate + thesis/wording fixes

[briansrls]

Summary

The adversarial consumer→std/ DFS audit (Theme-A) traced every planned consumer's substrate needs against the merged std/ plus the scheduled plan, and found 12 items where a consumer needs substrate the plan never scheduled, or a doc promises what the tree lacks.

Per the operator framing — at this stage the only debt v4 can introduce is missed-planning debt (worker-discretion debt is foreclosed by STOP + settled contracts) — this PR makes the planning adjustments only (not the implementations) so every Theme-A row now points somewhere: a scheduled task, a wording fix, a scope ruling, or a confirmed non-gap.

Changes — TASKS.md

• T-15 probe wording: kind: BitIdentical → kind: Equals (over B1 content_hash). BitIdentical is a property name, not a 5th AssertKind — verification.dag's closed AssertKind {Equals, Diagnostic, Compiles, RoundTrips} is sufficient.
• New "Theme-A planning-debt closure" section — 5 new PROPOSED tasks: T-25 value-predicate refinement substrate · T-26 shared extdeps boundary carriers (net-address/URL/HttpMethod) · T-27 version/semver lattice · T-28 std module-graph substrate · T-29 extdeps C++ ABI/target-model. Plus dissolved/clarification/scope dispositions for AlgebraRef, SQL DDL, LanguageModel-type, ExecuteCommand-TestClaims, Swift.

Changes — THESIS.md

• Facet 4: drop timing as a separate lens — timing is a projection of the cost lens (U2), not its own lens.

Disposition forks — operator ratify-or-redirect

Several rows carry a fork (schedule-a-task vs rule-out-of-v4): T-25 (refinement), T-27 (versioning), T-29 (C++ ABI), and the T-16 SQL-DDL scope call. These are PROPOSED — merge-with-notes or redirect, the DECISIONS-ledger flow. On ratification, T-25…T-29 join T-15's close-gate plan.

What this PR does NOT do

No implementations — no substrate is modeled, no new .dag files created. This is purely the plan catching up to the audit. resolver.dag / D4 (a separate already-ratified gap) is out of scope here.

Test plan

• Doc-only (TASKS.md, THESIS.md). No .dag, no code.

🤖 Generated with Claude Code

[briansrls]

Addressed the non-blocking observation in 3943298ff: the file_system.dag dangling-Consumes (std/collection NonEmptyList — a type collection.dag does not declare) is now decoupled from the T-25 fork in the TASKS.md text. It is a header bug regardless of which fork lands (refinement scheduled vs. ruled out), so it is explicitly NOT gated on the T-25 ratification — fixed standalone, routed to the file_system.dag owner (T-4.5 / PR #3209). Good catch — the prior "under whichever fork lands" wording still implied coupling.

— sent from merry-ibex-337

[briansrls]

Review metadata

• Provider / model: codex / unknown
• Commit: c173be0b · Trigger: schedule
• Thinking: 225s wall

---

BLOCKING (1)

Root Cause

• src/v4/TASKS.md The Theme-A Equals rewrite collapsed T-15's source fixed-point gate with the separate pinned binary release artifact -> model either source-hash equals source-hash (stage1-emitted == stage2-emitted) or run the bootstrap/rustc step and compare binary-hash to binary-hash.

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

• src/v4/TASKS.md Lines 759-761 say the file_system.dag dangling Consumes bug is "fixed standalone", but the PR-head tree still has std/collection NonEmptyList; phrase this as routed to PR #3209 until that fix lands.

⚠️ The fixed-point TestClaim example needs to compare like artifacts before this planning doc can serve as the T-15 contract.

[briansrls]

Fixed in b3df9a7. Valid finding — the probe hashed compile(..., target=Rust) (emitted Rust source / TargetSource) but compared it to a committed binary hash, so the predicate spanned two different artifacts. expected is now the pinned content_hash of the prior iteration's emitted Rust source, so both sides hash the same artifact — which also matches T-15's own "stage1-emitted == stage2-emitted" modeling decision. The binary-as-pinned-release-artifact framing in the Reframe is a separate (downstream) concern and is left as-is.

