Extends ADR-0131.G.5's total-across aggregate branch with the existential
verb frame 'How many <unit> are there <cue>?' over the SAME closed cue
vocabulary (no cue-set widening). The solver already sums entity=None
total-across; the wall was purely the question parser's verb-frame coverage.
- Composes end-to-end: 'Jamie has 28 marbles. Kyle has 12 marbles. How many
marbles are there in total?' -> 40.0 (load-bearing, not inert).
- 23 real GSM8K problems now parse the question (advance past the Q-wall to
the statement-parse wall).
- wrong=0 HOLDS on the full 7,473-q corpus; train_sample byte-identical
4/46/0; no metric delta (composition-wall lesson, third instance).
- Cross-ADR discipline: 'What is the total number of <unit>?' is DEFERRED,
not contradicted — ADR-0131.G.5 pins it as a refusal probe; promoting it
must amend that ADR's closed-cue contract. test_total_number_of_still_deferred
locks the boundary.
Firewall: completeness guard (ADR-0191) + question round-trip refuse conjoined
('dogs and cats') and derived ('animal legs') units. G5 lane + synthetic-registry
wrong=0 capability-axis gate green; smoke 67 passed.
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
The discrete_count matcher gated the counted noun on a CLOSED ratified set
(observed_counted_nouns): 'Betty has 24 marbles' matched, 'Randy has 60 mango
trees' / 'Sam has 12 red apples' did not — purely because the noun was unseen.
Open the single-anchor possession/acquisition path to an open noun phrase
(adjective* + 1-3 word head, bounded by a stop-word lookahead so it never
swallows a trailing PP), keeping every other narrowness layer (proper-noun
subject, verb whitelist, single numeric token, no clause-split). Closed
observed nouns still match (capitalized compounds preserved); compound
enumeration stays closed.
Safe because ADR-0191 moved the wrong=0 guarantee downstream: an open-vocab
mis-parse hits the completeness guard + round-trip + branch-disagreement.
Proof: full real corpus 61->494 discrete_count anchors (8x), wrong=0 HOLDS,
zero confabulations.
Substrate PR — 0 metric delta by design (train_sample byte-identical 4/46/0;
the problems still need composition downstream). Value: the foundation every
discrete_count flip consumes, and empirical proof open-vocab is firewall-safe.
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
* fix(adr-0191): candidate-graph completeness guard — real-corpus wrong 5->0
The candidate-graph reader (serving) checked grounding + round-trip but had
no completeness obligation, so problems whose later clauses failed to parse
emitted a partial reading. Over the full 7,473-question real GSM8K train
split this confabulated 5 answers (wrong!=0) the 47-case train_sample hid;
2 were regressions from #488.
Add the missing admissibility leg (mirrors the derivation reader's verify.py):
every source quantity (all statements + question) must be consumed by the
chosen reading, else refuse. Refusal-only -> cannot create a wrong answer.
Number-sense is pack-authoritative (en_numerics_v1 parse_compound_cardinal +
lookup_multiplier + all 6 currency symbols) so it never disagrees with the
engine; aggregating initials expose consumed_value_tokens provenance.
Evidence: real-corpus wrong 5->0, correct held at 4; train_sample byte-
identical 4/46/0; G1-G5+S1+G3.1 green; smoke 67 passed; math_teaching_corpus
lane byte-identical.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* feat(adr-0191): committed full-corpus GSM8K microscope (standing wrong=0 + coverage instrument)
Promotes the throwaway tmp/ microscope that found the 5 confabulations into a
committed tool. Runs the canonical serving reader over any GSM8K corpus and
reports failures-first: correct/wrong/refused, every wrong answer by name,
refusal families, and the no-injection per-category coverage map that ranks
which injector to build next by real frequency.
Default corpus is the committed 47-case train_sample (always available);
--corpus path runs the full real split. This is the ADR-0191 follow-up: re-run
after every capability PR, not just train_sample — a flip is only real if it
does not widen the confabulation surface.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
* feat(adr-0189): comparative reading — anchor-verb widening + multi-word units
The candidate-graph comparative extractor (ADR-0131.G.2) read only has/have +
single-word units, so real-GSM8K comparatives ('Brooke does three times as many
jumping jacks as Sidney') didn't parse — a dark statement in 17 places blocking
15 of the 47 refused train_sample cases, despite the ADR-0123 solver already
supporting compare_additive/compare_multiplicative.
Widens the anchor-verb set (reusing legacy vetted lemmas + does/collected/
gained/studied…), EXCLUDING polarity-inverting verbs (lose/spend/give/sell/win)
to preserve wrong=0; admits 1-2 word units via the existing multi-word
_unit_grounds branch. Feeds the existing solver unchanged.
wrong=0 proven: G2_comparatives 29/29, G3 20/0, G4 32/32, train_sample 3/47/0
byte-identical; polarity-inverting verbs proven refused (failing-under-violation).
Chain composes correctly in isolation (146 -> 438). Flips 0 cases ALONE — every
comparative case needs a composing partner (aggregation / multi-word-noun
injection); this ships the component, not yet a flip.
- generate/math_candidate_parser.py: _comparison_anchor_verb widening + 1-2 word
unit slots in the two multiplicative comparative regexes.
- tests/test_adr_0131_G2a_*: 5 tests incl. polarity-inversion wrong=0 guards.
- docs/decisions/ADR-0189: gap, change, wrong=0 evidence, honest scope.
* feat(adr-0189a): first metric move 3/47/0 -> 4/46/0 (case 0024, comprehension-composed)
Case 0024 now SOLVES (answer 438) by composing three general comprehension
capabilities feeding the unchanged ADR-0123 solver:
1. day-of-week count enumeration: Sidney = 20+36+40+50 = 146
(_day_enumeration_candidates; derived sum grounds via first count token,
mirroring _embedded_quantifier; closed to the 7 day names)
2. comparative reading (ADR-0189): Brooke = 3 x Sidney
3. activity question 'How many <unit> did <Entity> <verb>?' (_Q_DID_RE)
Plus do/does/did added to the CandidateInitial anchor whitelist (production-
possession), admitted only via the closed day-enumeration shape.
wrong=0 PROVEN across every lane: all 8 capability axes wrong=0 (G2_comparatives
29/29, G3 20/0, G4 32/32, G5/S1/S3/S4 all pass), train_sample 4/46/0 wrong=0,
verify_lane_shas exit 0 (no pinned lane changed), generate_claims --check OK.
872 tests pass; new tests are failing-under-violation incl. wrong=0 guards
(non-day comma list not summed; polarity-inverting comparative verbs refused).
Re-baselined report.json + train_sample_coverage_report.json (latter also clears
pre-existing reason drift) + CLAIMS.md to the new 4/46/0 metric. Decode-not-guess:
0024 solved by READING its structure, not storing an answer. Remaining pre-existing
failures (G3 committed-report, telemetry) unrelated, fail on pristine main.
- generate/math_candidate_parser.py: day-enum extractor + _Q_DID_RE + does-anchor.
- tests/test_adr_0189a_day_enum_activity.py: 5 tests (incl. end-to-end 0024=438).
- docs/decisions/ADR-0189a + report.json/coverage/CLAIMS re-baseline.
* docs(adr-0186): sealed candidate-graph injector lane (resume ADR-0170 W2-W5 under ADR-0175 seal)
Topology audit found two disjoint GSM8K readers: the candidate-graph reader
owns the official 3/47/0 metric and already has divide/multiply/compare; its
wall is the recognizer->injection coverage gap (ADR-0170 W2-W5 backlog), not
arithmetic. The derivation reader (resolve_pooled) is a separate sealed organ
that cannot reach the goal without an unbuilt Phase-5 bridge.
ADR-0186 reconciles ADR-0170's injector roadmap with ADR-0175's serving seal:
develop W2-W5 injectors behind a default-off 'sealed' flag on inject_from_match,
measured on a new report_sealed.json, with frozen 3/47/0 byte-identical until a
reviewed Phase-5 promotion. wrong=0 gated on both paths.
Live-loop instrumentation found a schema-vs-extraction split: rate_with_currency
and temporal_aggregation matchers already extract full anchors (blocked on the
CandidateRate union / apply_rate primitive), while discrete_count/currency/
multiplicative are blocked on matcher extraction. And no refused case is one
injector away - every case is multi-statement, so the unit of measurable
progress is a target case's complete injector+composition set. Sequencing:
seal mechanism first, CandidateRate schema next, then a first complete
target-case unlock.
