…@1 ≥86.00% (1.2 pp noise → 84.80%) (#1021)

* feat(ship-two-001): FALSIFY-SHIP-005 PARTIAL discharge — MODEL-1 HumanEval pass@1 ≥86.00% (1.2 pp noise → effective 84.80%) clean-branch rebuild

Wires MODEL-1 `apr eval --benchmark humaneval` ship floor (AC-SHIP1-005)
to a pure two-number threshold verdict fn. Clean-branch rebuild of the
SHIP-005 delta from the now-superseded stacked PR #1015 (commit
8c497a0 which layered on top of SHIP-007 PR #1019 prior to SHIP-002
landing). Re-based directly on current main (f615148) so SHIP-005
stands alone — SHIP-007 (#1019) remains open blocked on infra defect
#1020 (runner-disk-guard cross-runner race) and SHIP-018 (#1004) still
in flight.

MODEL-1 AC-SHIP1 coverage on main: 3/10 touched (SHIP-009 + SHIP-008 +
SHIP-006) pre-SHIP-002; 4/10 once SHIP-002 landed via f615148;
5/10 once this PR lands. Mirrors MODEL-2 SHIP-018 shape (pass@1
threshold) but uniquely carries a 1.2 pp noise allowance called out by
spec §4.2 AC-SHIP1-005 — MODEL-2 has no noise window.

contracts/qwen2-e2e-verification-v1.yaml v1.1.0 → v1.2.0:
  - Adds FALSIFY-QW2E-SHIP-005 binding
      AC_SHIP1_005_NOMINAL_HUMANEVAL_PASS_AT_1_PCT = 86.00
      AC_SHIP1_005_NOISE_ALLOWANCE_PP              = 1.20
      AC_SHIP1_005_EFFECTIVE_HUMANEVAL_PASS_AT_1_PCT ≈ 84.80
    to `verdict_from_pass_at_1(correct, total, threshold_pct) ->
    Ship005Verdict` in `crates/aprender-core/src/metrics/ship_005.rs`.
  - 8-section mutation survey:
      1. Safe-margin Pass above effective floor (85/100 = 85.0%)
      2. Above nominal floor (87/100 = 87.0%) Pass
      3. Noise-window Fail at nominal (85/100 Fails nominal)
      4. Below-effective Fail incl. HumanEval-canonical 139/164 = 84.756%
      5. Monotonicity sweep correct=0..=164 at effective
      6. Div-safety (total=0) + sanity (correct>total) → Fail
      7. Non-finite threshold (NaN, ±∞) → Fail conservatively
      8. Tolerance-bounded provenance pin on all three constants
         (86.0 − 1.2 in f32 yields ~84.79999924, not exact 84.80).
  - `ship_blocking: true`, `discharge_status: PARTIAL_ALGORITHM_LEVEL`,
    `full_discharge_blocks_on: live apr eval --benchmark humaneval ...`
    on RTX 4090; 6 named counter_example_classes.

crates/aprender-core/src/metrics/ship_005.rs (NEW, 305 lines):
  - Three-constant design unique to MODEL-1 (SHIP-007/018 had one).
  - `#[must_use] pub fn verdict_from_pass_at_1(...)` returns
    `Ship005Verdict::Fail` conservatively on: total=0 (div guard),
    correct>total (sanity), !threshold.is_finite() (NaN/±∞).
  - `falsify_ship_005_humaneval_pass_at_1_threshold_logic` — 1 passing.

Spec `docs/specifications/aprender-train/ship-two-models-spec.md`
  v2.26.0 → v2.27.0 annotates AC-SHIP1-005 + FALSIFY-SHIP-005 rows
  `**(PARTIAL_ALGORITHM_LEVEL v2.27.0)**` and appends amendment entry
  noting 11 PARTIAL + 3 DISCHARGED across both models, MODEL-1 5/10.

Authored self-contained because SHIP-018 PR #1004 and SHIP-007 PR
#1019 are not yet on main. Once they land, the two (or three)
`verdict_from_pass_at_1_*` fns should be dedup'd into a single
parameterized helper.

Full discharge blocks on: live `apr eval --benchmark humaneval
paiml/qwen2.5-coder-7b-apache-q4k-v1 --json` on RTX 4090 with
--features cuda; median pass@1 across 3 seed=0 runs ≥ 86.00 (or
≥ 84.80 under the 1.2 pp noise allowance).

Tests:
  cargo test -p aprender-core --lib \
    falsify_ship_005_humaneval_pass_at_1_threshold_logic
Contract:
  cargo run --quiet -p aprender-contracts-cli --bin pv -- validate \
    contracts/qwen2-e2e-verification-v1.yaml

Supersedes #1015 (stacked-branch original).