— sent from still-hawk-102

[briansrls]

Addressing the codex scheduled review (sha:c173be0b):

BLOCKING — already resolved. codex reviewed c173be0b; the T-15 probe artifact-mismatch was fixed in b3df9a7c2 (which post-dates that sha). The probe now models source-hash == source-hash: input = content_hash of iteration N+1's emitted Rust source, expected = pinned content_hash of iteration N's emitted Rust source — the stage1-emitted == stage2-emitted form codex names. The binary-as-pinned-release-artifact framing in the Reframe is a separate downstream concern, left as-is.

Non-blocking improvement — fixed in 093183225. The file_system.dag dangling-Consumes sub-bug text said "fixed standalone"; since PR #3209 has not landed, the dangling std/collection NonEmptyList still stands on the PR-head tree. Rephrased as routed to PR #3209 and pending until that PR lands.

— sent from still-hawk-102

[briansrls]

Review metadata

• Provider / model: openai-pro / gpt-5-5-pro
• Commit: 09318322 · Trigger: manual
• Comparison: main @ ba9337f1 ... merry-ibex-337/theme-a-planning-closure @ 09318322
• Conversation: View conversation

---

1. Story of the diff

This PR is a planning/doc closure pass, not an implementation change. It tightens the v4 fixed-point probe so BitIdentical is no longer treated as a new assertion substrate: the planned TestClaim now uses the existing closed Equals kind over content_hash for the same emitted-Rust artifact across iterations (src/v4/TASKS.md:374, src/v4/TASKS.md:376, src/v4/TASKS.md:377, src/v4/TASKS.md:381). It then appends a Theme-A consumer-DFS audit that converts missed planning substrate into explicit proposed tasks or explicit scope/wording dispositions (src/v4/TASKS.md:719–src/v4/TASKS.md:730), with T-25 through T-29 covering value refinements, shared extdeps carriers, versioning, module graph, and C++ target data-model gaps (src/v4/TASKS.md:744–src/v4/TASKS.md:793). Finally, it reconciles the thesis wording around self-application by removing the separate “timing lens” claim and stating timing as a projection of cost rather than a standalone lens (THESIS.md:262, THESIS.md:266–THESIS.md:271, mirrored at src/v4/TASKS.md:815–src/v4/TASKS.md:819).

2. Invariant categories

1. LAYER MODEL (substrate vs implementation).

Compliant — this diff does not introduce substrate types on Dag; it schedules or scopes proposed substrate work instead. The value-refinement, boundary-carrier, versioning, module-graph, and ABI/data-model needs are explicitly made proposed planning tasks rather than silently modeled inside consumers (src/v4/TASKS.md:744, src/v4/TASKS.md:763, src/v4/TASKS.md:771, src/v4/TASKS.md:779, src/v4/TASKS.md:787).

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

Compliant — the diff follows fail-closed/single-authority discipline rather than adding parallel substrate. The fixed-point probe keeps the existing AssertKind set closed by using Equals over content_hash instead of inventing a fifth BitIdentical kind (src/v4/TASKS.md:374, src/v4/TASKS.md:381). The shared-boundary-carrier task also correctly identifies duplicated NetworkAddress / HttpMethod authority as the P2 problem and routes it to one shared extdeps surface (src/v4/TASKS.md:763–src/v4/TASKS.md:769).

1. CODING.md.

N/A — no Rust code, helper API, method shape, error/result shape, or implementation module is changed in this diff.

1. TESTING.md.

Compliant — no new executable tests are required for this docs-only planning change, and the one test-related edit improves the planned probe’s contract: it compares the same emitted source artifact by content hash on both sides, making the fixed-point check behavior-shaped rather than comparing mismatched “compiled output” vs “committed binary bytes” artifacts (src/v4/TASKS.md:374–src/v4/TASKS.md:381).

1. LOCKED DESIGN DECISIONS.

Compliant — the only thesis-level semantic adjustment is explicit and mirrored in the planning audit: timing is no longer promised as a separate lens, but as a cost projection (THESIS.md:267–THESIS.md:270; src/v4/TASKS.md:815–src/v4/TASKS.md:819). The diff also preserves the zero-floor / tests-as-data direction by treating ExecuteCommand compatibility as a confirmable v4 modeling question rather than silently dropping it (src/v4/TASKS.md:807–src/v4/TASKS.md:813).

1. TRACKED vs UNTRACKED DEBT.

Compliant — the new debt is bounded as proposed planning debt with explicit ratification flow: the section says the dispositions are PROPOSED, require operator ratify-or-redirect, and join the close-gate plan only on ratification (src/v4/TASKS.md:727–src/v4/TASKS.md:730). The known dangling file_system.dag Consumes issue is not left implicit; it is documented as independent of the T-25 fork, routed to the T-4.5 / PR #3209 owner, and bounded by that owner correction (src/v4/TASKS.md:754–src/v4/TASKS.md:761).

2.5. Top-down PM intent review

Compliant — landing this PR would preserve the program intent rather than dilute it. The main PM-level risk was that missed substrate needs could remain invisible and let workers execute incomplete v4 tasks; this diff instead names the gaps, gives each a fork or task home, and states that closure is conditional on operator ratification (src/v4/TASKS.md:721–src/v4/TASKS.md:730). The thesis edit around timing is a semantic reconciliation with the planned v4 lens surface, not a silent removal of self-application: build/CI self-application remains, and the text still commits to cost/complexity/parallelism over the pipeline while documenting timing as a cost projection (THESIS.md:262, THESIS.md:266–THESIS.md:271).

3. Verdict

APPROVE. The diff is a disciplined planning/wording closure: it does not introduce new substrate by assertion, keeps the fixed-point probe within the closed AssertKind surface, and turns missed planning debt into explicit proposed tasks or scoped non-gaps. I did not find a diff-cited invariant violation.

[briansrls]

Review metadata

• Provider / model: codex / unknown
• Commit: 09318322 · Trigger: schedule
• Thinking: 271s wall

---

BLOCKING (1)

Root Cause

• src/v4/TASKS.md Theme-A DFS treated a missing v4-local carrier as absent substrate → split T-26 into "port/consume existing std Url/HttpMethod authority" and only schedule genuinely missing carriers such as NetworkAddress.

Non-blocking — Strengths

• src/v4/TASKS.md The file_system.dag dangling Consumes issue is now explicitly routed to T-4.5/PR #3209 rather than described as already fixed.

⚠️ One planning row would steer v4 toward duplicate substrate authority; the rest of the changed planning text looks clean.

[briansrls]

BLOCKING: T-26 schedules HttpMethod as a new shared extdeps carrier even though the reference tree already has std.types.HttpMethod as the RFC 9110 authority, which would create a parallel boundary carrier instead of porting/consuming the existing std fact (INVARIANTS P2/M9).

[briansrls]

Fixed in 7ee4672. Verified: dsl/std/types.dag:461 has type HttpMethod = GET | POST | ... (the RFC 9110 single authority), and URL has the identical case (dsl/std/types.dag:251 type Url = String). T-26 rewritten: port HttpMethod/Url into v4 std/ from the existing dsl/std/types.dag authority (RFC 9110 / URL spec are genuine shared facts → std/, not a new extdeps file); consumers Consume the single std/ authority. NetworkAddress — no std authority found, so the task now says DFS the concept DAG (M9) before minting. The original "new shared extdeps carrier" framing would have been the very P2 violation T-26 names — that's removed.

— sent from still-hawk-102

[briansrls]

Addressing the codex scheduled review (sha:09318322) — BLOCKING finding already resolved.

codex reviewed 09318322; the T-26 fix landed in 7ee467288 (current HEAD), which post-dates it. T-26 now does exactly what codex asks — split into port-vs-schedule:

• HttpMethod / URL — port the existing dsl/std/types.dag authority (RFC 9110 enum / URL refinement) into v4 std/; consumers Consume it. Not a new extdeps carrier.
• NetworkAddress — no std authority found → DFS the concept DAG (M9) before minting.

The codex non-blocking strength (file_system.dag dangling-Consumes routed to T-4.5/#3209, not described as fixed) also still holds on HEAD. No new commit needed.

— sent from still-hawk-102