* feat(adr-0186): sealed candidate-graph injector lane mechanism (default-off)
Adds the seal mechanism for developing ADR-0170 W2-W5 injectors without
mutating the frozen serving metric:
- inject_from_match(match, sentence, *, sealed=False): when sealed=True,
consults the new _SEALED_INJECTORS table first; default-off never touches it.
- _SEALED_INJECTORS: empty at land (this PR ships the mechanism; the first
sealed capability is the CandidateRate schema, ADR-0186 §5.3).
- parse_and_solve(text, *, sealed=False): threads the flag to the per-statement
injection site. The seal is injector eligibility, not a forked reader.
wrong=0 guarantees (tests/test_adr_0186_sealed_injector_lane.py, failing-
under-violation): empty seal is a strict no-op; a registered sealed injector
admits ONLY under sealed=True and is invisible to the frozen path; train_sample
report stays 3/47/0. Verified: coverage probe 3/47/0 byte-identical, smoke +
architectural invariants green, 288 candidate-graph/recognizer tests pass (2
pre-existing failures unrelated, confirmed on unmodified main).
* docs(adr-0185): mark superseded by ADR-0186 (premise refuted by topology audit)
The topology audit proved ADR-0185's premise ('the engine cannot divide') is
true only of the derivation reader, which is disjoint from the candidate-graph
reader that owns the official 3/47/0 metric. The goal organ already divides;
its wall is injection coverage. ADR-0185 is retained as a record, not
implemented; cross-referenced from ADR-0186 §header.
* docs(adr-0184): scope the distinct-unit product rule — cut the product-of-all over-commit
The 47-refusal coverage diagnostic surfaced that the headline 3/47/0 (serving
recognizer) hides the sealed comprehension reader's real state: resolve_pooled over
the 50 real train_sample cases is 2 correct / 13 WRONG / 35 refused. The confuser
probe's wrong=0 was a misleading proxy — all 13 real wrongs are the whole-text
product-of-all, the unique complete candidate, committed unopposed.
Scopes the first lever, decided by MEASURING candidate refusal rules against the real
metric (correct up, wrong down on train_sample):
baseline 2 / 13 / 35
distinct-unit product (chosen) 2 / 8 / 40 <- cuts 5, zero coverage loss
product spans >1 clause 1 / 4 / 45 <- destroys correct 0003
drop all products 0 / 2 / 48
The distinct-unit rule: multiply/divide may compose DISTINCT units but a multiply
step whose operand repeats a non-empty unit already in the product (apples x apples,
cards x cards) is unit-incoherence -> refuse (unit^2 is never the answer). Empty-unit
operands exempt (0003 multiplies a blank-unit 0.75). Dimensional, not lexical
(ADR-0165-safe); refines verify.py clause 3 shared by self_verifies + classify.
Honest scope: 13->8, NOT 0. The remaining 8 are distinct-unit products in the wrong
shape (rate problems) = cue precision (ADR-0177 CP-2b), the next lever, NOT to be
faked with a per-case rule. Establishes the real scoreboard (resolve_pooled over
train_sample) and notes the ratification bridge (ADR-0175 Phase 5) as the separate
dependency for any of this to reach the serving headline.
Spec only; serving 3/47/0 untouched (verify is not on the serving path).
* feat(adr-0184): distinct-unit product rule — sealed reader real-GSM8K wrong 13->8
Cuts the over-eager product-of-all on real GSM8K. The sealed comprehension reader
(resolve_pooled over train_sample) was 2 correct / 13 WRONG / 35 refused; all 13 are
the whole-text product-of-all committed unopposed (0042->2.4M, 0048->19200,
0001->14400). This is the first lever measured against the REAL metric (resolve_pooled
over train_sample), not the curated confuser count.
Mechanism (verify._is_repeated_unit_product + classify_derivation downgrade):
a pure multiplicative chain whose operands repeat a non-empty unit forms unit^2
(apples x apples, cards x cards) -- never the answer; it is the product-of-all
multiplying independent groups. Such a product is classified `exempt`
(commit-INELIGIBLE), NOT removed. Empty units exempt (0003 multiplies blank-unit
0.75); divide exempt (feet/feet = legitimate count). Dimensional, not lexical
(ADR-0165-safe).
Implementation finding (folded into ADR §3.1): the naive version put the predicate
in the shared _base_reasons gate, which DROPPED the product and regressed the
confuser probe 1->3 -- the disguised-polarity 0001/0003 refuse only because the
coins x coins product DISAGREES with the coins + coins accumulation reading; dropping
it unmasked the additive reading as a unique wrong commit (80/30). The fix is the
downgrade: keep it as a commit-ineligible `exempt` candidate so it still forces the
disagreement. Pinned by test_downgrade_not_removal_preserves_disagreement_refusal.
Evidence (sealed lane; chat/ does not import verify -> serving frozen):
- resolve_pooled over train_sample: 2 correct / 8 wrong / 40 refused (was 2/13/35);
the 5 repeated-unit products (0001/0017/0042/0045/0048) now refuse, 0003/0021 kept.
- confuser probe: wrong unchanged (no 0001/0003 regression), positives still solve.
- serving train_sample 3/47/0 and practice (accumulation + search) 3/47/0
byte-identical; self_verifies/_base_reasons unchanged so search lanes are untouched.
- 171 derivation/pool/verify tests + 40 architectural invariants green.
Honest scope: 13->8, NOT 0. The remaining 8 (0011/0016/0018/0019/0025/0028/0032/0047)
are distinct-unit products in the wrong shape (rate problems) = cue precision
(ADR-0177 CP-2b), the next lever -- not to be faked with a per-case rule. Carries the
corrected ADR-0184 (supersedes the spec-only #484).
* feat(adr-0182): anchor-skip + intra-clause accumulation — distractor 0016 refuses, twin 0017 solves (confuser wrong 1->0)
The last confuser wrong. 0016 ("A train travels 60 miles per hour for 2 hours. Tom
has 8 tickets and buys 4 more tickets. How many tickets?") committed the blunt
product 60x2x8x4=3840 because it was the unique complete reading: the train sentence
is an all-foreign anchor-position block (2 quantities, can't seed an anchor) and the
Tom sentence packs state+change in ONE sentence ("has 8 ... and buys 4 more"), which
the sentence-level reader couldn't decompose -> accumulation_candidates was empty ->
no rival reading -> product committed.
The microscope confirmed intra-clause state+change is a REAL GSM8K pattern (train-0010
"Yun had 20 paperclips initially, but then lost 12"; practice-0121 "Sam has 30 apples
and gives 10 to Anna") -- so this is genuine comprehension, not a 0016-only patch.
Mechanism (added to accumulation_candidates ONLY -> feeds only the pool; train_sample
serving + practice use compose_accumulation, unchanged -> 3/47/0 byte-identical by
construction):
- _sub_clauses: sentence clauses further split on coordinating conjunctions
(and / then / and then). LOCAL to the ungated candidate generator -- the global
segmenter (GB-1/GB-2/serving) is untouched.
- _build_accumulation_anchor_skip: anchor = first single-quantity sub-clause (leading
non-anchorable all-foreign blocks are skipped); chain the conjunction-mate change
("buys 4 more" -> +4). Referent guard + polarity-cue requirement still gate it
(a no-cue sub-clause -> refuse), so GB-2 same-unit lists ("6 apples and 4 apples")
produce no spurious candidate.
Result (sealed lane):
- confuser wrong 1 -> 0. distractor-quantity 0 wrong / 2 refused: 0016 refuses
(product 3840 [complete] vs additive 12 [exempt] disagree). BONUS: the clean twin
0017 ("Tom has 8 tickets and buys 4 more tickets", no distractor) now *solves* 12
-- genuine comprehension of a real positive, the comprehension-vs-surface-match
discrimination the corpus exists to measure. genuine positives 7 -> 8 solved.
- the only non-clean verdict left is 1 spurious (0010 multi-referent "altogether" --
a separate H1 graduation question, not this lever).
- train_sample 3/47/0 and practice 3/47/0 byte-identical; 243 derivation/pool tests
+ 40 architectural invariants green.
Tests:
- test_adr_0182_pool.py TestAnchorSkipIntraClause: intra-clause twin resolves to 12;
the 0016 anchor-skip candidate classifies `exempt`; 0016 refuses via disagreement;
no spurious extra candidate without a conjunction.
- test_adr_0163_f2_confusers.py: baseline wrong 1->0, positives_solved 7->8; renamed
test_distractor_quantity_refuses asserts BOTH 0014 and 0016; new
test_intra_clause_twin_0017_solves.
