Repository navigation
Conversation
…§11 subsumption model (operator-direct emit_rust.dag analysis) Operator delivered substantive algebraic analysis of emit_rust.dag's dissolution coverage gaps. This PR lands the three architectural moves they identified: [A] Three new producer stages (§10.1 + §10.2) — the algebraic move 'broader rules prevent repeats': - v4.lens.decision_tree_shape — normalizes closed-vocab branch dispatch REGARDLESS of source spelling (match / if-else chain / string-eq chain). Replaces match-only detection with shape-detection-regardless- of-syntax. Carries KeyVocabulary (resolved ConstructorId set, authoritative not by-spelling), per-branch BodyShape (α-renamed expression tree, same algebra as L1.12.b's body normalization), missing-key behavior, and source-form fact-of-finding. - v4.lens.generated_forest_shape — the META scope of L1.13's skeleton- collapse: catches fns that walk a closed coproduct's variant-set AND emit a per-variant target artifact (text or AST). Reuses SkeletonGroup + NormalizedArmBody from match_arm_skeleton at the emission scope (Practice 11 — same mechanism, two scopes). - v4.lens.target_syntax_string_shape — classifies string-construction graphs as DataString vs TargetGrammarTokenSequence vs Unknown. Gated on a per-target TargetGrammarTokenSet substrate carrier; rust.dag's LanguageModel (PR #3476) is the canonical first instance. Targets without a TargetGrammarTokenSet → Unknown → lens fails closed (no false positives). §10.3 dissolution stage's consumes: list updated for all three. [B] Six sub-signature lens bindings — all bound under existing L1.x parents (no new top-level IDs per the standing 2026-05-19 directive); each shares the parent's coverage_defect_* acceptance key with a new diagnostic-payload Trigger discriminator (same pattern as L1.12.b under L1.12): - L1.4.c PolicyTableAsFunctions — function whose body IS a typed table over closed vocabulary. Consumes decision_tree_shape. Seed kill: rust_string_policy_for_naming vs rust_internal_policy_for_naming. Subsumes the prior 'L1.13.b match-as-typed-table' reservation (L1.13's TotalMap dependency text retired to L1.4.c). - L1.5.b GeneratedForestCollapse — meta-walk emitting per-variant target artifacts that collapse to K skeletons. Consumes generated_forest_shape. Seed kill: emit_enum_shared_accessors. - L1.10.c NameDiscriminantBypass — spelled-name dispatch where resolved identity is available. Consumes decision_tree_shape with source_form = StringEqChain. Seed kill: variant_encoding_is_string_variant shape predicates. - L1.10.d TargetSyntaxByConcat — grammar-token concat that should be a TargetSurfaceNode. Consumes target_syntax_string_shape. Seed kills: emit_rust_block_stmts / emit_typed_record_lit / emit_rest_url_line / emit_rest_body_line; whole emit_rust.dag is a pile-up of L1.10.d findings. - L1.11.b PlausibleScalarFallback — missing-arm fabricates scalar literal used as target authority. Consumes decision_tree_shape + target_syntax_string_shape. Seed kills: emit_rest_auth_line's None=>'Authorization' / None=>'x-api-key' defaults; service base URL 'http://localhost' fallback. - L1.13.b DecisionTreeCollapse — generalizes parent L1.13's skeleton-collapse beyond match expressions to any closed-vocab decision tree. Same classification (PureTemplate / Outlier / etc.) regardless of source-form. Consumes decision_tree_shape. §5.0 / §5.1 cross-reference rows added/extended for all six. [C] §11 Subsumption model — new section addressing the operator's side-thought: 'what if the dag has multiple lens violations simultaneously — diagnosis should find the minimal pattern immediately, not surface one-at-a-time creating a fix → reveal → fix → reveal cycle.' Introduces DissolutionSubsumption substrate carrier with two verification kinds (MechanicalReverification via TestClaim re-run / ProducerStageDerivation via consumes: graph). Diagnostic surface ordering: roots first (Apply X — closes 47 other findings), leaves on root-rejection, subsumed-leaf preview. Worked example: the rust.dag LanguageModel subsumption-root for emit_rust.dag's 100+ findings under L1.5.b / L1.10.c / L1.10.d / L1.13.b / L1.4.c / L1.11.b. Renumbered prior §11 'Open — audit of current coverage' to §12. All sub-signatures share the producer-stage architecture under Facts Flow Forward / P2: one shared producer (decision_tree_shape / generated_forest_shape / target_syntax_string_shape) consumed by multiple downstream lenses. No new producer stages are minted per lens — same discipline that made L1.12.b + L1.4.b co-define cleanly. Enforcement gates uniformly require: (i) producer stage lands with the documented schema, (ii) parent diagnostic payload extends with the appropriate Trigger discriminator, (iii) for sub-signatures needing substrate carriers (TargetGrammarTokenSet, TotalMap), at least one canonical first instance lands as migration target. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
…2-decision-tree-producer-families
….13.c relocation)
Operator delivered substantive review with 7 specific tightenings. All
addressed in this commit:
[1] L1.13.c TotalMap too concrete for payload-bearing variants
(rust_serde policy's StripPrefixAndSnakeCase { prefix } etc. would
imply infinite table). Two clean shapes now distinguished:
- Pattern A: TotalMap<K, V> — for finite payload-free keys + fully
materialized values
- Pattern B: TotalPolicy<K, Context, RowTemplate> — for payload-
bearing key vocabularies; row templates reference payload fields
by name and materialize per-call by substituting runtime payload
into the template's holes
Auto-fix infers the pattern from KeyVocabulary's payload-bearing-ness.
[2] decision_tree_shape needs 04_infer + typed-subject rule.
StringEqChain now produces KeyVocabulary ONLY when the compared
subject has a UNIQUE resolved closed carrier per 04_infer. Ambiguous
or open cases get a new OpenOrAmbiguousNameDispatch source_form
variant; lens consumers (L1.10.c) do NOT auto-fire on it (open-
vocabulary escape preserved). Producer consumes: list extended.
[3] target_syntax_string_shape classification inconsistency fixed.
'Absence of TargetGrammarTokenSet → DataString' was wrong (absence of
substrate is not evidence of data). Now: no TargetGrammarTokenSet for
any target the string might reach → Unknown. Spelled out the four-
case closed predicate (1-no-substrate / 2-no-grammar-token-evidence-
in-data-sink / 3-grammar-token-evidence-in-target-sink / 4-other).
NO numeric token-density threshold — entirely closed structural test
(addresses tightening #6).
[4] L1.11.b needs broader use-site producer than target_syntax_string_shape.
New producer added: v4.lens.scalar_authority_use_shape produces
Map<ScalarRhsId, ScalarAuthorityUseShape> with closed ScalarUseRole
coproduct: TargetSyntaxUse / IdentifierUse / UrlUse / HeaderNameUse /
FilePathUse / ResourceCoordinateUse / DataStringUse / UnknownUse.
Gated on scalar_authority_use_registry substrate carrier (per-
consumer ScalarUseRole declarations). L1.11.b consumes this producer
NOT target_syntax_string_shape (different classification axes:
production-site vs use-site role).
[5] Generated-forest asymmetric coverage Escape now requires substrate-
derived subset witness (VariantCoverage row citing structural
predicate). Local unregistered filters like
'children |> filter(child => child.children |> count > 0)' do NOT
pass — those are L1.5.b / L1.10.c / L1.13.b territory depending on
the filter's shape. The escape requires structural authority.
[6] TargetSyntaxByConcat threshold language replaced with closed
structural predicate (addressed in #3 via the four-case predicate in
producer-stage spec; L1.10.d's Signature text also updated to
reference the closed predicate explicitly, not 'classification
threshold').
