cycle(54): self-coherence cleanup + β APPROVED#59
Merged
Conversation
…tch contract γ creates cycle/54 from origin/main @ 3efde94 (post-S1+S2+S3+S4 merge). Scaffold names scope, AC mapping (8 ACs), α/β/operator/δ dispatch contract, known constraints. α and β are dispatched as separate subagent sessions per CDD §1.4 Triadic rule — no γ+α collapse this cycle. Wave-1 carry-over disclosed to α/β briefs: session-bound git proxy 403 recovery pattern; OCaml toolchain absent in dispatch sandbox (B-severity ci-status finding is γ-authorized).
…tric c_sigma_num) Each kata gains a [baseline] block capturing: - baseline_engine_commit/version/command/mode/config_hash/input_file_hashes - per-axis α, β, γ (recorded triples where available; inferred where not) - c_sigma_math and c_sigma_num (geometric (α·β·γ)^(1/3)) - zero_component_present, numeric_floor_applied - rationale_category (aggregate-correction or frontier-tightening) Kata-04 and kata-05 carry frontier-tightening rationale to retain documented 'moving frontier' margins. Kata-01/02/03 inherit aggregate-correction with conservative range retention (per-axis triples on those katas were inferred from README signal narrative, not explicitly recorded; exact c_sigma_num CI re-baseline pending).
THESIS, QUICKSTART, ARCHITECTURE, OPERATOR-MANUAL, katas/README, and all five per-kata READMEs now: - describe geometric C_sigma_math and C_sigma_num (canonical v3.2, v0.10.0) - reference provenance.aggregate_math.C_sigma_math and provenance.aggregate_numeric.C_sigma_num in JSON examples and score-range references - remove the flat top-level c_sigma example - remove the arithmetic-mean (alpha+beta+gamma)/3 language Per-kata READMEs additionally record both the pre-cutover arithmetic reading (preserved as historical calibration) and the post-cutover geometric C_sigma_num value, plus a pointer to kata.toml [baseline] for the v0.10.0 provenance block.
…ing CI rule, RELEASE.md v0.10.0 AC5: CHANGELOG.md gets a 0.10.0 ledger row + detail section with a prominent migration note stating that pre-v0.10.0 frozen reports quote arithmetic aggregates and are not directly comparable to post-cutover output. References #49 master + four predecessor cycles (#50, #51, #52, #53). AC7: scripts/check-forbidden-wording.sh — bash script (no external deps; uses git-diff -U0 hunk parsing) that compares HEAD to a base ref (default origin/main) and rejects newly-added 'Operational acceptance', 'Operationally accepted', 'self-coherence ACCEPT', 'release criteria satisfied' outside docs/{tier}/{bundle}/{X.Y.Z}/, docs/archive/, .git/, _build/, .cdd/, CHANGELOG.md, RELEASE.md, and this script itself. Wired into .github/workflows/ci.yml as a new forbidden-wording job that runs on push + pull_request. Existing historical occurrences in excluded paths pass — the check is strictly forward-only. AC8: RELEASE.md rewritten as the v0.10.0 release notes describing the canonical-v3.2-cutover wave end-to-end (engine cutover via sub-cycles #50-#53, cleanup pass via #54). References #49 master.
engine/ocaml/test/test_target_registry.ml: 4 hermetic assertions
covering AC6's 5 bullet points:
AC6.1 parse_registry returns Ok for targets/registry.tsc and
asserts registry_format = "tsc-target-registry/0.1"
AC6.2 exactly the three names {spec, engine, repo} are registered
AC6.3 resolve_target_path returns Ok for each name -> the canonical
targets/{spec,engine,repo}.tsc manifest path
AC6.4 parse_manifest returns Ok with a non-empty manifest_include
(or manifest_include_targets) for each target
AC6.5 file-expansion produces > 0 files for each target via the
same root/path semantics main.ml uses (the expand_glob helper
is reproduced verbatim in the test so the runtime semantics
are exercised end-to-end)
The test walks up from cwd to find repo root (VERSION + targets/
registry.tsc co-located) which matches the dune runtime convention.
Wired into engine/ocaml/test/dune as test_target_registry. Existing
tests (test_response_schema, test_ood, test_cross_target,
test_mechanical, test_coherence, test_kata) coexist unchanged.
…review-readiness All eight ACs addressed and mapped to commit SHAs: AC1 katas re-baselined commit 1fa777e AC2 active docs canonical commit 2f45da8 AC3 project.tsc removed commit d6e8453 AC4 frozen-snapshot verified clean (no edits — banner absent) AC5 CHANGELOG migration note commit a274923 AC6 target-registry smoke tests commit 2242680 AC7 forbidden-wording CI rule commit a274923 AC8 release artifacts at 0.10.0 commit d6e8453 + a274923 Known debt (B-severity, explicit): - OCaml toolchain absent; dune build/runtest deferred to CI - Session-bound git proxy 403; cycle/54 stuck at 0981855 on origin, work pushed to cycle/54-fix as operator mirror - Kata-01/02/03 baseline triples inferred (cycle-34 recorded only the arithmetic-mean aggregate); pending CI re-baseline - Kata-04/05 retain frontier-tightening rationale (intentional) - Historical-report migration explicitly out of scope per #54 Ready for β.
