core-labs/core - Forgejo: Beyond coding. We forge.

core-labs/core

Fork 0

Commit graph

Author	SHA1	Message	Date
Shay	5a78b0e37b	feat(register): ADR-0077 — substantive register knobs + layering boundary (R6) R5 (ADR-0072) shipped the register machinery; ADR-0074's orthogonality tour proved the axis was decoratively orthogonal to anchor-lens but inspection of the cognition-eval surfaces revealed two structural gaps: * On pack-grounded DEFINITION/RECALL/COMPARISON composers, the only realizer override any register consumed was `disclosure_domain_count` — which only fires on the no-gloss disclosure path. Under terse_v1, every gloss-DEFINITION cell was byte-identical to default_neutral_v1. * The register-tour's `surfaces_vary_at_least_once` gate could be satisfied by convivial's decorative wrapper alone, masking that regression in CI. R6 closes both: Layering separation (the load-bearing fix): * New TurnEvent/ChatResponse field `register_canonical_surface` carries the composer output BEFORE any register transformation. The pipeline hashes this field for `trace_hash`, preserving R5's invariant that per-prompt trace_hash is CONSTANT across registers even while substantive transforms produce visibly different surfaces. Substantive transforms (`chat/register_substantive.py`): * terse_v1 gains 3 bool knobs: `drop_provenance_tag`, `compress_gloss`, `drop_articles` — all pure regex transforms on the canonical surface. * convivial_v1 gains `append_semantic_domain_clause` — appends a single bounded "Related: <atom>." clause using the lemma's pack atoms. * default_neutral_v1 leaves overrides empty; substantive transform is byte-identical no-op (preserves `byte_identity_null_lift`). * C1 (ADR-0075) safety preserved: drop_articles refuses to drop articles following `not` (avoids R3 violations); no knob combination trips R2/R3. Strengthened tour gate (`evals/register_tour/run_tour.py`): * Replaces `surfaces_vary_at_least_once` with two falsifiable claims: - `terse_substantively_differs_from_neutral_on_pack_grounded_definition` - `convivial_substantively_differs_from_neutral_on_pack_grounded_definition` Both restrict to DEFINITION+pack-grounded cells and require difference beyond whitespace/punctuation. * New claim `register_canonical_surfaces_identical` directly proves the layering separation. * Preserves R5's `all_grounding_sources_identical` + `all_trace_hashes_identical`. Pack ratification: * Loader widened to accept `bool` for closed-set R6 keys (drop_provenance_tag / compress_gloss / drop_articles / append_semantic_domain_clause). * `_KNOWN_OVERRIDE_KEYS` ratify gate extended with same. * terse_v1 + convivial_v1 reratified with new knobs; companion mastery reports re-sealed. default_neutral_v1 unchanged. Invariants pinned: * `invariant_register_canonical_surface_constant_across_registers` (new) * `invariant_terse_substantively_distinct_from_neutral` (new) * `invariant_convivial_substantively_distinct_from_neutral` (new) * `invariant_realizer_no_illegal_articulation` (C1, preserved) * `invariant_realizer_guard_byte_identity_on_currently_passing_cases` (C1, preserved) Verification: * `core eval cognition`: 100.0% / 91.7% / 100.0% / 100.0% — byte- identical under default_neutral_v1. * `core demo register-tour`: all 5 claims green, exit 0. * `core demo anchor-lens-tour`: green (no anchor-lens code touched). * `core demo orthogonality-tour`: green (5/5 claims). * Full lane: 2858 passed, 1 pre-existing failure (test_all_preamble_explains_combined_run, carried forward unchanged from main). 56 new R6 tests across three files.	2026-05-19 23:39:11 -07:00
Shay	6207b5fd0e	feat(register): R1–R4 register pack subsystem — deterministic surface variation Introduces the presentation axis as a fourth pack class (sibling to identity / safety / ethics), orthogonal to the truth path. Same input + same packs + same register ⇒ bit-for-bit reproducible surface; varying any of the three ⇒ genuinely different output. No stochastic sampling. ADR-0068 (R1): RegisterPack frozen dataclass, loader, ratify script, seam test. - default_neutral_v1 ratified as null register. ADR-0069 (R2): realizer register parameter threaded through 9 composer entry points; RuntimeConfig.register_pack_id; three byte-identity invariants (A: None ≡ pre-R2 unregistered; B: None ≡ default_neutral_v1; C: trace_hash invariant under register). Amended to default-with-lint after 167-call-site scout: composers default to UNREGISTERED, AST lint enforces explicit register= at runtime call sites. ADR-0070 (R3): terse_v1 register, first non-neutral pack. realizer_overrides schema with known-keys allow-list (disclosure_domain_count ∈ {1,2,3}). build_pack_surface_candidate reads override with fail-soft clamp. New invariant register_invariant_grounding asserts grounding_source + trace_hash byte-identical across {None, neutral, terse}. ADR-0071 (R4): seeded surface variation via convivial_v1. chat/register_variation.py applies SHA-256-seeded marker selection from bounded discourse-marker buckets. ChatResponse.pre_decoration_surface routes truth-path surface to core/cognition/pipeline.py so trace_hash stays invariant under register (the load-bearing architectural fix — initially invariant C failed under convivial because decoration was leaking into trace_hash via response.surface). Empty-string marker entries now legitimate ("no marker this turn" is a valid seed pick). realizer_overrides schema widened with per_intent nested block (validated against IntentTag whitelist; wired but not exercised by convivial). Two new invariants: seeded_variation_replay_equivalence (fresh runtimes → byte-identical) and seeded_variation_turn_distinct (same prompt across turns → ≥2 distinct surfaces). ADR-0072 (R5, draft): telemetry + operator surface — TurnEvent gains register_id and register_variant_id, core chat --register flag, core demo register-tour. Status: Proposed; not yet implemented. Three ratified register packs ship: default_neutral_v1 (null), terse_v1 (disclosure_domain_count=1), convivial_v1 (3 openings × 3 closings). Verification: - 84 register tests pass + 1 documented skip - Curated lanes green: smoke 67, cognition 120+1s, teaching 17, packs 6, runtime 19, algebra 132 - Cognition eval byte-identical to pre-register baseline: public 100/100/91.7/100, holdout 100/100/83.3/100 - Full lane: 2608 passed, 4 skipped, 1 failed (pre-existing test_cli_demo.py "Combined Demo" → "Run Every Demo" rename, unrelated) Truth-path isolation: chat/register_variation.py is realizer-side; the seam test (tests/test_register_pack_seam.py) refuses imports of packs.register from intent classification, propagation, vault recall, trace hashing, and algebra.	2026-05-19 16:52:36 -07:00