Stacked on #480 (prior-state guard); merge #480 first. Confuser arc wrong 7->5->2->1->0.
* test(adr-0182): close anchor-skip refuse-branch coverage gap (lookback finding)
The four-PR lookback review (EX-6/pooling/prior-state/anchor-skip) found the
anchor-skip refuse branches in _build_accumulation_anchor_skip untested: the
referent guard (new named actor -> refuse) and the polarity-cue requirement were
asserted but not proven (no test would fail if removed). Per the schema-obligation
discipline, add NON-VACUOUS failing-under-violation tests:
- test_anchor_skip_referent_guard_discriminates_actor: identical structure, change
sub-clause subject is a pronoun ('he', same referent -> 12 IS produced) vs a new
name ('Sara' -> guard suppresses it). Removing _same_referent makes the new-actor
case also produce 12 -> the second assertion fails. (Non-vacuous: the positive
control proves the path is reachable, so the negative isn't trivially-empty.)
- test_anchor_skip_requires_a_polarity_cue: 'gets 4 more' (cue -> 12) vs 'owns 4'
(no cue -> not guessed). Removing the polarity gate makes the no-cue case produce
a reading -> fails.
- test_anchor_skip_refuses_without_single_quantity_anchor: no single-quantity
sub-clause -> () (does not force a multi-quantity clause to anchor).
No code change; behavior unchanged. Confuser wrong stays 0; 281 derivation tests +
40 invariants green.
The wrong=0-critical clause in pool.resolve_pooled (an exempt-only answer
never commits; a complete reading is required to resolve) had no test that
failed when it was removed: the _EXEMPT_ONLY fixture's pool also contained a
complete product (20*3*5=300), so refusal came from the disagreement rule,
not commit-ineligibility. Mutation-disabling the clause left all tests green.
Inject a single-exempt pool directly (the aggressive composers manufacture a
competing complete product for any natural text, so a corpus fixture cannot
isolate the branch). Removing the clause now commits 25 and fails loudly.
Rename the old fixture/test to state honestly that it refuses via disagreement.
The "before/left" reading-rule lever. The microscope showed only the question-time
reading is cleanly achievable: "for N money = spend" is cue-precision-blocked (the
`for` cue is overloaded across train_sample -- `for $2`, `for 14 days`, `for 10 reps`
-- so a spend rule risks regressing train-0021/0003 and is the overfitting trap),
and the disguised-polarity cases already refuse via pooling. "left" is already handled
by loss verbs.
What ships: a question-scope guard. A question asking for a state *before* a stated
change ("How much did Lisa have before lunch?", gold 50) asks for a temporal point
the forward composers do not compute -- they derive the final/net state (50-20=30).
Until a question-time reader exists that is a refusal, never a guess at the wrong
point. target.asks_prior_state detects before/initially/originally/at first/to begin
with/to start with/at the start in the QUESTION CLAUSE only (the last `?`-sentence),
so body narrative does not trip it -- verified safe against train-0003 ("sells before
school starts"), 0010 ("had 20 initially, then lost 12"), 0028. `used to` is excluded
(the purpose infinitive "beads used to make a bracelet" is a false positive).
resolve_pooled refuses when asks_prior_state holds.
Result (sealed lane; chat/ does not import these -> serving 3/47/0 frozen):
- confuser wrong 2 -> 1 (only 0016 distractor-anchor-skip remains). temporal-scope
category cleared (0 wrong / 2 refused). pair-tells 1 -> 0: 0020 ("before lunch")
refuses while its minimal-pair twin 0021 ("left", gold 30) still solves the net --
the discrimination the corpus is built to measure.
- genuine positives still 7 solved, 0 wrong.
- train_sample 3/47/0 and practice 3/47/0 byte-identical.
- 205 derivation/target/pool tests + 40 architectural invariants green.
Tests:
- test_adr_0182_pool.py TestPriorStateQuestionGuard: detector true/false edges
(before-in-question vs before-in-body vs `used to make` false positive vs the
`left` twin) + resolve_pooled refuses the before-question while the left-twin
resolves forward to 30.
- test_adr_0163_f2_confusers.py: baseline wrong 2->1, pair_tells 1->0; new
test_temporal_scope_does_not_misfire + test_before_left_minimal_pair_discriminated.
Stacked on #476 (pooling); merge #476 first. 0016 anchor-skip is the next lever.
Pairs with ADR-0163-F2 graduation amendment (#478).
Implements ADR-0181 PR-6 (eval-plan §4): teachers label or align; they never
define the substrate and never fold embeddings into the versor path.
- sensorium/audio/teachers.py:
- TeacherHint: typed, versioned, checksummed annotation (no raw embeddings).
- AudioTeacher protocol (pure on the signal).
- attach_teacher_hints: the ONLY admission path — appends content.* anchors to
the IR's content_anchors (immutable, recomputes ir_sha256). content.* is not
an operator key, so compile_events skips it: versor + projection_sha256 stay
byte-identical; only the ir leg of the merge_key moves (evidence recorded).
- KNOWN_TEACHER_LANES (whisper/nemo/clap/encodec): declared + gated behind
optional extras; load_teacher import-guards and fails loudly (never a silent
fallback). StubTranscriptTeacher is the deterministic reference instance.
- parser.py: extract _ir_payload + ir_sha256_of (DRY single source of truth for
ir_sha256; byte-identical to parse() output — regression-guarded).
- pyproject.toml: audio-whisper/nemo/clap/encodec optional extras (never
runtime-required).
16 failable proof tests in tests/test_audio_teachers.py. Load-bearing:
test_teacher_hint_does_not_change_versor. Mutation-verified — giving a teacher
anchor an operator event_type (folding it into the versor) fails the
versor-invariance proof; reverted, all pass.
Additive only (ADR-0013): no core layer touched. Audio suite 57/57; eval-gate
ir_sha256 pins unchanged by the parser refactor; architectural invariants 40/40.
Real model adapters are deferred until extras+weights are present; this PR ships
the policy, the typed-hint contract, and the shadow-only guarantee.
Implements ADR-0182's first win on top of EX-6 (#473). The distractor-quantity
confuser 0014 misfired (20x3x5=300) because a blunt product-of-all was the *unique*
self-verifying reading: the completeness clause forces the distractor into it, and
no rival reading existed to trigger the wrong=0 disagreement rule. No tight cue rule
separates it from the legitimate cross-unit products (`for` licenses both the 0014
distractor AND the correct train-0021 product) -- that is the deferred cue-precision
problem. So instead of a reactive patch, let the disagreement rule do the refusing.
Mechanism (sealed lane; chat/ does not import these -> serving 3/47/0 frozen):
- verify.classify_derivation: a derivation is `complete` (commit-eligible),
`exempt` (verified but for an isolated-foreign unused quantity -> commit-
INELIGIBLE), or None. Refactored self_verifies into _base_reasons +
_unused_quantities so the two share logic and cannot drift (behavior identical;
385 derivation tests + smoke 67 green).
- accumulate.accumulation_candidates: exposes the ungated readings, incl. a
distractor-skip reading that drops an isolated-foreign quantity from a multi-
quantity change clause (20+5=25). compose_accumulation is byte-identical
(drop_isolated_foreign=False + the same gate).
- search.multiplicative_candidates / multistep.candidate_chains: ungated candidates.
- pool.resolve_pooled: pools every composer's readings; disagreement -> refuse; a
single answer commits only if a `complete` candidate produced it (exempt-only ->
refuse, so the commit-path completeness guarantee from ADR-0175 is untouched).
- confuser runner: _engine_answer now delegates to resolve_pooled (the prior
first-composer-wins order could not notice it held two incompatible readings).
Result (the microscope):
- confuser wrong 5 -> 2. distractor 0014 refuses (product 300 vs additive 25);
BONUS: both disguised-polarity cases (0001/0003) refuse -- the spurious
"buys X for N coins" product disagrees with the accumulation reading. Remaining
wrong: 0016 (distractor in the anchor clause -> needs anchor-skip, separate step)
and 0020 (temporal-scope). pair-tells 4 -> 1.
- genuine positives still 7 solved, 0 wrong.
- train_sample 3/47/0 and practice 3/47/0 byte-identical (they call
compose_accumulation/search directly -- unchanged -- not the pool).
- smoke 67, architectural invariants 40, lane-SHA freeze 8/8.
