R3 gate #87: lens_cementing_test_discipline_complete

[briansrls]

Auto-opened by session-dashboard for session sleek-swift-736.
Pushing to session/sleek-swift-736 advances this PR.

Closes #2609

Worker attestation

Before flipping this PR to ready for review, confirm each item:

• Title describes the change (not the session id or branch).
• PR body summarises what and why (replace the TODO below).
• Tests run: name the command (e.g. npm test, cargo test) and the result.
• If this closes a work item, the body contains a Closes #N directive.
• No commits on this branch are surprises (no fork/cherry-pick I did not make).
• No secrets / credentials / large binaries staged.

Summary

TODO: replace this paragraph with one or two sentences naming the change and its motivation. Reviewers read this first.

Test plan

• TODO: list the commands that ran (or "no tests changed; relied on CI") and the outcome.

[briansrls]

codex REQUEST_CHANGES (cost_target_realization receipt) — valid.

`r3_gate_87_cost_target_realization_rust_receipt_callable` only invoked `type_realization_meta` and discarded the result, so it violated `TESTING.md` §2 (no observable contract).

Fix (pushed): renamed to `r3_gate_87_cost_target_realization_rust_receipt_resolves_type_realization_row`, compile a minimal program with `compile_to_dag`, then assert `type_realization_meta` is `Some` and the resolved declaration name is `TypeRealization` (same `declaration_by_name` contract as `lens_cost_target_realization_test::type_realization_meta_resolves_against_bootstrap`, but on the user-pipeline dag).

Commit: `a8d0910bd`.

[briansrls]

Dashboard relay (briansrls PR comment @ 17:47:25Z) — already addressed in-tree; no new commit.

That text is the resolution note for codex’s cost_target_realization receipt: HEAD still has `r3_gate_87_cost_target_realization_rust_receipt_resolves_type_realization_row` (`compile_to_dag` + `Some(TypeRealization)` assertions) from `a8d0910bd`. Nothing further to change for this item.

— sent from sleek-swift-736

[briansrls]

claude-opus-4-7 REQUEST_CHANGES — addressed.

1. §1.8 gate RT1-RT4: model correctness — credential wiring, fail-closed auth, fil… #87 / semantic dilution — Status is CONSUMER_LANDED only; Notes restore canonical §Acceptance (every .dag lens cemented against frozen v2-oracle on same source, r3-structure.md) and explicitly withhold PASSING while eight regen harnesses remain Compiles placeholders (§1.7). Lane status row + Cluster M risk table updated (on branch via commit 953227736).

2. Lane-E cost harness — miss_int_lookup() bodies are stubs for DeclarationRef lowering, same as r1_gates.dag; TestRunner implements real parity via host forward-fold vs emit lens_cost::cost_of (test_runner.rs ~3487–3513). t_r3_gate_87_cementing_regen_cost.dag header comments now document that (953227736).

3. INVARIANTS P5(b) — In-source receipt on r3_gate_87_lens_cementing_regen_receipts_test.rs + cross-ref on t_pb_b_1_dag_runner_test.rs (1392975d3).

Force-with-lease push: 1392975d3.

[briansrls]

Follow-up (rebase onto main @ 459b89e + Cluster M row): resolved docs/r3-program-plan.md conflicts vs landed #86 (program_generator_carrier_landed CONSUMER_LANDED + PASSING, PR #2645). Q-PB0-ClusterM row no longer claims #86 is DECLARED. Branch force-pushed: c00d71cb9.

If GitHub still shows mergeable=CONFLICTING, refresh — local git merge origin/main is clean.

[briansrls]

Dashboard relay (briansrls PR comments @ 18:04 / 18:08) — already reflected at origin/session/sleek-swift-736 c00d71c (rebased onto current main; branch contains origin/main). No further code or doc edits required for these items.

• §1.8 RT1-RT4: model correctness — credential wiring, fail-closed auth, fil… #87: CONSUMER_LANDED + Notes withholding PASSING — present in docs/r3-program-plan.md:313.
• Lane-E cost harness stub truth — t_r3_gate_87_cementing_regen_cost.dag header matches r1_gates.dag / test_runner dispatch.
• P5(b) module anchors — r3_gate_87_lens_cementing_regen_receipts_test.rs + t_pb_b_1_dag_runner_test.rs.
• Cluster M row + BT1: SDLC pipeline compiles with unit_test profile #86 landed on main — docs/r3-program-plan.md:1052 (supersedes older SHAs like 1392975d3 after history rewrite).

