[noahgift]

Executes the operator ruling of 2026-09-20 on #3430 (clauses 1–6, amendment 2'/2a/2b). #3452's plan is ruled; this is M1's leaf. Behaviour is unchanged — no consumer imports it yet.

What lands

crates/aprender-quant/src/ggml_type.rs	GgmlType (35 live variants at upstream ids), QuantTraits, TRAITS[43] dense and indexed by id, GgmlFamily, GgmlTypeError, and the complete API the three enums it will replace carry today
scripts/extract_ggml_traits.sh + .py	the committed extraction command (clause 3)
crates/aprender-quant/fixtures/ggml_traits.json	the fixture, recording the sha it was read from
contracts/ggml-type-v1.yaml	2 equations, 5 obligations, 5 falsifiers, 1 kani harness — pv validate clean, pv audit no findings
scripts/extract_ggml_traits_selftest.sh	the extractor's 7-case table, synthetic inputs, no llama.cpp checkout needed

Clause 2': the oracle is named by reference, never as a sha

The extractor reads build_commit from scripts/llama_pin.toml through llama_pin_get — the repo's one reader of that file — and refuses a checkout that is not at that commit (exit 3) or is dirty (exit 4). Today it resolves to d1d3c3396; the ruling's 3173a5647 is struck by amendment 2' and appears nowhere here.

Clause 3: why a probe, not a parse

ggml's table says .type_size = sizeof(block_q4_1) — a C expression. Every in-tree table that got these numbers wrong got them wrong by hand. So the extractor links the pinned checkout's own libggml-base and calls ggml_type_size / ggml_blck_size / ggml_type_name / ggml_is_quantized.

It still parses the header, as a cross-check that can fail: a row is Removed because its line is commented, never because the comment says so (// GGML_TYPE_Q4_0_4_8 = 32, says nothing — the pass that read wording returned 44/41/3 instead of 43/35/8 during the ruling), and that must agree with the compiled table's blck_size == 0 or nothing is emitted (exit 7).

Independently reproduces clause 2b: 43 slots, 35 live, 8 removed.

Clause 3's mutation proof — all RED, every control GREEN

mutation	what turned red
TRAITS[9].type_size 36 → 40 (gguf-py's Q8_1, refuted by ggml's static_assert)	traits_table_equals_the_upstream_fixture_row_for_row, computes_fifteen_sizes_are_unchanged
TRAITS[30].name BF16 → Bf16	serves_sixteen_strings_still_mean_what_they_meant + 1
TRAITS[4].family Removed → Affine	the row comparison and the is_quantized cross-check, independently
a fake id 43 in the fixture	3 tests
fixture pin → 39173bcac (the superseded comparator)	falsify_ggml_001, naming both shas and the regeneration command
a row deleted from the fixture	falsify_ggml_002 ("the fixture lists 42 of 43 ids")
(control)	8 + 3 GREEN

Clause 2a: a pin bump turns this RED in the bump's own PR

falsify_ggml_001 asserts fixture.pin_build_commit == llama_pin.toml build_commit. It lives in aprender-core/tests/monorepo_invariants.rs, not in aprender-quant, because trueno_quant is published and a packaged crate has no scripts/llama_pin.toml — a test there could only skip, and a skip is how this becomes decoration. That target is already CI-wired; the new quant target is wired by a ci/explicit-test-commands.d/ fragment, so no workflow file is touched.

Found while extracting — NOT fixed here (clause 4 input for #3431)

• aprender-core::ggml_dtype_element_size is wrong in 10 of 31 rows. From index 16 the table follows the order of its own comment (… I8, I16, BF16, I32, I64, F64, IQ1_M) instead of ggml id order (24 I8, 25 I16, 26 I32, 27 I64, 28 F64, 29 IQ1_M, 30 BF16), and the IQ rows carry values belonging to no type. Measured consequence: BF16 read 0.375 B/elem instead of 2.0, a 5.3× under-estimate, so a truncated BF16 GGUF passed the apr tensors reports a declared tensor table that does not fit the file, while apr validate rejects it #2569 extent check (fail-open) and apr tensors printed a size 5.3× too small. PMAT-869 corrected four K-quant rows of this same table in 2026-06 and stopped.
• GgufTensor::byte_size sizes Q4_1 at 18 where upstream (and core's own shape.rs) say 20 — two tables in one crate disagreeing. It has no non-test caller, so no GGUF was written wrong by it; clause 5's "refusal on load" has no load path to attach to. Stated rather than silently skipped.

Both fixes land next, as lookups into TRAITS — not a fourth hand-typed table. That is also why they are not in this PR: doing both at once would make the leaf's "changes no behaviour" claim unverifiable.

Verification

cargo test -p aprender-quant                 19 lib + 8 fixture + 1 doc, all pass
cargo test -p aprender-core --test monorepo_invariants   15 pass (3 new)
bash scripts/extract_ggml_traits_selftest.sh 7 cases, 0 failures
cargo clippy -p aprender-quant --all-targets -- -D warnings   clean
cargo fmt --all -- --check · cargo deny check advisories      clean
pv validate contracts/ggml-type-v1.yaml      0 errors · pv audit: no findings
cargo test -p aprender-contracts --lib       1665 pass (corpus lint, sigma gate included)
bash scripts/check_explicit_test_commands.sh · check_shell_lint_ratchet.sh · check_include_files.sh   PASS
cargo package -p aprender-quant --list       fixture + module + test all packaged

Receipt with the full measurement table and every deviation from plan v4: docs/audits/impl-PMAT-3430-m1-leaf-receipt.md.