[7] L1.4.c taxonomy move: L1.4.c (Policy-table-as-functions) relocated
to L1.13.c (Table decision-tree) under the Skeleton-collapse family.
Operator: 'function body is a typed table is not really a carrier
clone; closer to decision-table dissolution.' Mechanism is
decision_tree_shape (same as L1.13.b), not type-declaration-shape
(L1.4/L1.4.b's). L1.4.c original site replaced with a 'Moved' notice;
L1.13.c added under L1.13.b with the payload-aware TotalPolicy
refinement from item #1 baked in.
Tables + cross-refs updated in lockstep:
- §5.0 L1.4 row (L1.4.c removed); L1.13 row (L1.13.c added)
- §5.1 L1.4.c row removed; L1.13.c row added; L1.13.b discriminator
extended to include TableDecisionTree variant
- §10.1: L1.4.c→L1.13.c across producer-row consumers
- §10.2: producer-shape spec edits per items 2/3/6; new
scalar_authority_use_shape producer added; stage count
'Eight'→'Nine'
- §10.3: dissolution consumes list extended with scalar_authority_use_shape
- §10.2 closing paragraph: cross-stage routing updated to reflect the
relocations + new producer
- §11 subsumption worked example: L1.4.c→L1.13.c references updated
All seven tightenings preserve the producer-stage architecture under
Facts Flow Forward / P2 — no producer-stage duplication; consumers
cleanly partitioned per classification axis.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
…table' prose (cursor APPROVE_WITH_COMMENTS doc-hygiene) cursor APPROVE_WITH_COMMENTS flagged INVARIANTS P2 (single authority) issue: I rebound L1.13.b to DecisionTreeCollapse and moved the F16 match-as-typed-table case to L1.13.c, but several pre-existing prose passages from origin/main still described L1.13.b as the 'future per-arm-name / match-as-typed-table sub-signature.' Two incompatible authorities for the same sub-signature ID in one design doc. Reconciled four residual references: 1. L1.13's borderline-case discussion (:2227-2233 originally) — the prose listed 'L1.13.b (future)' OR 'separate match-as-typed-table lens' as the two possible bindings for the manual_test_claim_for_manual_anchor pattern. Both pointers now unified under L1.13.c. Added a historical-traceability note explaining the original two-option phrasing for readers who hit the prior text. 2. L1.13's Kills section (:2103) — 'proposed L1.13.b that detects arm-name-parameterizes-reference pattern' → 'L1.13.c Table decision-tree' with the same historical note inline. 3. v4.lens.match_arm_skeleton producer comment (:2789-2794) — 'L1.13 auto-fix and any future L1.13.b sub-signature' + 'reusable by: L1.13, future L1.13.b (per-arm-name-parameterized-reference sub-signature), future match-as-typed-table lens' → 'L1.13 auto- fix needs (and the L1.13.c table-decision-tree sub-signature reuses it for per-arm-RHS-referencing-typed-data cases)' + 'reusable by: L1.13 (base, MatchExpr-scoped), L1.13.c (table decision-tree)'. 4. v4.lens.decision_tree_shape producer comment (:2881-2882) — 'reusable by: L1.13, L1.13.b, L1.13.c, L1.10.c, future match-as-typed-table sub-signature' → cleanly enumerated by binding without the residual 'future match-as-typed-table' pointer (which is now L1.13.c). After reconcile, the only remaining 'future L1.13.b' mentions in the doc are inside the explicit historical-traceability notes (intentional — they explain the binding history for readers who encounter the prior text). Net effect: L1.13.b now consistently binds to DecisionTreeCollapse throughout; L1.13.c consistently binds to TableDecisionTree throughout; the match-as-typed-table case has one single authoritative pointer (L1.13.c) across the whole doc. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