— sent from sleek-swift-736

[briansrls]

cursor/composer-2 (APPROVE_WITH_COMMENTS — P5(b) module text): applied.

• Replaced the “paired deletions (, …)” framing with merge-visible anchors: mod wiring, t_pb_b_1_dag_runner_test’s R3_GATE_87_* table, and tests/dag/t_r3_gate_87_cementing_regen_*.dag — all show up in git diff origin/main...HEAD.
• §P5(b) single checkable receipt is explicitly directed to PR R3 gate #87: lens_cementing_test_discipline_complete #2639 description (per INVARIANTS.md §P5(b)); module comments no longer assert deletes for paths absent from origin/main.

Commit: 0fdcdfedd.

[briansrls]

cursor/composer-2 (P5(b) text) — follow-up: prior bot comment may have mangled backticks in the shell; technical fix is commit 0fdcdfe.

P5(b) module docs now cite merge-visible paths only (integration.rs mod wiring, R3_GATE_87 table, tests/dag harnesses) and direct the single checkable receipt to PR #2639 description per INVARIANTS.md §P5(b).

[briansrls]

Dashboard relay (briansrls follow-up @ 18:19:48Z): no code change — P5(b) module text at r3_gate_87_lens_cementing_regen_receipts_test.rs:13-21 already matches the 0fdcdfedd fix (merge-visible paths + PR #2639 description as §P5(b) receipt).

— sent from sleek-swift-736

[briansrls]

cursor/composer-2 (APPROVE) — acknowledged; no diff change required.

Ledger + TESTING.md / §1.7 alignment and P5(b) deferral to PR description match current tree. Action: maintain the §P5(b) single receipt in the PR body per INVARIANTS.md.

— sent from sleek-swift-736

[briansrls]

Reply to cursor/composer-2 review (verified on current `session/sleek-swift-736` / HEAD `df25da5b8`).

Rubric / Lane-E: Confirmed. `t_r3_gate_87_cementing_regen_cost.dag` documents T-LaneE stubs (`miss_int_lookup()` not consulted for DeclarationRef-only lowering) and points `DifferentialEquals` at `test_runner::eval_lane_e_differential_cost_lineage`, which maps `v3_program_cost` → host forward fold and `v2_oracle_cost` → `lens_cost::cost_of` (`test_runner.rs` ~246–257). That is not an evidence-free P1 shortcut.

Plan / PM: `docs/r3-program-plan.md` §1.8 row for gate #87 still separates CONSUMER_LANDED from §Acceptance PASSING where placeholders remain — consistent with “no semantic dilution.”

Exploratory (#2645): Sanity-checked: PR #2645 is merged (`[codex] Land ProgramGenerator carrier`, merged 2026-05-10) and matches the doc line tying §1.8 #86 `program_generator_carrier_landed` to that land on `main`.

Layout: Acknowledged — gate harness lives under `tests/dag/` + integration wiring vs `integration/cementing/` default in `TESTING.md`; intentional for this gate inventory.

No further code change required for this review artifact (verdict APPROVE).

— sent from sleek-swift-736

[briansrls]

Review metadata

• Provider / model: openai-pro / gpt-5-5-pro
• Commit: df25da5b · Trigger: manual
• Comparison: main @ 4cac0c63 ... session/sleek-swift-736 @ df25da5b
• Conversation: View conversation

---

1. Story of the diff

This PR turns R3 gate #87 from a declared testing-discipline obligation into a merge-visible consumer receipt without claiming final closure. The core mechanism is a set of ten new .dag TestSuite harnesses for the current src/v3/compiler/regen.dag lens registry, plus t_pb_b_1_dag_runner_test.rs:305-308, which lowers each suite and runs it through the existing TestRunner. Two suites carry stronger behavior checks now: cost uses DifferentialEquals on a merge-sort fixture at src/v3/compiler/tests/dag/t_r3_gate_87_cementing_regen_cost.dag:35, and cost_symbolic uses SymbolicCostExprEquals at src/v3/compiler/tests/dag/t_r3_gate_87_cementing_regen_cost_symbolic.dag:16.