…wording check The workflow that invokes scripts/check-forbidden-wording.sh names the phrases it checks for in its job-comment header (lines 102-108). That self-reference is documentation, not a new occurrence of the retired wording — analogous to how the script itself is on the exclusion list. Without this exclusion, running the script against the cycle/54 head finds 4 'forbidden' phrases in the ci.yml comment block, which would fail the very job that runs the script. Self-application now passes: $ bash scripts/check-forbidden-wording.sh main check-forbidden-wording: pass (base ref: main)
Contract integrity: all 11 rows yes/n/a. AC1–AC8 verified at HEAD 4d293ff. Honest-claim sub-checks (13) all pass — AC4 banner absence structural (empty diff), kata baseline geometric c_sigma_math reproduced to 4dp, AC7 self-application defect fixed at 431293f and re-run exit 0, AC8 version-consistency PASSED. CI gate (rule 3.10) deferred per γ-authorization — no OCaml toolchain in dispatch sandbox; classified B-severity `ci-status: defer to CI run`, operator verifies on the PR. γ-scaffold present (rule 3.11b satisfied). No D / C / B / A findings. Merge instruction: open PR cycle/54-closeout → main with `Closes #54`.
The new test_target_registry.ml (commit 2242680) used unwrapped nested match expressions and the `fail (...) ; []` sequence pattern. Under dune's default :standard flags on the CI OCaml 5.2 toolchain, this emits build-breaking warnings: - Inner `match ... with | ... | ...` without parens/begin..end risks fragile pattern-arm attribution to the outer match. - `fail (Printf.sprintf ...) ; []` sequences fail (which returns 'a via exit 1) with a list literal; warning 21 'this statement never returns' / warning 10 'this expression should have type unit' may be promoted to errors. This fix: - Wrap every nested `match` in `(match ... with | ... | ...)` parens, matching the convention used in engine/ocaml/test/test_cross_target.ml (which has been green on main throughout this wave). - Drop trailing `; []` after `fail` calls; since `fail : string -> 'a`, the empty list is unreachable and `'a` already unifies to the expected list type. No behavior change: every test still calls Target_registry.parse_registry / resolve_target_path / parse_manifest with the same arguments and checks the same invariants. AC6 oracle preserved. Addresses PR #59 build job failure (no other OCaml file changed in this cycle vs origin/main; test_target_registry.ml is the only diff). Refs: #54 Co-authored-by: alpha <alpha@tsc.cdd.cnos>
Documents the post-CI drift diagnosis and fix: - δ-at-gate symptoms triaged (c_sigma rename, comparison_to_json arity, w21) — none present in this cycle's diff. - Actual drift located in test_target_registry.ml: unwrapped nested match expressions and fail; [] sequence pattern, both triggering build-breaking warnings on OCaml 5.2 with dune's default :standard flags. - Fix applied at 7a23890; AC6 oracle unchanged. - Toolchain still absent; verification via convention parity with test_cross_target.ml which has been green on main throughout the wave. Refs: #54 fix-r1 Co-authored-by: alpha <alpha@tsc.cdd.cnos>
Owner
Author
α fix-round R1 — pushedFix SHA: Drift diagnosisPR #59's only OCaml change vs
Actual drift — OCaml grammar in
|
added 3 commits
May 14, 2026 07:01
…ggregate The v0.10.0 canonical-v3.2 cutover (cycle/50, commit 93c662c) replaced the flat [c_sigma : float] field on [Mechanical_scoring.result] with an [aggregate : Mechanical_scoring.aggregate] sub-record carrying [c_sigma_math], [c_sigma_num], [epsilon], [zero_component_present], [numeric_floor_applied]. The synthetic test fixture [mk_result] in test_cross_target.ml still used the pre-cutover [c_sigma = ...] form and would not compile against the current type. This fix constructs the aggregate via [Coherence.aggregate] (the same path [Cross_target.row_of_mechanical] uses internally) and assigns it to the [aggregate] field. The synthetic value carries: - [c_sigma_math]: strict geometric mean (collapses to 0 on zero α/β/γ) - [c_sigma_num]: ε-floored geometric mean (well-defined under degeneracy) - [epsilon]: from [Coherence.epsilon_default] - [zero_component_present] / [numeric_floor_applied]: as computed No behavior change to the test oracle: the test still calls [Cross_target.row_of_mechanical] / [report_from_results] with the same synthetic (α, β, γ) triples, and Cross_target re-derives the aggregate from r.alpha/beta/gamma.score regardless of r.aggregate, so AC2/AC3/AC4 oracles are preserved bit-for-bit. R1 disclosure: this is one of the three failures R1 mis-diagnosed as "not present in this cycle's diff." It was present on cycle/54-closeout HEAD d6a48b2 (and on origin/main itself); R1's grep was too narrow. Refs: #54 Co-authored-by: alpha <alpha@tsc.cdd.cnos>