Tests:
- test_adr_0182_pool.py: classify (complete/exempt/None incl. narrow-exemption
edges) + resolve_pooled, with the wrong=0 obligation test_exempt_only_never_commits
(a distractor with no multiplicative cue must refuse, not commit 25 -- fails loudly
if the exemption is made commit-eligible).
- test_adr_0163_f2_confusers.py: baseline tightened wrong 5->2, pair_tells 4->1;
new test_distractor_0014_refuses_via_pooling + test_disguised_polarity_does_not_misfire.
Stacked on #473 (EX-6); merge #473 first. 0016 + the remaining wrongs are follow-ups.
The confuser probe's two pseudo-accumulation misfires (0005 ->796, 0007 ->996)
both traced to the same extraction blind spot: a number bonded to its unit by a
hyphen (`25-foot sections`, `20-inch pieces`) was invisible to the base
`number + space + word` pattern, so the self-verification completeness clause
never saw the divisor and the bare `buys ... gives` accumulation read as
"complete". This is the highest-leverage lever the 2026-05-29 session named
("the gate must see the fractions/25-foot it currently misses").
EX-6 adds a tight, ADR-0165-safe lexeme pass: a digit run, a single hyphen, an
alphabetic unit word. The alphabetic-only unit group keeps numeric ranges (`3-5`)
out; taking only the first hyphen segment keeps the postmodifier tail
(`25-year-old`) from inflating the unit — so it stays clear of the deferred EX-3
multi-word-unit traps. Over-extraction here is strictly refuse-preferring: making
the divisor visible drives 0005/0007 to refuse via the polarity-None
`cuts`/`splits` clause, never to a wrong answer.
Evidence (deterministic, the microscope):
- confuser probe: wrong 7 -> 5; pseudo-accumulation 0 wrong / 4 refused;
genuine positives still 7 solved; pair-tells unchanged (4).
- train_sample (capability): 3/47/0 byte-identical.
- practice accumulation: 3/47/0 (wrong=0) byte-identical.
- smoke 67 passed; lane-SHA freeze 8/8 (serving frozen).
Tests:
- TestEX6HyphenatedUnitNumbers pins the new lexeme (value+unit, decimal, no
double-count, word-compound unaffected, numeric range not read as a unit).
- TestProbeBaseline tightened wrong 7->5; new test_pseudo_accumulation_does_not_misfire
is the failing-under-violation obligation (fails loudly if the pass regresses).
- TestSlashFractionLeakHazard pins the deferred `1/4`->`4` denominator leak: not
fixed here because suppressing the leaked operand *removes* a quantity and can
unblock the completeness clause (not unambiguously refuse-preferring), so it
needs its own train_sample + probe validation.
PR-4 of ADR-0181. The acceptance-gate lane that decides whether audio_core_v1's
gate may open. Deterministic synthesis-spec fixtures (no .wav blobs) with
predicted parses, so the gates grade parser semantics as well as determinism.
evals/audio_sensorium/:
- synth.py deterministic fixture synthesis (PCG64 + float32-at-boundary)
- fixtures.json 5 specs: silence, rising-pitch question, falling statement,
noise burst, speech-then-pause
- generate_expected.py reproducible pin generator (uv run -m ...)
- expected_ir.jsonl frozen canonical_sha256 + ir_sha256 + event_type_counts
- expected_projection.json frozen projection_sha256 + reference versor
tests/test_audio_eval_gates.py (12): the gate table per fixture —
shape/dtype, versor_condition<1e-6, within-run replay, canonical-checksum
stability (hard int/cast-stable pin), IR-replay + frozen ir_sha256, semantic
event_type_counts (parser-accuracy gate), and cross-platform versor stability
within atol=1e-6 of the reference (float-safe per eval plan); plus trace
hygiene and gate-closure refusal.
Verified semantics: rise→prosody.rise, fall→prosody.fall, silence→pause.long+
turn.boundary, noise→nonspeech.noise, speech_then_pause→all three.
Cross-platform note: int/quantized-derived hashes are pinned hard; the float
versor is compared within tolerance rather than hash-pinned, since cos/sin/
geometric_product can differ by a ULP across arches. This is the eval-plan's
"equal within declared numeric tolerance" reading — keeps CI stable.
All audio 44 + arch-invariants 40 + smoke 67 green. No core mutation.
Builds the corpus from the ADR-0163-F2 spec: 30 hand-curated, real-sourced cases
across the proven misfire categories (disguised-polarity, pseudo-accumulation/
fractions, multi-referent H1, multi-actor-pronoun ADR-0174, distractor-quantity,
temporal-scope H3, comparative-referent H2, unit-confuser) + genuine-positive
minimal-pair twins. Schema carries category/surface_trap/expected/pair_id/source.
The runner scores OPPOSITE to a coverage lane: the bar is `wrong` -> 0 (a confuser
*answered* is a defect regardless of value) plus pair-consistency (solving a twin
but answering its confuser = a surface-matching tell). It runs the realistic sealed
attempt (accumulation -> multiplicative -> chain, first to resolve).
Honest measured baseline (the probe's whole point — these are the defects the
templated corpus hid): 30 cases -> 7 solved / 15 refused / 7 WRONG / 1 spurious;
4 pair-tells (0001/0003/0014/0020). Wrong by category: disguised-polarity 2
(buys-a-toy-for-30 -> +30), pseudo-accumulation 2 (the 0002 cable/fraction),
distractor-quantity 2, temporal-scope 1 (before-giving -> gave the now-value).
Per the overfitting lesson, the composers are NOT reactively patched to pass the
probe (that is the trap). The baseline is pinned as a no-regression gate (wrong
<= 7, pair-tells <= 4, positives keep solving); future fixes must be GENERAL
mechanisms validated on train_sample, driving wrong down. Sealed: serving 3/47/0
byte-identical (lane-SHA 8/8, claims OK); architectural invariants green.
The first cross-clause comprehension reading: one actor's quantity changes over
successive clauses ("Sam has 14 apples. He buys 9 more." -> 14 + 9). It is the
safe specialisation of the cross-clause sum that GB-3a refuses wholesale (the
Alice/Tom hazard) — we chain only when (same referent) AND (a licensed change
cue of unambiguous polarity), else refuse.
generate/derivation/accumulate.py — compose_accumulation:
- anchor on clause 1's single quantity; apply +M (gain) / -M (loss) per later
change clause, operand taken in the anchor's unit (accumulation is same-dimension);
routed through the unchanged self-verification gate.
- polarity (ordered, so ambiguous "gives" is resolved not guessed): "more" -> gain;
else unambiguous loss verb -> loss; else gives/gave + to/away -> loss; else
unambiguous gain verb -> gain; else REFUSE.
- referent guard (the ADR-0174 multi-actor hazard's defensive fix, built minimally
in the clean lane — NOT the retired gender-blind resolver): a later clause's
subject token must be a pronoun or the anchor's name; a NEW named subject (Tom)
-> refuse. Pronoun gender/number is not matched; a new name is the only signal.
evals/.../accumulation_runner.py — practice scorer: on a base refusal, attempt
compose_accumulation and gold-check (mirrors search_runner). Sealed: fires only
on already-refused cases, never alters serving.
Measured (sealed practice additive lane): 0 -> 55 correct, wrong unchanged at 1
(the base scorer's pre-existing one; accumulation added 55 correct, 0 wrong). The
36 still-refused are multi-change (GB-3b.2) or unrecognised verbs (vocab growth) —
conservative, never wrong.
Proof obligations (tests fail under the violation): new-named-actor refuses (H1),
no/ambiguous change cue refuses, list anchor refuses, multi-change refuses,
determinism. 136 targeted tests + architectural invariants green; serving 3/47/0
byte-identical (lane-SHA 8/8, claims --check OK).
ADR-0180 §1.5.4 + CLAUDE.md work-sequencing item 5 require these four
properties green on main before any core-rs/src/vault.rs change. They are
also the foundation ADR-0181 PR-5 (audio Delta-CRDT wiring) rides on.
T-1 set-equality of vault writes under shuffled ingest (+ idempotent
re-ingest at the content-addressed layer)
T-2 trace-hash invariance to vault order, + recall result-set invariance
to insertion order (the genuinely-failable half)
T-3 versor_apply non-commutativity (negative guard)
T-4 ProjectionHead.project purity across calls and threads
Findings (docs/audit/ADR-0180-t1-t4-findings.md):
- compute_trace_hash folds only vault_hits (a count), NOT vault contents, so
ADR-0180 §1.5.3's "re-sort vault state in content-addressed order" is
currently vacuous at the trace-hash layer; the live order-invariance
obligation is at recall() (result-set + count). Recommend amending §1.5.3.
- equal-score recall ties are index-sensitive; the Merge Kernel needs a
content-addressed tiebreak (mirrors ADR-0181 §2.2 merge key). Recommend
amending §2.2.
- append is genuinely semilattice-eligible; versor_apply is non-commutative.
7 passed; smoke suite green. No runtime/core mutation — tests + audit only.
CP-2a populates the CP-1 ledger from gold-labelled candidate readings and reports
per-pattern reliability — the measurement the cue-precision thesis rests on. Plus
the function-word unit filter, whose value this measurement makes concrete (clean
unit_shape labelling).
What landed (all sealed; serving 3/47/0 byte-identical):
- generate/cue_precision/trainer.py — train_from_cases(cases, enumerators): folds
gold-labelled candidate chains into the ledger via record_case. Decoupled (the
candidate enumerators are injected, so the package still imports nothing from
search). candidates_for dedupes a reading shared by two enumerators.
- generate/derivation/multistep.py — extracted the enumeration half of search_chain
into public candidate_chains(problem_text); search_chain now delegates (verified
byte-identical: ms3 tests + practice counts unchanged). CP-2 needs the readings
the search weighs, not just the one it resolves.
- generate/derivation/extract.py — function-word unit filter (_NON_UNIT_WORDS):
blanks spurious function-word units ($0.75 each -> "", 3/4 of -> "") that
corrupt same-unit detection and unit_shape. Closed lexeme set, ADR-0165-safe.
- evals/gsm8k_math/practice/v1/cue_precision_report.py — trains over 200 sealed
cases (50 train_sample + 150 ADR-0163-F additive) with the real enumerators and
prints the per-pattern reliability table.
- tests/test_adr_0177_cp2a_training.py — trainer obligations (credit/dedupe/
determinism/empty) via synthetic enumerators; real-measurement well-formedness;
search_chain parity.
Load-bearing finding (recorded in ADR-0177): over 200 cases EVERY (cue,op,unit_shape)
pattern floors at ~0.0 reliability (best: for-multiply-cross_unit 0.0116 at 2/34).
The blunt product/sum-of-all readings are almost always wrong vs gold, so the
conservative floor correctly trusts nothing. => CP-2b (trust reliable cues) is
blocked on candidate GENERATION, not the ledger: candidate readings must get less
crude (clause/referent structure, ADR-0178 GB-3b) before any cue earns reliability.
Cue-precision and compositional structure are coupled; structure comes first.
Verification: 107 targeted tests green (CP-2a/CP-1/extract/ms3/GB-1/2/3/MS-1/2) +
architectural invariants; serving CLAIMS.md sha unchanged; practice 4/1/45 and
0/1/149 unchanged. Inert: trains/reports only, consulted by no search/gate.
Track C of docs/handoff/PARALLEL-WORK-PLAN-2026-05-29.md asked for a tight
EX-3 multi-word-unit redo satisfying (a) "12 jumping jacks." -> "jumping
jacks", (b) "6 apples and 4 apples." -> two apples, (c) all GB-1/2/3 tests
green. The cleanest tight rule that satisfies all three —
(?<![\w.])(\d+(?:\.\d+)?)\s+([a-z]+\s+[a-z]+)(?=\s*[.?!,]|\s*$)
— was implemented and passed the four pinned test files. Full-suite
verification then surfaced a second trap the audit at
docs/handoff/AUDIT-ADR-0179-EX-RECONCILE.md did not anticipate:
postmodifier-adjective tails. "25 years old?" fires the tight rule and
produces unit "years old" rather than "years", regressing
test_adr_0176_ms1_question_target.py::TestQuestionQuantities::
test_extracts_quantity_stated_in_question and the "X years old" pattern
in tests/test_adr_0176_ms2_chain.py. The pattern is endemic in GSM8K
(cases 0006 and 0033 both use "X years old"); closing it would need a
second closed lexeme set ({old, tall, long, wide, deep, away, ago, ...})
which the brief judged too open-ended to enumerate responsibly.
Per the brief's escape hatch ("If no rule satisfies all of (a)-(c) without
a grammar template, write a note and ship no code — a refusal is fine")
this commit:
* updates extract.py's module docstring to name BOTH known traps
(connective-crossing AND postmodifier-adjective tails);
* adds tests/test_adr_0179_extract.py::TestEX3StillDeferred with two pins
asserting the postmodifier-adjective shape stays at unit "years" alone,
so no future redo silently re-introduces the regression;
* ships NO extractor code change — the regex remains exactly as on main.
Scope/safety:
* Files touched are within Track C's allowed set (extract.py + its test).
* Zero functional change: extract_quantities byte-identical to main.
* Serving lane untouched (chat/ does not import this module).
* Safe alongside GB-3b on compose.py / clauses.py.
CP-1 of ADR-0177: the per-(cue, op, unit_shape) reliability ledger + credit
assignment mechanism. Mirrors the ADR-0175 per-class ledger discipline
(core/reliability_gate/ledger.py): counts-only integers, reliability via the
pinned conservative_floor, refusals never counted as commitments.
- generate/cue_precision/ledger.py
- CuePattern: (cue, op, unit_shape) key; op in VALID_OPS, unit_shape closed-set.
- pattern_for_step / patterns_in_chain: per-step extraction. unit_shape compares
the operand unit to the model's running (primary/start) unit; a dimensionless
comparative scalar scales within the dimension -> same_unit.
- PatternTally: counts-only (correct/wrong, no refused axis); reliability =
conservative_floor(correct, committed); 0.0 while cold/below N_MIN.
- CuePrecisionLedger: immutable pattern->tally map (canonical sorted tuple);
record_chain / record_case credit candidate chains by gold label, independent
of whether the search resolved or refused.
Inert substrate: not wired into the gate, any scorer, or the search (CP-2/CP-3).
Imported by nothing outside its own tests (asserted by a source-tree scan).
Tests (tests/test_adr_0177_cp1_ledger.py, 27 passing): pattern validation;
unit_shape classification; cold ledger -> 0 reliability; credit assignment;
refusals-not-counted; reliability earned by volume; determinism/replay;
immutability; inertness scan. Smoke suite green (67 passed).
The mandated lookback review before GB-3 (CLAUDE.md §Lookback Review Discipline)
confirmed the audit's hazards H1/H2/H3 were LIVE: compose_sequential summed
same-unit quantities from the whole problem, merging unrelated referents/scopes
and admitting wrong structures whose value happened to ground:
H1 (second actor's apples) -> 6+4+2 = 12
H2 (comparative on other actor) -> (6+4)*2 = 20
H3 (later depletion event) -> 6+4+3 = 13
Root cause is the audit's G1/D2 drift: GB-2a re-extracts from the whole text and
ignores GB-1's clause structure. The fix is the GB-3 increment — make the composer
clause-scoped (consume segment_clauses), refusing when the licensed structure spans
clauses, because this slice cannot model referents:
- quantities must live in exactly ONE clause (0 or >1 -> refuse);
- a comparative outside that clause -> refuse (unmodelled referent binding).
All three hazards now refuse; all 7 GB-2 single-clause structures preserved
(list-sum, three-item, sum-then-scale, and the mixed-unit/disagreement/too-few
refusals). tests/test_adr_0178_gb3_referent_guard.py would fail against the
pre-guard code (12/20/13), so the obligation is proven, not decorative.
Scope/safety:
- compose_sequential is sealed substrate, not wired to a scorer -> serving
byte-identical 3/47/0 (lane-SHA 8/8, generate_claims --check OK); practice
unchanged 4/1/45. No new test failures (2 pre-existing on main).
- ADR-0178 amended: GB-2 relabelled GB-2a (list slice, drift G1 recorded);
GB-3 split into GB-3a (this referent guard, landed) and GB-3b (constructive
cross-clause chaining, next).
Reconciles ChatGPT's four independently-branched extraction PRs (#451/#452/
#453/#454) into one coherent generate/derivation/extract.py. They each rewrote
the same file + same new test off main, so they conflicted pairwise and needed
integration, not a merge.
Integrated (span-tracked, most-specific-pass-first so numbers are never double
counted):
- EX-1 word-numbers (#452): reuses WORD_NUMBERS; tens-one hyphen compounds;
factor-bearing half/third/quarter excluded.
- EX-4 list-unit inheritance (#451): bare numeric list with one trailing unit.