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

* ci: retrigger after disk-guard stuck workspace-test (#1021)

Previous run 24806028080 workspace-test stuck at 37+min on lib-tests step
(vs typical 19min). Canceled. Re-triggering on fresh runner.

Tracking: infra issue #1020 second incidence.

---------

Co-authored-by: Claude Opus 4.7 <noreply@anthropic.com>

…RL + SHA-256 verdict rules (6/10) (#1022)

Wires MODEL-1 AC-SHIP1-010 ("published artifact URL resolves; SHA-256
matches manifest") to pure algorithm-level decision rules in
`crates/aprender-core/src/format/ship_010.rs` via two verdict fns and
a 7-section mutation survey per side.

Two constants bind the ship rules:

- `AC_SHIP1_010_SHA256_HEX_LEN = 64` — sha256 = 32 bytes = 64 lowercase
  hex chars, per canonical `sha256sum` output. Any digest whose length
  differs, whose case differs, or whose bytes are not `[0-9a-f]` is
  rejected before any equality comparison.

- `AC_SHIP1_010_REQUIRED_URL_SCHEME = "https://"` — TLS floor codified
  as a byte-literal per §4.2 (plaintext `http://` is MITM-spoofable
  and is a ship-blocker, not a warning).

Two pure verdict fns in `format/ship_010.rs`:

- `verdict_from_sha256_match(expected_hex, actual_hex) -> Ship010Verdict`
  — length gate + canonical-lowercase-hex gate + byte-equal compare.
  Short-circuits to `Fail` before any string comparison when either
  input is malformed.

- `verdict_from_manifest_url(url) -> Ship010Verdict` — starts-with
  `https://` + non-empty host + ASCII-whitespace/control byte rejection.
  Accepts `huggingface.co/...` and `...amazonaws.com/...` canonical
  forms; rejects plaintext, scheme-less, empty-host, and
  whitespace/control-poisoned URLs.

7-section mutation survey per fn (proves each precondition is
load-bearing):

- SHA-256: identical-hex Pass / single-hex-flip Fail / wrong-length
  Fail / uppercase-hex rejected / non-hex (`g`..`z`) rejected / all-zero
  guard / provenance pin on constant `AC_SHIP1_010_SHA256_HEX_LEN`.

- URL: HF canonical Pass / S3 canonical Pass / plaintext `http://`
  Fail / scheme-less Fail / empty-host (`https://`) Fail /
  whitespace-control `\n \t \r` rejected / provenance pin on constant
  `AC_SHIP1_010_REQUIRED_URL_SCHEME`.

Contract `publish-manifest-v1.yaml` v1.3.0 → v1.4.0 adds a new
`FALSIFY-SHIP-010` block under `falsification_tests:` binding the
parent AC (`parent_acceptance_criteria: AC-SHIP1-010`), listing the
two constants under `binds_constants:`, and pointing
`evidence_discharged_by:` at the three Rust test fns. Status stays
ACTIVE; discharge level is `PARTIAL_ALGORITHM_LEVEL` — full discharge
blocks on a live `curl -sSI <artifact_url>` 200-OK + `sha256sum
<local_file>` against a freshly-pulled `paiml/qwen2.5-coder-7b-apache-q4k-v1`
file, verified against the manifest SHA-256 on a host with HF
network egress.

Coverage math post-landing:

- MODEL-1: 5/10 → **6/10** touched (1 DISCHARGED from SHIP-001, plus
  five PARTIALs on SHIP-002 / SHIP-005 / SHIP-006 / SHIP-007 /
  SHIP-008 and now SHIP-010). First MODEL-1 network-dependent PARTIAL
  — others have been format / algorithm / threshold rules.

- Combined both-models tally: 12 PARTIAL + 3 DISCHARGED (was 11 + 3).

Why self-contained (not stacked on PR #1019 SHIP-007): SHIP-010 lives
in `format/` next to LAYOUT contracts, not in `metrics/` or `qa/`;
the two domains are orthogonal, so the PR is based on fresh main
rather than stacked. `publish-manifest-v1.yaml` has no overlap with
`qwen2-e2e-verification-v1.yaml` (SHIP-007 home).

Dogfood evidence:

- `cargo build -p aprender-core --lib` → green (14.83s)
- `cargo test -p aprender-core --lib format::ship_010` →
  `3 passed; 0 failed; 0 ignored`
- `pv validate contracts/publish-manifest-v1.yaml` →
  `0 error(s), 0 warning(s). Contract is valid.`
- `cargo fmt -p aprender-core -- --check` → clean

Spec bump: v2.27.0 → v2.28.0 (entry added at top of header; AC table
row for AC-SHIP1-010 tagged `PARTIAL_ALGORITHM_LEVEL v2.28.0`).