… arg [Mechanical_scoring.compare] is declared as: val compare : ?config:config -> old_:Bundle.t -> new_:Bundle.t -> comparison With an optional argument followed only by labeled arguments (no unit or positional terminator), OCaml cannot decide when application is "complete" — the call [compare ~old_:bundle_a ~new_:bundle_b] returns a curried function value (still expecting either ?config or another positional), not a [comparison]. Passing that function value to [comparison_to_json] produces a type error: function-typed value where [Yojson.Safe.t] is expected. The fix passes [~config:Mechanical_scoring.default_config] explicitly at the call site so the optional is bound and [compare] resolves to a fully-applied [comparison]. Behavior is unchanged because [default_config] is the same value the optional would default to internally; the AC3 oracle (form-suffixed delta field names + value parity) is preserved bit-for-bit. This is failure #2 of the three the operator's δ-at-gate observed on cycle/54-closeout HEAD d6a48b2 — R1's grep missed it because the symptom is whole-call-site shape, not a single literal token. The brief's hint ("Pass ~config:Mechanical_scoring.default_config explicitly, OR terminate the partial application with ()") was acted on via the first alternative since the current .mli has no unit terminator. Refs: #54 Co-authored-by: alpha <alpha@tsc.cdd.cnos>
Records the three failures the δ-at-gate operator named on PR #59 build log at HEAD d6a48b2, the R2 fixes applied (commits 09842e5 + 2ab45c2), honest acknowledgment of R1's mis-diagnosis, and CI-green verdict on 2ab45c2. R1 had dismissed the operator-named symptoms as "not present in this cycle's diff" — F1 (c_sigma in test_cross_target.ml) and F2 (compare partial application in test_mechanical.ml) were in fact literal compile errors on the branch. F3 (warning 21 in test_target_registry.ml) was already resolved by R1 commit 7a23890; R2 re-verified. The lesson recorded plainly: audit by interface contract (does the literal call/access type-check against the current .mli?), not by file-diff (the failing site may be in a file untouched this cycle). CI verification on 2ab45c2: - build (job 75944432772): success at 07:11:05Z - run-katas (auto-discovered) (job 75944432816): success at 07:10:43Z Refs: #54 Co-authored-by: alpha <alpha@tsc.cdd.cnos>
Owner
Author
α fix-round R2 — CI greenFinal SHA:
Three fixes
R1 vs R2 — honest disclosureR1's drift section in The lesson, recorded in §Fix-round-2 of self-coherence: audit by interface contract (does the literal call/access type-check against the current What's on the branch now
Toolchain still unavailable in dispatch sandbox; both fixes derived by Ready for β R2. Generated by Claude Code |
Appends Round 2 section to .cdd/unreleased/54/beta-review.md covering the focused re-review after α fix-round R2. Round 2 verdict: APPROVED, zero findings. V1 — fix-round addresses the three named failures: ✅ F1 (09842e5): test_cross_target.ml c_sigma → aggregate sub-record F2 (2ab45c2): test_mechanical.ml compare ~config explicit binding F3: already resolved by α R1 commit 7a23890; re-verified clean V2 — rule 3.10 CI-green gate satisfied: ✅ build + run-katas success on fix-bearing commit 2ab45c2. HEAD 2c8dee7 is doc-only OCaml-identical; non-OCaml jobs already green. Reclassifies R1's deferred B-severity ci-status finding to satisfied. V3 — α R1 mis-diagnosis disclosed honestly: ✅ Self-coherence §Fix-round-2 + PR #59 comment both name the wrong inference and record the lesson (audit by interface contract, not by file-diff). V4 — frozen snapshots untouched in fix-round: ✅ git diff 4d293ff..HEAD on frozen dirs = 0 lines. V5 — release artifacts consistent at 0.10.0: ✅ VERSION, dune-project, opam all stable at 0.10.0. Scope discipline: git diff 4d293ff..HEAD -- engine/ocaml/lib/ is empty. No production code modified in fix-round. α stayed inside test-alignment authority. Merge instruction: merge PR #59 (cycle/54-closeout → main) via UI with Closes #54 in the merge commit. main is branch-protected; β does not direct-push. Refs: #54 Co-authored-by: beta <beta@tsc.cdd.cnos>
Owner
Author
β R2 review — APPROVEDVerdict: APPROVED, zero findings. Fix-round verification
Scope discipline
Findings (R2)None. Zero R2 findings; R1's sole B-severity Merge instruction (rule 3.11)Merge PR #59 ( Notes for cdd-iteration
— β R2 Generated by Claude Code |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Closes #54
Cherry-picks β review artifact (67dd306) onto closeout branch.