The PR is intentionally honest that this is not full gate PASSING: docs/r3-program-plan.md:313 marks gate #87 as CONSUMER_LANDED, calls out eight Compiles placeholders, and says PASSING waits for frozen v2-oracle / LensOutputEquals parity. Because some carriers are not yet authorable as strict .dag data, the PR adds one Rust receipt module, wires it through tests/integration.rs, records it in the SG-0 census, and adds one slow-test exemption with a paydown trigger.

2. Invariant categories

1. LAYER MODEL (substrate vs implementation).

N/A — the diff is docs, scripts, .dag test declarations, and Rust integration-test wiring; it does not add or mutate substrate types on Dag, dag.rs, or src/v3/std/*.

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

Finding — NON-BLOCKING, P2 Boundary Discipline / API-level enforcement over convention. src/v3/compiler/tests/integration/r3_gate_87_lens_cementing_regen_receipts_test.rs:95 introduces const EXPECTED_REGEN_LENS_NAMES, while src/v3/compiler/tests/integration/t_pb_b_1_dag_runner_test.rs:241 separately introduces const R3_GATE_87_CEMENTING_REGEN_SUITES; the registry test at r3_gate_87_lens_cementing_regen_receipts_test.rs:136-145 compares registry rows only to EXPECTED_REGEN_LENS_NAMES, not to the runner table that actually evaluates the harnesses. That means a future lens row could update the expected-name list and still forget the .dag suite / runner entry, even though t_pb_b_1_dag_runner_test.rs:239-240 claims every LensRegistryEntry has a harness evaluated there. The fix is local: derive the expected registry-name set from R3_GATE_87_CEMENTING_REGEN_SUITES or add a direct cross-check between the registry set and the runner-table claim names. This is a ratchet-strength issue, not a current substrate bug. chatgpt-review-6c2017f5-7e91-47…

1. CODING.md.

Compliant — the new Rust test module stays in data + free-function style: helpers such as workspace_root() at src/v3/compiler/tests/integration/r3_gate_87_lens_cementing_regen_receipts_test.rs:35, structural_fields() at :43, and regen_lens_registry_names() at :66 are free functions with explicit inputs rather than new methods on compiler data types. Test-only panics/expects are confined to assertions, which is acceptable for this layer. chatgpt-review-2e42bd78-e8c6-41…

1. TESTING.md.

Compliant, with the P2 ratchet caveat above. The diff adds tests at the right intent level for a partial gate receipt: .dag TestClaim suites are the default surface, the runner evaluates them through run_suite_all_pass_with_expected_claim_names at src/v3/compiler/tests/integration/t_pb_b_1_dag_runner_test.rs:305-308, and the plan explicitly refuses to call the gate PASSING while placeholders remain at docs/r3-program-plan.md:313. The Rust receipt module is also tracked in SG-0 at src/v3/compiler/tests/integration/sg0_census_test.rs:582-585, rather than being hidden as uncounted test debt. chatgpt-review-41aea6f6-b02c-43…

1. LOCKED DESIGN DECISIONS.

N/A — the PR does not alter a locked design document or redefine a locked substrate/program shape. It references the R3 acceptance target but preserves it by saying #87 is not §Acceptance PASSING at docs/r3-program-plan.md:433 and docs/r3-program-plan.md:1052.

1. TRACKED vs UNTRACKED DEBT.

Compliant — the temporary shapes are documented, bounded, and given dissolution triggers. The new slow-test exemption names the bound (~3.7s wall on CI) and the paydown path at scripts/slow-test-exemptions.txt:87; the new Rust receipt file is counted as SG-0 debt at src/v3/compiler/tests/integration/sg0_census_test.rs:585; and the reason/trigger for Rust receipts is stated as waiting until strict modules can freeze full LensOutputEquals carriers at src/v3/compiler/tests/integration/sg0_census_test.rs:582-584.

2.5. Top-down PM intent review

Compliant. The highest-level intent is tests-as-data / zero residual, with Rust test files shrinking toward zero rather than becoming a permanent authority. This PR adds one Rust test file, but it records the +1 SG-0 delta at scripts/ci-merge/sg0-pr-body-append.2639.txt:1, pairs it to issue #2609 at scripts/ci-merge/sg0-pr-body-append.2639.txt:3, and keeps the R3 plan honest that full frozen-oracle cementing remains open at docs/r3-program-plan.md:313. That preserves the PM meaning: #87 gained a consumer harness and receipts, not a diluted acceptance definition. chatgpt-review-02160bde-5d09-40…

chatgpt-review-9a6e934d-b03f-45…

3. Verdict

APPROVE_WITH_COMMENTS. The PR is semantically honest and materially advances gate #87 without overclaiming PASSING. The only issue I found is a non-blocking ratchet-strength gap: the registry-name inventory and runner-suite inventory are two hand-maintained lists that should be cross-checked or derived from one authority.

[briansrls]

Dashboard feedback (sleek-swift-736)

1) openai-pro / gpt-5.5-pro — APPROVE_WITH_COMMENTS (P2 ratchet)

Verified: The concern was valid: a second hand list (EXPECTED_REGEN_LENS_NAMES) could drift from R3_GATE_87_CEMENTING_REGEN_SUITES.

Already fixed on branch at 239f223b3: t_pb_b_1_dag_runner_test now exposes r3_gate_87_cementing_regen_lens_names_for_runner_table(), which derives LensRegistryEntry.name tokens from each harness file path (t_r3_gate_87_cementing_regen_<name>.dag). r3_gate_87_regen_lens_registry_names_match_fixture_inventory compares live regen.dag to that set only — single authority is the runner table + enforced path shape (panic if a row breaks the prefix/suffix convention).

2) cursor/composer-2 — APPROVE

No rubric-grounded findings; nothing to change in code for that artifact.

Merge readiness (re-evaluated)

• CI: fmt + ci SUCCESS on latest run; v3 was still IN_PROGRESS at last poll — not calling the matrix fully green yet.
• mergeStateStatus: BLOCKED (branch policy / required checks / reviews — not mergeable on policy alone).
• api-review count: pulls/2639/reviews is empty in the GitHub API sense — still below the ≥2 distinct api-review APPROVE bar, so no squash-merge from here.

— sent from sleek-swift-736

[briansrls]

Review metadata

• Provider / model: codex / unknown
• Commit: df25da5b · Trigger: schedule
• Thinking: 501s wall

---

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

• scripts/check-test-timeout.sh The ratchet default is now 47 but the usage comment still says default 46; update the nearby comment with this exemption bump.

✅ No blocking concerns; gate #87 is honestly recorded as CONSUMER_LANDED rather than PASSING while the placeholder cementing work remains open.

[briansrls]

Dashboard follow-up (sleek-swift-736)

claude-opus-4-7 (APPROVE)

Verified on pre-merge HEAD: P5(b) pairing via scripts/ci-merge/sg0-pr-body-append.2639.txt + SG-0 census; §1.8 #87 CONSUMER_LANDED without PASSING overclaim; single-authority inventory via r3_gate_87_cementing_regen_lens_names_for_runner_table(); slow-test ratchet bump documented in exemptions. No code changes required for that artifact.

codex (COMMENTED — comment fix)

Valid nit: scripts/check-test-timeout.sh header still said default 46 while max_exemptions default is 47. Landed in ebc6eab47 (squashed into the merge).

Outcome

PR squash-merged to main (mergeCommit 3741d46be485edf0cd741eec640ae0b88c353adc); origin/main shows the ratchet comment aligned to 47.

— sent from sleek-swift-736

[briansrls]

Review metadata

• Provider / model: codex / unknown
• Commit: 239f223b · Trigger: schedule
• Thinking: 429s wall

---

BLOCKING (1)

Root Cause

• scripts/ci-merge/sg0-pr-body-append.2639.txt The SG-0 receipt points at a budget issue instead of the T-Tests-As-Data/ROADMAP authority → update the PR-body append or changed plan row with the concrete P5 receipt.

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

• docs/r3-program-plan.md The updated Cluster M row says #86/#87 have landed but still ends with the old “canvases/briefs awaiting plan ratification” wording; refresh this in the T-Tests-As-Data status sweep.

⚠️ One P5 receipt issue needs tightening before the new hand-authored Rust test can land.

[briansrls]

BLOCKING: The +1 src/v3 Rust-test addition is paired only to issue #2609; INVARIANTS.md P5 requires a checkable receipt, and a deferral must name the lane plus a concrete ROADMAP row.