feat(falsify-ship-016): PARTIAL discharge — apr qa 8-of-8 aggregate-AND verdict fn by noahgift · Pull Request #1008 · paiml/aprender

noahgift · 2026-04-22T15:01:37Z

Summary

Wires GATE-ARCH-370M-008 (AC-SHIP2-006, FALSIFY-SHIP-016) to a pure
verdict_from_qa_gates(&[bool]) -> Ship016Verdict aggregate-AND fn in
crates/aprender-train/src/models/llama_370m.rs, proven today by:

Exhaustive 2^8 = 256-combination sweep — exactly one input (all-true) yields Pass; the other 255 yield Fail
Single-gate-flip falsifiability — flipping any one gate true→false flips aggregate to Fail (8 cases)
Monotonicity — flipping false→true never regresses Pass→Fail (pairwise sweep)
3 contract-drift guards — slice length 0/7/9/16 → Fail conservatively even when all-true
Provenance pin — AC_SHIP2_006_REQUIRED_QA_GATE_COUNT const == 8 matches spec §7

discharge_status: PARTIAL_ALGORITHM_LEVEL. Full discharge blocks on real 370M .apr + 8-gate apr qa harness invocation (fixture-swap only, no harness rewrite).

Pattern note

SHIP-016 is the first aggregate-AND shape — SHIP-017/018/020 were single-threshold shapes. The proof pattern now covers two distinct decision-rule shapes (threshold + aggregate-AND), confirming decision-rule/compute-harness separation is a reusable pattern, not a one-off.

5th PARTIAL — "exhausted" verdict now falsified 4×

Chain: SHIP-019 → SHIP-017 → SHIP-020 → SHIP-018 → SHIP-016. Each was found after a prior "search exhausted" verdict. Lesson: re-run counter-example survey before declaring search space exhausted.

MODEL-2 ship-gate coverage

Status	Count	Gates
ACTIVE	3/12	001, 011, 012
PARTIAL	7/12	002 (SHIP-012), 005 (SHIP-015), 006 (SHIP-016, new), 007 (SHIP-017), 008 (SHIP-018), 009 (SHIP-019), 010 (SHIP-020)
Truly compute-blocked	2/12	003 (CE ≤ 2.2 val loss), 004 (≤21-day wall-clock)

10/12 touched (83.3%) — up from 9/12 (75.0%) after SHIP-018.

Files changed

contracts/model-families/llama-370m-sovereign-v1.yaml v1.5.0 → v1.6.0 (GATE-ARCH-370M-008 block added; stays ACTIVE)
crates/aprender-train/src/models/llama_370m.rs — const + enum + pure fn + 2 tests
docs/specifications/aprender-train/ship-two-models-spec.md v2.23.0 → v2.24.0 (amendment block)
crates/aprender-train/src/train/device.rs — pre-existing fmt fixes bundled per Toyota Way

Test plan

cargo test -p aprender-train --lib models::llama_370m — 12/12 PASS including 2 new SHIP-016 tests
cargo fmt -p aprender-train --check — clean
cargo clippy -p aprender-train --lib -- -D warnings — clean
pv validate contracts/model-families/llama-370m-sovereign-v1.yaml — 0 errors, 0 warnings
CI ci / gate + workspace-test green

Refs #152

🤖 Generated with Claude Code

…ND verdict fn Wires GATE-ARCH-370M-008 (AC-SHIP2-006) to a pure verdict_from_qa_gates(&[bool]) -> Ship016Verdict aggregate-AND fn in aprender-train/src/models/llama_370m.rs, proven today by exhaustive 2^8 = 256-combination sweep + single-gate-flip falsifiability + monotonicity + 3 contract-drift guards (slice length 0/7/9/16 → Fail even when all-true). Discharge marker: PARTIAL_ALGORITHM_LEVEL. Pattern note: SHIP-016 is the first aggregate-AND shape — SHIP-017/018/020 were single-threshold shapes. The proof pattern now covers two distinct decision-rule shapes, confirming decision-rule/compute-harness separation is a reusable pattern, not a one-off. **5th PARTIAL after "exhausted" verdict falsified 4× already** (SHIP-019 → SHIP-017 → SHIP-020 → SHIP-018 → SHIP-016). **MODEL-2 ship-gate coverage: 3/12 ACTIVE + 7/12 PARTIAL = 10/12 touched (83.3%).** Remaining 2 truly compute-blocked (003 CE ≤ 2.2, 004 ≤21-day wall-clock) have no fixture-swap trick. Changes: - contracts/model-families/llama-370m-sovereign-v1.yaml v1.5.0 → v1.6.0 (GATE-ARCH-370M-008 block added; stays ACTIVE) - crates/aprender-train/src/models/llama_370m.rs: + AC_SHIP2_006_REQUIRED_QA_GATE_COUNT = 8 const + Ship016Verdict enum + verdict_from_qa_gates(&[bool]) pure fn with aggregate-AND + falsify_ship_016_apr_qa_aggregate_and_logic test (2^8 sweep + single-gate-flip + monotonicity + 3 contract-drift guards) + falsify_ship_016_gate_arch_370m_008_has_partial_discharge_marker test (YAML binding: binds_to AC-SHIP2-006, falsification_id FALSIFY-SHIP-016, discharge_status PARTIAL_ALGORITHM_LEVEL) - docs/specifications/aprender-train/ship-two-models-spec.md v2.23.0 → v2.24.0 (amendment block documenting 5th PARTIAL, first aggregate-AND shape) - crates/aprender-train/src/train/device.rs: pre-existing fmt fixes bundled per Toyota Way "all defects are your defects" Full discharge blocks on: real 370M .apr from AC-SHIP2-003/004 compute-dispatch + 8-gate apr qa harness invocation with exit 0 → feed the 8 gate-result booleans into verdict_from_qa_gates and require Ship016Verdict::Pass. Fixture-swap only — no harness rewrite. Refs #152 Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