- EX-5 sentence-final numbers (#454): bare final number with empty unit.
Deferred: EX-3 multi-word units (#453). Its greedy lowercase span reads
"6 apples and 4 apples" as unit "apples and", regressing GB-2's
test_same_unit_list_sums, and still can't recover real multi-word units from
0024-class text ("jumping jacks on"). Needs a tighter rule; see
docs/handoff/AUDIT-ADR-0179-EX-RECONCILE.md.
Verification (sealed lane only; chat/ does not import this module):
- Serving frozen: lane-SHA 8/8 match, generate_claims --check OK -> 3/47/0
byte-identical, wrong=0 held.
- Sealed practice improved 4/2/44 -> 4/1/45: case 0025 flips wrong->refused.
EX-1 reads "three", so completeness sees a quantity the 6x50 chain omits and
refuses the spurious 300 (gold 1200) instead of committing it.
- No new test failures (3 pre-existing on main).
Also fixes stale test drift from EX-2 (#447): TestDecimalGroundingGapIsDeferred
asserted decimals still refuse, but #447 made $0.75-class resolve to 864.
Renamed to TestDecimalGroundingResolves and updated to assert the flip.
Honest scope note: EX-4 does NOT unblock real case 0024 (its PR test used a
fabricated bare-list paraphrase). TestRealCase0024StillBlocked pins the true
boundary.
EX-2 — the ONE shared-primitive (serving-path) touch of extraction richness.
_value_grounds already grounds the symbol form $N.NN (currency) and N/M (fraction);
a decimal written WITHOUT a symbol ('0.75') is never a single token (the tokenizer
splits on '.') so it failed to ground — refusing correct products like 0003
(48*24*0.75=864). Now a bare decimal 'N.M' grounds when both digit-runs appear,
symmetric with the $N.NN and N/M branches. Only returns True on a match; non-matching
decimals fall through (refuse).
wrong=0 (the load-bearing obligation, this being a shared/serving-path change):
- serving 3/47/0 BYTE-IDENTICAL (verified).
- round-trip + candidate-graph tests 51/51.
- lane-SHA gate (pinned eval lanes unchanged) — verified before merge.
Payoff: 0003 unblocked in sealed practice (search_chain -> 864 = gold; +1 flip).
Decimal cases that now ground but mis-compose become sealed eliminations (learning
signal); serving untouched. 6 EX-2 tests (decimal grounds/refuses, integer
unchanged, serving byte-identical, 0003 resolves).
GB-2 first increment (ADR-0178). compose_sequential() adds the structure the blunt
MS-3 shapes couldn't reach: a same-unit quantity LIST sums (additive cue), and any
stated comparative scales the sum (sum-then-scale, 0024-family). Op-per-step from
text structure (list => add; comparative => scale); operands are text quantities
(grounded) + comparative steps (cue-grounded) on the flat left-fold — no derived-
intermediate model needed (running value is the intermediate).
Deliberately narrow: same-unit lists only. A stated comparative is ALWAYS applied
(no bare-vs-scaled self-disagreement). A product base over the same list is added
WITHOUT a comparative tail purely as a disagreement-safety candidate -> a same-unit
list that also carries a mult cue (ambiguous) REFUSES. Product-of-all/cross-unit
products stay MS-3's job (avoids the product x comparative blowups a blunt all-bases
composer produced: 0024 -> 4.3M).
Clean-case capability proven: 8 tests (list-sum, sum-then-double/triple, mixed-units
refuse, ambiguous-disagreement refuse, determinism). Honest practice result: 3/2/45
— NO new flips (extraction wall: real cases like 0024 extract non-uniform units
'36 on' so they aren't seen as same-unit lists), 2 sealed eliminations (0037/0039:
list-sum was the wrong structure -> learning signal). Coverage gated by extraction
richness + cue precision, as predicted.
Sealed; serving untouched. Full derivation surface 53/53; ruff clean; smoke 67.
Continuation: richer relational ops (per/each->multiply, more/older->add), branch/
DAG (0033), and the extraction richness (uniform-unit extraction) that unblocks this
on real cases.
GB-1 — first slice of the comprehension-guided composer (ADR-0178). Reads the
problem one clause at a time and derives each clause's LOCAL contribution; GB-2
combines them across clauses.
generate/derivation/clauses.py:
- segment_clauses(text): sentence-level orthographic split (ADR-0165; not grammar).
- clause_local_results(text) -> tuple[ClauseResult]: per clause, 0 quantities =
context (hold), 1 = leaf (its value), >=2 = bounded local search (reuses MS-3
search_chain). Refuse-preferring: ambiguous multi-quantity clause -> unresolved
hold, not guessed.
Locality is the guidance that bounds the search + steers grouping. 9 GB-1 tests
(segmentation, leaf/context/local-product, ambiguous-holds, determinism,
per-clause structure of a multi-sentence problem). Full derivation surface 86/86;
ruff clean; smoke 67. Sealed; not wired into serving (ClauseResults ready for
GB-2 sequential combination).
MS-3 composes MS-1 (Target) + MS-2 (comparative chains + completeness) + the gate.
generate/derivation/multistep.py: search_chain(problem_text, target=None).
Shape-based, NOT blind enumeration: enumerates a small principled candidate set
(product-of-all if a multiplicative cue is present; sum-of-all if an aggregation
hint is present; each optionally + comparative scalars), each using all
quantities, routed through select_self_verified (grounding ∧ cue ∧ unit ∧
completeness ∧ uniqueness). Bounded (MAX_QUANTITIES, refuse-on-overflow) +
deterministic. Target supplies the aggregation cue + question quantities; target-
UNIT matching is deferred (answer_unit=start.unit is wrong for cross-unit products
-> a unit gate would over-refuse; documented).
Honest practice measurement (sealed lane): 4 correct / 9 wrong / 37 refused
(baseline 3/0/47). +1 flip is the unambiguous whole-problem product (0021); the 9
wrongs are product-of-all eliminations on multi-step problems (caught by gold,
the learning signal). Whole-problem shapes add no coverage beyond the unambiguous
product WITHOUT cue precision: when product and sum both self-verify they disagree
-> uniqueness refuses (safe-but-low-coverage by design). The lever remains cue
precision (the ADR-0175 learning loop).
Microscope finding: 0003-class flips (48*24*0.75=864=gold) are blocked by a
DECIMAL/currency grounding gap -- '$0.75' tokenizes to 0/75 so '0.75' is not
grounded by the shared round-trip primitive. Not a search bug; deferred
extraction-richness work (won't casually change the serving round-trip primitive).
A test documents the current refusal so the fix is detectable.
wrong=0: serving untouched (sealed); ambiguity + no-licensed-cue refuse; routes
through the proven gate. 8 MS-3 tests; full derivation surface 77/77; ruff clean;
smoke 67.
MS-2 of multi-step composition. Extends the derivation model so a chain mixes
text-quantity operands and COMPARATIVE-scalar operands (twice->x2, 'N times'->xN,
half->x0.5), self-verifying the whole chain with completeness over body+question
and question-target matching.
- model.py: Step gains comparative flag.
- comparatives.py: ComparativeScalar gains number_token (the '<N> times' number,
so completeness counts the consumed body quantity); comparative_step(cs) bridges
a scalar into a Step (operand grounded by cue, not a text value token).
- verify.py: self_verifies exempts comparative operands from value-grounding
(clause 1) — they are cue-grounded (clause 2); completeness (Counter) counts a
digit comparative's number_token as consuming the body quantity. Adds target_units
to select_self_verified: a chain whose answer_unit isn't the asked unit is dropped
(question-target match; empty target_units imposes no constraint).
Proves the multi-step shapes from the gold structures: 0024 (text sum then 'three
times' scale -> 438), 0033 father-chain (digit-comparative '7 times' + fixed 'half'
+ text add -> 47). Full 0033 DAG (quantity reuse + the question's 25) deferred.
25 MS-2 tests; full derivation surface 69/69 (3a/3b/comparatives/ms1/ms2); ruff
clean; smoke 67. Not wired into serving (model ready for MS-3 target-guided search).
MS-1 of multi-step composition. Turns the question into a Target = what the
problem asks for, the search's pruning signal + stopping criterion (MS-3).
Lexeme-level only (ADR-0165): the existing question parser returns nothing on
these GSM8K questions, and 0165 forbids new question-shape grammar regex. Three
robust signals:
- quantities: numbers stated IN the question (0033's 'when she is 25') via the
body's lexeme extractor — they participate in the derivation.
- aggregation: presence of an aggregation lexeme (total/altogether/combined/sum/
'in all'/'in total') — soft hint the final step is a sum.
- units: asked units resolved by INTERSECTION with the body's known units
(precise lexeme match, e.g. 'jumping'). Superordinates (weight<->pounds) are
NOT faked — deferred to a curated superordinate-units pack; until then the unit
signal is precise-but-incomplete and the search leans on completeness.
Refuse-preferring: empty target field is not an error, just a weaker prune.
generate/derivation/target.py: Target + extract_target(question, known_units=()).
12 MS-1 tests (question-quantity, aggregation, body-unit intersection,
superordinate-not-faked, determinism, frozen). Verified: derivation suite 57/57;
ruff clean; smoke 67. Not wired into serving (Target ready for MS-2/MS-3).
The curated, irreducible world-fact primitives multi-step composition needs
(ADR-0175 section 10: the engine can't derive 'twice = 2' from arithmetic). The
microscope flagged these via the 0015/0025/0024/0033 wrongs.
language_packs/data/en_core_comparatives_v1/: 9 closed-set multiplicative
comparatives (twice/double/triple/quadruple/half/quarter + inflections) -> scalar
ops. manifest.json with sha256 of the bytes on disk (CLAUDE.md pack rule).
Refusal-preferring: non-terminating/ambiguous comparatives (a third, several)
deliberately excluded; expansion via HITL corridor.
generate/derivation/comparatives.py: extract_comparative_scalars() ->
ComparativeScalar(op, scalar, span, cue). Fixed lexemes + the '<number> times'
pattern (digit or word-number via WORD_NUMBERS). Lexeme-level (ADR-0165);
deterministic (text-order); supplies only the SCALAR primitive — referent
binding is the multi-step search's job (ADR-0176).
14 tests incl. refusal-preferring discipline + pack integrity (manifest checksum
matches bytes on disk). Verified: derivation suite 45/45; ruff clean; smoke 67;
packs 141. Not wired into serving (data + extractor ready for ADR-0176 MS phases).
Self-verification strengthening (microscope-driven). The Phase 3b measurement
showed self-verification was necessary-but-not-sufficient: 9/13 self-verified
attempts were wrong. Inspecting them deterministically revealed most were
correct FIRST STEPS of multi-step problems that ignored numbers stated elsewhere.
Adds clause 5 to self_verifies: a derivation must account for every quantity the
problem states (problem quantities subset of used). Refuse-preferring: unused
quantities -> not self-verified. This catches the multi-step-incomplete attempts
the grounding/cue/unit clauses cannot (their operands ARE grounded).
Practice measurement: wrongs 9 -> 2 (4 correct / 2 wrong / 44 refused). The 2
survivors (0015, 0025) are COMPLETE but wrong due to missed WORD-quantities
('twice', 'her friends') -> the microscope points the next change at extraction.
Updated the disagreement test to use two complete derivations; added an
incomplete-refusal test. 32 tests pass; smoke green; serving untouched (sealed).
ADR-0175 Phase 3b — the first live attempt generator. Runs only in the sealed
practice lane, only on cases the engine refused; every proposal is gated by the
Phase 3a self-verification gate.
generate/derivation/:
- extract.py: extract_quantities() — lexeme-level (number + unit word; ADR-0165).
- search.py: search_multiplicative() — one in-clause product candidate per
sentence with >=2 quantities + a present multiplicative cue; gated by
select_self_verified. Per-sentence scope + multi-candidate disagreement give
the uniqueness gate real teeth (two qualifying sentences -> refuse). The cue
set {each,every,for,per,times} is an explicit PROVISIONAL hypothesis the
practice loop refines, not a claimed-correct grammar.
evals/gsm8k_math/practice/v1/search_runner.py: search_augmented_scorer +
build_search_report — base scorer, then a practice-only attempt on refusals.
MEASUREMENT (the deliverable, per the breadth-of-impact test):
practice with search: correct=4 wrong=9 refused=37 (baseline 3/0/47)
- Flips +1 (0021, the clean in-clause aggregate) and its renumbered/reworded
variants (ADR-0114a perturbation guard) -> a real capability, not memorisation.
- 9 wrong attempts -> elimination records (§9), the learning signal. The naive
full-product cue model over-attempts; the eliminations are exactly the signal
that refines it.
HONEST FINDING: self-verification (grounding ∧ cue ∧ unit ∧ uniqueness) is
NECESSARY but NOT SUFFICIENT — 9/13 self-verified attempts were wrong vs gold.
The gap is cue PRECISION / which-quantities-compose (the knowledge axis), not
'can we multiply' (skill). This is why the search runs sealed: gold catches the
9, and case 0050 (canary) attempted-and-failed IN PRACTICE without touching
serving -> validates the seal.
Invariants: #1 seal (serving still 3/47/0; 0050 refuses in serving; no
generate/chat import of the lane), #3 determinism. Serving wrong=0 untouched.
Verified: 3a+3b 31/31; ruff clean; serving lane 4/4; smoke 67/67.
ADR-0175 Phase 3 splits wrong=0-first: build the gate (3a) and PROVE invariant #2
before the bounded search (3b) that could exploit gaps.
generate/derivation/:
- model.py: Quantity / Step / GroundedDerivation. A derivation is a left-fold over
text-sourced quantities; each Step carries its licensing cue (the lexeme the
search claims licenses the op).
- verify.py: self_verifies() — grounded operands ∧ grounded operation cues ∧ unit
consistency ∧ no divide-by-zero. Grounding REUSES the canonical primitives from
math_roundtrip (_tokens/_token_in/_value_grounds) so the gate cannot drift from
the round-trip contract. select_self_verified() adds the uniqueness rule:
unique self-verifying answer resolves; zero or disagreeing refuse (wrong=0).
INVARIANT #2 proven (TestInvariant2_NoSpuriousSelfVerification): the gate refuses
to self-verify a derivation that is not grounded+unit-consistent+unique even when
its value coincides with gold — the 20/5==4 class:
- invented operand not in text -> refused
- operation cue not in text -> refused (division not licensed by any present cue)
- value coincidence (20/5=4) with ungrounded op -> still refused
- add across units (pounds + reps) -> refused
- divide-by-zero -> refused
Plus uniqueness: disagreeing grounded derivations -> refuse; agreeing -> resolve.
Phase 3a is inert (nothing wires generate.derivation into serving). 3b is the
bounded search that produces derivations for this gate + measures the flip-curve
in the practice lane under perturbation.
Verified: 16/16; ruff clean; smoke 67/67; no serving import.
ADR-0175 Phase 2 — a NEW lane (evals/gsm8k_math/practice/v1/), separate from the
wrong=0-pinned serving runner which is NOT modified. Runs the 50 cases in
practice mode: scores correct/wrong/refused as practice metrics, feeds per-class
counts into the Phase 1 ledger, diagnoses every refusal (§8), emits an
elimination record per wrong.
- classify_operation: gold-derived primary op class {multiplicative,divisive,
additive} from <<a*b=c>> calc annotations (Tier-1 checkable in practice).
- diagnose_refusal (§8): skill_gap / knowledge_gap / genuine_ambiguity router.
- EliminationRecord (§9): wrong attempt gold caught -> pruning signal.
- PracticeReport: counts + per-class ledger + diagnoses + eliminations; as_dict.
- run_practice(cases, scorer=...): injectable scorer for tests; defaults to the
candidate-graph scorer (read-only — never alters serving).
Live result mirrors serving (3 correct / 0 wrong / 47 refused of 50) because the
engine still refuses rather than guesses — attempts/eliminations go live in
Phase 3. But the diagnosis is already actionable: 35 skill_gap / 12 knowledge_gap
/ 0 genuine_ambiguity — 74% of refusals are skill gaps (Phase 3's search target),
quantifying the skill-vs-knowledge split.
Invariants: #1 seal (serving still 3/47/0; no generate/chat import of the lane),
#3 determinism (report byte-identical across runs). Elimination + wrong-tolerance
paths unit-tested via injected scorer (no live wrongs yet).
Verified: Phase 1+2 53/53, serving train_sample tests 4/4 (seal), smoke 67/67,
ruff clean.
ADR-0175 Phase 1 — standalone, deterministic, zero serving change. Nothing in
the serving/eval path imports it.
core/reliability_gate/:
- floor.py: conservative_floor(s,k) — pinned one-sided Wilson lower bound over
COMMITTED trials. z=2.576, N_MIN=10; range [0,1) (never exactly 1.0); float64
rounded half-to-even to 1e-9 for cross-backend replay. Perfect record reduces
to k/(k+z²) (earned by volume).