Author

SHA1

Message

Date

Shay

5a78b0e37b

feat(register): ADR-0077 — substantive register knobs + layering boundary (R6)

R5 (ADR-0072) shipped the register *machinery*; ADR-0074's orthogonality
tour proved the axis was decoratively orthogonal to anchor-lens but
inspection of the cognition-eval surfaces revealed two structural gaps:

* On pack-grounded DEFINITION/RECALL/COMPARISON composers, the only
  realizer override any register consumed was `disclosure_domain_count`
  — which only fires on the no-gloss disclosure path.  Under terse_v1,
  every gloss-DEFINITION cell was byte-identical to default_neutral_v1.
* The register-tour's `surfaces_vary_at_least_once` gate could be
  satisfied by convivial's decorative wrapper alone, masking that
  regression in CI.

R6 closes both:

Layering separation (the load-bearing fix):
* New TurnEvent/ChatResponse field `register_canonical_surface` carries
  the composer output BEFORE any register transformation.  The pipeline
  hashes this field for `trace_hash`, preserving R5's invariant that
  per-prompt trace_hash is CONSTANT across registers even while
  substantive transforms produce visibly different surfaces.

Substantive transforms (`chat/register_substantive.py`):
* terse_v1 gains 3 bool knobs: `drop_provenance_tag`, `compress_gloss`,
  `drop_articles` — all pure regex transforms on the canonical surface.
* convivial_v1 gains `append_semantic_domain_clause` — appends a single
  bounded "Related: <atom>." clause using the lemma's pack atoms.
* default_neutral_v1 leaves overrides empty; substantive transform is
  byte-identical no-op (preserves `byte_identity_null_lift`).
* C1 (ADR-0075) safety preserved: drop_articles refuses to drop
  articles following `not` (avoids R3 violations); no knob combination
  trips R2/R3.

Strengthened tour gate (`evals/register_tour/run_tour.py`):
* Replaces `surfaces_vary_at_least_once` with two falsifiable claims:
  - `terse_substantively_differs_from_neutral_on_pack_grounded_definition`
  - `convivial_substantively_differs_from_neutral_on_pack_grounded_definition`
  Both restrict to DEFINITION+pack-grounded cells and require
  difference beyond whitespace/punctuation.
* New claim `register_canonical_surfaces_identical` directly proves
  the layering separation.
* Preserves R5's `all_grounding_sources_identical` +
  `all_trace_hashes_identical`.

Pack ratification:
* Loader widened to accept `bool` for closed-set R6 keys
  (drop_provenance_tag / compress_gloss / drop_articles /
  append_semantic_domain_clause).
* `_KNOWN_OVERRIDE_KEYS` ratify gate extended with same.
* terse_v1 + convivial_v1 reratified with new knobs; companion
  mastery reports re-sealed.  default_neutral_v1 unchanged.

Invariants pinned:
* `invariant_register_canonical_surface_constant_across_registers` (new)
* `invariant_terse_substantively_distinct_from_neutral` (new)
* `invariant_convivial_substantively_distinct_from_neutral` (new)
* `invariant_realizer_no_illegal_articulation` (C1, preserved)
* `invariant_realizer_guard_byte_identity_on_currently_passing_cases`
  (C1, preserved)

Verification:
* `core eval cognition`: 100.0% / 91.7% / 100.0% / 100.0% — byte-
  identical under default_neutral_v1.
* `core demo register-tour`: all 5 claims green, exit 0.
* `core demo anchor-lens-tour`: green (no anchor-lens code touched).
* `core demo orthogonality-tour`: green (5/5 claims).
* Full lane: 2858 passed, 1 pre-existing failure
  (test_all_preamble_explains_combined_run, carried forward
  unchanged from main).  56 new R6 tests across three files.

2026-05-19 23:39:11 -07:00

Shay

6207b5fd0e

feat(register): R1–R4 register pack subsystem — deterministic surface variation

Introduces the presentation axis as a fourth pack class (sibling to identity /
safety / ethics), orthogonal to the truth path. Same input + same packs +
same register ⇒ bit-for-bit reproducible surface; varying any of the three ⇒
genuinely different output. No stochastic sampling.

ADR-0068 (R1): RegisterPack frozen dataclass, loader, ratify script, seam test.
  - default_neutral_v1 ratified as null register.

ADR-0069 (R2): realizer register parameter threaded through 9 composer entry
  points; RuntimeConfig.register_pack_id; three byte-identity invariants
  (A: None ≡ pre-R2 unregistered; B: None ≡ default_neutral_v1; C: trace_hash
  invariant under register). Amended to default-with-lint after 167-call-site
  scout: composers default to UNREGISTERED, AST lint enforces explicit
  register= at runtime call sites.

ADR-0070 (R3): terse_v1 register, first non-neutral pack. realizer_overrides
  schema with known-keys allow-list (disclosure_domain_count ∈ {1,2,3}).
  build_pack_surface_candidate reads override with fail-soft clamp. New
  invariant register_invariant_grounding asserts grounding_source +
  trace_hash byte-identical across {None, neutral, terse}.

ADR-0071 (R4): seeded surface variation via convivial_v1.
  chat/register_variation.py applies SHA-256-seeded marker selection from
  bounded discourse-marker buckets. ChatResponse.pre_decoration_surface routes
  truth-path surface to core/cognition/pipeline.py so trace_hash stays
  invariant under register (the load-bearing architectural fix — initially
  invariant C failed under convivial because decoration was leaking into
  trace_hash via response.surface). Empty-string marker entries now
  legitimate ("no marker this turn" is a valid seed pick). realizer_overrides
  schema widened with per_intent nested block (validated against IntentTag
  whitelist; wired but not exercised by convivial). Two new invariants:
  seeded_variation_replay_equivalence (fresh runtimes → byte-identical) and
  seeded_variation_turn_distinct (same prompt across turns → ≥2 distinct
  surfaces).

ADR-0072 (R5, draft): telemetry + operator surface — TurnEvent gains
  register_id and register_variant_id, core chat --register flag, core demo
  register-tour. Status: Proposed; not yet implemented.

Three ratified register packs ship: default_neutral_v1 (null), terse_v1
(disclosure_domain_count=1), convivial_v1 (3 openings × 3 closings).

Verification:
  - 84 register tests pass + 1 documented skip
  - Curated lanes green: smoke 67, cognition 120+1s, teaching 17, packs 6,
    runtime 19, algebra 132
  - Cognition eval byte-identical to pre-register baseline:
    public 100/100/91.7/100, holdout 100/100/83.3/100
  - Full lane: 2608 passed, 4 skipped, 1 failed (pre-existing
    test_cli_demo.py "Combined Demo" → "Run Every Demo" rename, unrelated)

Truth-path isolation: chat/register_variation.py is realizer-side; the seam
test (tests/test_register_pack_seam.py) refuses imports of packs.register
from intent classification, propagation, vault recall, trace hashing, and
algebra.

2026-05-19 16:52:36 -07:00

2 commits