…nter, not authority (codex APPROVE_WITH_COMMENTS Practice 9 / P2 reconcile) codex APPROVE_WITH_COMMENTS on PR #3489 flagged §5.1 as a Practice 9 / INVARIANTS P2 violation: the section opened with 'Downstream consumers MUST use the canonical key names enumerated here. The lens suite is the single authority...' That framing makes the design doc authoritative for coverage_defect_* keys when the actual substrate authority is src/v4/lens/coverage.dag's declared rows. Classic parallel-authority / maintained-ledger-doc anti-pattern (operator- direct standing 2026-05-19 retired this class). Fix: - §5.1 header retitled 'L1.x acceptance-key names — substrate registry pointer.' - Opening prose rewritten: explicit 'Authority pointer (not authority itself)' framing — src/v4/lens/coverage.dag is the authoritative registry; this doc describes but does NOT own. If table and substrate disagree, substrate wins. - Added explicit cite of the no-prose-ledger discipline (Practice 9 / INVARIANTS P2 / operator-direct standing 2026-05-19). - Reframed table purpose as 'reading scaffold for readers walking the lens family' — claims about substrate state at this revision, not definitional rules. - 'Downstream-consumer guidance' clarified: code reads keys from src/v4/lens/coverage.dag directly, NOT from this table. Substrate- driven projection, not doc-rebase-driven. - 'Migration notes' closing line rewritten: future taxonomy changes update the substrate first; this table follows. Net effect: doc describes the registry; substrate IS the registry. No change to row content (the keys are the same and they're what src/v4/lens/coverage.dag already declares); just the authority framing reconciled. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
…t + §11.4 TotalPolicy refinement) codex BLOCKING wrap on PR #3489 sha 6c3df18 raised three findings; finding #1 (substrate row landing for coverage_defect_skeleton_collapse) was addressed in auto-snapshot WIP commit b5c6996 (added SkeletonCollapse variant + data row to src/v4/lens/coverage.dag + updated the integration test ratchet's EXPECTED_ACCEPTANCE_ROWS). This commit addresses findings #2 + #3: [#2] L1.11.b enforcement-gate text still referenced target_syntax_string_shape as the prerequisite producer after I'd split the use-site classification out to scalar_authority_use_shape. Fix: Enforcement gate text rewritten to require BOTH - (a) v4.lens.decision_tree_shape lands (for None-arm scalar-RHS detection) - (b) v4.lens.scalar_authority_use_shape lands (for use-site authority- role classification) plus per-ScalarUseRole-variant gating via scalar_authority_use_registry. Explicit statement: target_syntax_string_shape is NOT a prerequisite — it classifies the production-site (L1.10.d's axis); L1.11.b consumes the orthogonal use-site-role axis. Only referenced for the TargetSyntaxUse cross-reference variant. [#3] §11.4 subsumption worked example R-root-B still described the root fix as TotalMap-only after the payload-aware TotalPolicy refinement landed in earlier commit. Fix: R-root-B prose rewritten to distinguish TotalMap (finite payload- free) from TotalPolicy<K, Context, RowTemplate> (payload-bearing policies, per the L1.13.c payload-aware refinement). The serde-policy case specifically called out as TotalPolicy (because VariantNaming has payload-bearing constructors like StripPrefixAndSnakeCase { prefix }); auth-defaults case may be TotalMap or TotalPolicy depending on auth-source vocabulary. All three codex BLOCKING findings now reconciled across substrate + design doc. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Contributor
Author
briansrls
deleted the
session/smart-lynx-142-decision-tree-producer-families
branch
June 1, 2026 18:43
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.
Auto-opened by session-dashboard for session
smart-lynx-142.Pushing to
session/smart-lynx-142-decision-tree-producer-familiesadvances this PR.Worker attestation
Before flipping this PR to ready for review, confirm each item:
npm test,cargo test) and the result.Closes #Ndirective.Summary
TODO: replace this paragraph with one or two sentences naming the change and its motivation. Reviewers read this first.
Test plan