- ledger.py: ClassTally — immutable per-class counts; reliability = commitment
precision (refusals excluded so coverage never penalizes reliability);
t2_precision over the anchor set; coverage tracked separately.
- ceilings.py: Action{PRACTICE,PROPOSE,SERVE} + Ceilings — human-set θ
(practice=0, propose=.85, serve=.99). Frozen; with_override returns a NEW
instance (no in-place self-authorization).
- gate.py: license_for() — deterministic gate, measured/required≥1 (≡ measured≥
required; required=0 ⟹ always). Pure; never mutates/emits ceilings.
34 tests, each ADR invariant exercised by a test that fails under its violation:
#3 determinism/replay (idempotent, pre-rounded, deterministic decisions),
#4 no self-authorization (frozen ceilings; gate never emits/mutates them),
#1 proxy (zero serving coupling). Plus the §4a worked examples (38 clean
commitments clear propose; one wrong in 40 drops below; serve needs ~657).
Verified: 34/34 pass; architectural invariants 40/40; smoke 67/67; ruff clean;
no serving/eval import of the package.
The recognizer/candidate-graph path is the single canonical reader.
Retires the flag-gated incremental-reader dispatch that admitted 0/50 on
train_sample and only added a dead fall-through:
- remove _try_comprehension_reader, _try_reader_for_question, _tokenize_sentence
and both dispatch blocks from generate/math_candidate_graph.py
- delete generate/comprehension/lifecycle_runtime_adapter.py (402 LOC,
used only by the question-reader dispatch)
- drop the comprehension_reader_questions config flag and the parse_and_solve
/ _score_one_candidate_graph config threading
- remove the --use-reader runner plumbing + flag-ON/OFF delta report from
the train_sample runner; refresh report.json (drops stale use_reader field
and a stale refusal-reason; verdicts unchanged at 3/47/0)
- remove the now-dead use_reader field from teaching/coverage.py
CoverageReport + the core teaching coverage CLI flag
- delete tests/test_reader_coexistence.py (flag-ON/OFF premise dissolved);
fix 3 ADR-0174 build_report calls and 2 subprocess invocations
lifecycle.py and audit.py are KEPT — they are load-bearing for the ADR-0172
math-contemplation teaching corridor (audit_problem -> teaching/math_*),
which a pre-deletion trace surfaced. The parent ADR's plan to delete
lifecycle.py was wrong; only its GSM8K scoring dispatch was inert.
Net -1,038 LOC (code + tests). Behavior-preserving:
- train_sample 3/47/0, byte-identical verdicts to pre-5a baseline
- determinism holds; smoke/packs/runtime/cognition/teaching lanes green
- contemplation corridor + lifecycle/audit tests pass
Pre-existing (NOT introduced here; reproduce on base with changes stashed):
5 out-of-curated-lane stale committed-artifact / stale-assertion failures
(test_math_evidence_e2e, test_adr_0126_runner_wiring, G3/coverage_probe
report-match, test_refusal_taxonomy_lane rebuild).
ADR-0174 Phase 3b — emit N anchors for compound-clause discrete-count
sentences sharing one subject + one verb. Architectural substrate;
score on train_sample preserved at 3/47/0 (compound cases like 0027
admit past the recognizer-injection refusal but the rest of the
problem still has downstream complexity — fractions, percent — that
needs Phase 4 + solver work).
generate/comprehension/state.py:
HYPOTHESIS_CAP raised 4 → 8. Case 0040 emits 5 anchors; cap=8
gives headroom (7-item lists) without becoming permissive.
generate/recognizer_match.py:
_try_extract_compound_discrete_count_anchors() — new extractor
emitting tuple of anchors for compound sentences. Refusal-
preferring on:
- no conjunctive separator (single-anchor path)
- multiplicative/percent/fraction markers
- head verb not in whitelist
- any tail clause without grounded (count, observed_noun) pair
- exceeding HYPOTHESIS_CAP
- unaccounted digit in tail (wrong=0 hazard defense surfaced by
2026-05-28 implementation review: bogusnoun would silently fail
to produce anchor while leaving the digit unaccounted, admitting
partial state)
Wired into _match_discrete_count_statement dispatch as fallback when
single-anchor extraction fails.
tests/test_adr_0174_phase3b_compound_clause.py:
11 acceptance tests passing — pure conjunctive lists (proper-noun
+ pronoun-subject + single-actor antecedent), refusal-preferring
discipline (mixed-verb, multiplicative-tail, non-whitelisted-head,
partial-grounding all-or-nothing), HYPOTHESIS_CAP enforcement,
multi-actor pronoun defense preserved on compound, wrong=0 +
case-0050 canary.
tests/test_adr_0174_phase1_held_hypothesis_state.py:
Updated test_hypothesis_cap_is_four → test_hypothesis_cap_is_eight
with rationale for the raise.
Phase 3b implementation lookback review (per CLAUDE.md doctrine):
- Surfaced silent-partial-admission hazard in tail extraction;
fixed with digit-accounting check before commit
- Surfaced LATENT regex-path multi-actor pronoun hazard (not
introduced by Phase 3b; documented in test docstring with
cross-reference to project-adr-0174-multi-actor-pronoun-hazard
memory for follow-up)
- case 0040 ('He now has...') remains refused — 'now' adverb between
subject and verb defeats the existing canonical regex. Adverb-
stripping is separate scope (not Phase 3b).
Acceptance:
- 258/258 ADR-0174 + math_problem_graph tests pass
- Smoke 67/67, packs 141/141
- train_sample 3/47/0 preserved (wrong=0 held)
- Case 0027 'Malcolm has 240 followers on Instagram and 500 followers
on Facebook' now admits via the compound extractor — verified by
refusal moving to the next sentence (which has 'half' fraction)
All findings from the 2026-05-28 Phase 1-3a lookback review addressed
in one commit on the Phase 3a branch:
Wrong=0 hazard defense (the load-bearing fix):
- generate/math_candidate_graph.py: Phase 3a wiring now collects the
set of distinct proper-noun subjects seen in prior context. When
more than one exists, refuses with no_antecedent_ambiguous trace
event rather than guessing the most-recent (which was gender-blind
single-binding — wrong attribution in multi-actor problems).
- Refusals from the statement loop now preserve _statement_trace via
reader_trace in CandidateGraphResult (pre-existing latent issue:
Phase 2/3 trace events were dropped on early statement refusal).
- New tests assert: ambiguous case refuses with correct trace; single-
actor case still resolves normally.
Test coverage backfills (closes the 13 untested predicate-name gaps):
- TestCheckConstraintsInitialPredicateNames — 3 tests asserting the
exact predicate name on initial.value_grounds / initial.unit_grounds
/ initial.entity_grounds failure paths.
- TestCheckConstraintsOperationPredicateNames — 3 tests asserting
operation.verb_grounds / operation.value_grounds / operation.unit_grounds
failure-predicate-name parity.
- TestCheckConstraintsComposedInitialPath — 4 tests for the RAT-1
composed_initial path which was entirely untested in Phase 2
(parity manually verified during lookback review; now automated).
ADR amendment (honest doc vs impl drift):
- docs/decisions/ADR-0174-held-hypothesis-comprehension.md: appended
'Implementation Notes' section documenting:
- reevaluate signature differs from spec text (shipped is more
composable; treat as amended)
- Phase 2 wires per-candidate, not per-token (per-token is Phase 5)
- Lookback recompute is candidate-level, not token-level
- Hypothesis.constraint_state is never populated by Phase 2
- Multi-actor pronoun hazard defense rationale
- Honest LOC accounting: Phases 1-3a net +1,500 lines (Phase 5
delivers the projected net removal)
- Test coverage backfill summary
Cosmetic:
- lookback.py:297 unreachable raise — added # type: ignore[unreachable]
with comment explaining defensive future-proofing for Phase 3b.
Acceptance verified:
- 124/124 Phase 1+2+3a + reader tests pass (was 95/95 before backfills)
- Smoke 67/67, packs 141/141
- train_sample 3/47/0 preserved (wrong=0 invariant held)
- Multi-actor hazard live-tested: parse_and_solve refuses the
Alice/Bob/She case with no_antecedent_ambiguous trace event
See CLAUDE.md §Lookback Review Discipline and memory
feedback-lookback-review-discipline for the doctrine that surfaced
all of these issues at the right time.