Next (this ticket): the two table fixes above, then Phases 3–4 (the three enums become re-exports, each crate keeps its admitted id set exactly, check_one_ggml_type_enum.sh ships green with them).

Refs #3430, #3421, #3431, #3452, #3563

ont-delta: none this PR touches README.md only to update a GENERATED CONTRACT_COUNT block (1799 -> 1800, written by make readme-sync, required by guard-cargo's FALSIFY-README-002); it adds no prose, no concept and no new ontology surface, and the one contract it adds is already a row in contracts/census.json and contracts/contracts.nt.

keep-open: every reference here is to work this PR does not finish — #3430 still owes Phases 3-4 (the three enums become re-exports and each crate keeps its admitted id set), #3421 is the epic, #3431 is M2 and this PR only supplies its input rows, #3563 is the general form of the pin assertion this PR instantiates once, and #3452 is the plan PR whose disposition is tab 1's call.

🤖 Generated with Claude Code

[noahgift]

quorum-review (AD-04): NOT agreed (auto_merge: checked=true was_armed=false disarmed=false)

{
 "ticket": "PMAT-3430",
 "head": "862f7fa6543b35eb0e73fc2ac91d10d059a86037",
 "width": 3,
 "executor": "agy",
 "agreed": false,
 "auto_merge": {
  "checked": true,
  "was_armed": false,
  "disarmed": false,
  "note": "auto-merge not armed"
 },
 "lanes": [
  {
   "lane": 1,
   "verdict": "NO-VERDICT",
   "findings": 0
  },
  {
   "lane": 2,
   "verdict": "NO-VERDICT",
   "findings": 0
  },
  {
   "lane": 3,
   "verdict": "PASS",
   "findings": 5
  }
 ]
}

[noahgift]

quorum-review (AD-04): NOT agreed (auto_merge: checked=true was_armed=false disarmed=false)

{
 "ticket": "PMAT-3430",
 "head": "862f7fa6543b35eb0e73fc2ac91d10d059a86037",
 "width": 3,
 "executor": "agy",
 "agreed": false,
 "auto_merge": {
  "checked": true,
  "was_armed": false,
  "disarmed": false,
  "note": "auto-merge not armed"
 },
 "lanes": [
  {
   "lane": 1,
   "verdict": "PASS",
   "findings": 0
  },
  {
   "lane": 2,
   "verdict": "NO-VERDICT",
   "findings": 0
  },
  {
   "lane": 3,
   "verdict": "PASS",
   "findings": 5
  }
 ]
}

[noahgift]

quorum-review (AD-04): three PASS — agreed (auto_merge: checked=true was_armed=false disarmed=false)

{
 "ticket": "PMAT-3430",
 "head": "9bbcaa42b7ada310dcdbb5900b9e42fe6e86acd6",
 "width": 3,
 "executor": "agy",
 "agreed": true,
 "auto_merge": {
  "checked": true,
  "was_armed": false,
  "disarmed": false,
  "note": "auto-merge not armed"
 },
 "lanes": [
  {
   "lane": 1,
   "verdict": "PASS",
   "findings": 0
  },
  {
   "lane": 2,
   "verdict": "PASS",
   "findings": 0
  },
  {
   "lane": 3,
   "verdict": "PASS",
   "findings": 5
  }
 ]
}

[github-actions]

§13.11 rung 1 — quorum shadow verdict

S13-SHADOW pr=3581 head=d51343bee9aedd099e7aa944996007f1e18dcd2a verdict=REFUSE class=Q1 arm_rc=1

Shadow mode: this records a verdict and merges nothing. A refusal
to arm is not a block (§13 adds zero rows to §7) — the pull request is
exactly as green as it was.

[noahgift]

quorum-review (AD-04): NOT agreed (auto_merge: checked=true was_armed=false disarmed=false)

{
 "ticket": "PMAT-3430",
 "head": "d51343bee9aedd099e7aa944996007f1e18dcd2a",
 "width": 3,
 "executor": "agy",
 "agreed": false,
 "auto_merge": {
  "checked": true,
  "was_armed": false,
  "disarmed": false,
  "note": "auto-merge not armed"
 },
 "lanes": [
  {
   "lane": 1,
   "verdict": "PASS",
   "findings": 1
  },
  {
   "lane": 2,
   "verdict": "NO-VERDICT",
   "findings": 0
  },
  {
   "lane": 3,
   "verdict": "NO-VERDICT",
   "findings": 0
  }
 ]
}

[noahgift]

quorum-review (AD-04): three PASS — agreed (auto_merge: checked=true was_armed=false disarmed=false)

{
 "ticket": "PMAT-3430",
 "head": "d51343bee9aedd099e7aa944996007f1e18dcd2a",
 "width": 3,
 "executor": "agy",
 "agreed": true,
 "auto_merge": {
  "checked": true,
  "was_armed": false,
  "disarmed": false,
  "note": "auto-merge not armed"
 },
 "lanes": [
  {
   "lane": 1,
   "verdict": "PASS",
   "findings": 3
  },
  {
   "lane": 2,
   "verdict": "PASS",
   "findings": 2
  },
  {
   "lane": 3,
   "verdict": "PASS",
   "findings": 6
  }
 ]
}