13 workspace Cargo.toml edits converting pinned crates.io versions of renacer, entrenar, entrenar-lora, batuta, and trueno-graph to `{ workspace = true }` aliases pointing at the in-tree crates (aprender-profile, aprender-train, aprender-orchestrate, aprender-graph). Root cause: after APR-MONO Phase 2/3 consolidation, several crates still pinned old crates.io sibling versions. Each carried a transitive dep on an older `aprender` (0.14.1, 0.27.8), producing a diamond that broke `cargo clippy -p aprender` with an ambiguous-package error. Post-fix metadata: aprender: 1 package (workspace 0.31.2) renacer / entrenar / trueno-graph: 0 crates.io packages Unblocks main-andon (required CI checks) and PRs #1008 (FALSIFY-SHIP-016 PARTIAL) and #1009 (FALSIFY-SHIP-009 PARTIAL). Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

* fix(apr-mono): eliminate sibling-pin diamond dep (muda) 13 workspace Cargo.toml edits converting pinned crates.io versions of renacer, entrenar, entrenar-lora, batuta, and trueno-graph to `{ workspace = true }` aliases pointing at the in-tree crates (aprender-profile, aprender-train, aprender-orchestrate, aprender-graph). Root cause: after APR-MONO Phase 2/3 consolidation, several crates still pinned old crates.io sibling versions. Each carried a transitive dep on an older `aprender` (0.14.1, 0.27.8), producing a diamond that broke `cargo clippy -p aprender` with an ambiguous-package error. Post-fix metadata: aprender: 1 package (workspace 0.31.2) renacer / entrenar / trueno-graph: 0 crates.io packages Unblocks main-andon (required CI checks) and PRs #1008 (FALSIFY-SHIP-016 PARTIAL) and #1009 (FALSIFY-SHIP-009 PARTIAL). Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> * fix(ci-lint): unblock workspace clippy exposed by sibling-alias activation The sibling-alias MUDA fix in 3f38207 activates crates that were never clippy-gated under main's red CI. This commit adds the minimum lint-unblock changes needed for `cargo clippy --all-targets -- -D warnings` to pass. Code fixes: - aprender-compute GPU path (wgsl_forward, cached_matmul, dispatch, parallel): underscore-prefix unused vars, `.unwrap()` → `.expect()` with descriptive messages, collapsible_if, doc_lazy_continuation, allow map_entry, drop two bogus `contract_post_*!()` macro calls (type-mismatched, never compiled). - aprender-graph/shortest_path.rs: add `#[must_use]` + doc on dijkstra_path, `.map_or(false, |&d| cost > d)` → `.is_some_and(|&d| cost > d)`. - aprender-graph/pattern.rs: `mapping.iter()` → `mapping`. Crate-level lint allows (pedantic lints carried over from pre-monorepo crates — to be fixed per-lint in dedicated follow-up PRs): - aprender-profile/src/lib.rs: 9 pedantic lints (renacer legacy). - apr-cli/tests/*.rs (13 files): `disallowed_methods` etc for integration tests where `unwrap()` is idiomatic. - apr-cli/examples/tool_calling_demo.rs: `disallowed_methods` for serde_json `json!` macro expansion. Dead code (MUDA-aligned cleanup): - Delete apr-cli/examples/probar_tui_testing.rs + federation_tui_demo.rs — both reference the removed `ratatui` crate and deleted `federation::tui` module. - Drop matching `[[example]]` entries from apr-cli/Cargo.toml. Verification: cargo clippy --all-targets -- -D warnings # clean in 36s Ported from closed PR #1010 (commit 4338cbf on fix/apr-mono-muda). Keeps #1011's Cargo.toml changes intact; no conflicts with the MUDA fix commit. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.7 <noreply@anthropic.com>

noahgift · 2026-04-23T18:17:18Z

Superseded by #1035 — rebased onto MODEL-1 stack + SHIP-017 + SHIP-020 + SHIP-018 at v2.37.0.

noahgift mentioned this pull request Apr 22, 2026

fix(apr-mono): eliminate sibling-pin diamond dep (muda) #1011

Merged

4 tasks

noahgift mentioned this pull request Apr 23, 2026

feat(falsify-ship-016): MODEL-2 AC-SHIP2-006 PARTIAL discharge (restacked) #1035

Closed

5 tasks

noahgift closed this Apr 23, 2026

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

feat(falsify-ship-016): PARTIAL discharge — apr qa 8-of-8 aggregate-AND verdict fn#1008

feat(falsify-ship-016): PARTIAL discharge — apr qa 8-of-8 aggregate-AND verdict fn#1008
noahgift wants to merge 1 commit into
mainfrom
feat/falsify-ship-016-partial-discharge

noahgift commented Apr 22, 2026

Uh oh!

noahgift commented Apr 23, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Conversation

noahgift commented Apr 22, 2026

Summary

Pattern note

5th PARTIAL — "exhausted" verdict now falsified 4×

MODEL-2 ship-gate coverage

Files changed

Test plan

Uh oh!

noahgift commented Apr 23, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant