…scharges (#1555)

* docs(contracts): SHIP-007 §22 upstream blocker RESOLVED — evidence pins on 5 MODEL-1 PARTIAL discharges

After M-FFN-GGUF-5 fix MERGED on aprender main 2026-05-07 (PR #1550
squash e856eb9), the §27 layer-3 ffn_swigl APR-vs-GGUF divergence
is closed: live H1 CONFIRMED at layer-3 ratio 1.245× (was 18.23×
pre-methodology-fix). 5 MODEL-1 PARTIAL discharges become live-
dispatch-ready: SHIP-002, SHIP-005, SHIP-006, SHIP-007, SHIP-008.

This PR adds evidence-pin annotations to each of the 3 contracts that
hold those discharges, citing PR #1550 as upstream §22 blocker
resolution. Pure additive YAML — no behavioral or test changes.

Contracts touched (3 contracts × 5 ACs):
- contracts/qwen2-e2e-verification-v1.yaml v1.10.0 → v1.11.0
  (FALSIFY-QW2E-SHIP-002, FALSIFY-QW2E-SHIP-005, FALSIFY-QW2E-SHIP-007)
- contracts/apr-model-qa-v1.yaml v1.2.0 → v1.3.0
  (FALSIFY-QA-SHIP-006)
- contracts/chat-template-v1.yaml v1.1.0 → v1.2.0
  (GATE-CHAT-SHIP-008)

Each contract's full_discharge_blocks_on clause now includes:
"Upstream blocker SHIP-007 §22 RESOLVED 2026-05-07 (aprender PR #1550
squash e856eb9; M-FFN-GGUF-5 fix); live discharge is now dispatch-
ready — no further upstream blockers."

This is bookkeeping work that captures the cascade outcome in the
contract surface so the next operator-dispatched LIVE-run session
has the citation ready. Each individual discharge still requires its
own LIVE run on RTX 4090 per the canonical command in
full_discharge_blocks_on (apr run / apr eval / apr bench / apr qa).
This PR does NOT promote PARTIAL_ALGORITHM_LEVEL → DISCHARGED — that
needs the LIVE evidence files.

Companion: scripts/ship-discharges/ship-XXX-discharge.sh dispatch
scripts authored in parallel by sub-agent (separate PR).

Test plan:
- [x] pv validate contracts/qwen2-e2e-verification-v1.yaml → 0 errors
- [x] pv validate contracts/apr-model-qa-v1.yaml → 0 errors
- [x] pv validate contracts/chat-template-v1.yaml → 0 errors
- [x] No code changes; production hot paths byte-unchanged

Refs PMAT-CCPA, SHIP-007 §22, M-FFN-GGUF-5 PR #1550.

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

* feat(ship-discharges): 5 live-dispatch scripts for MODEL-1 PARTIAL discharges

After SHIP-007 §22 upstream blocker resolved (PR #1550 merged 2026-05-07),
SHIP-002/005/006/007/008 are LIVE-dispatch-ready. Each script runs the
canonical command from its contract's `full_discharge_blocks_on:` clause,
parses output, emits evidence JSON, and prints Pass/Fail verdict.

Scripts (978 LOC total, all bashrs lint clean):
- ship-002-discharge.sh — `apr run` + python AST parse, 0 syntax errors
- ship-005-discharge.sh — 3 HumanEval runs (seed=0), median pass@1 ≥ 86.00%
  (or ≥ 84.80% with 1.2 pp noise allowance)
- ship-006-discharge.sh — `apr qa --json`, all 8 gates pass
- ship-007-discharge.sh — `apr bench`, median ≥ 30.0 tok/s on RTX 4090
- ship-008-discharge.sh — `apr run --print-prompt`, byte-exact ChatML golden

Each script:
- Defaults to /mnt/nvme-raid0/targets/aprender/release/apr (lambda-vector
  canonical), accepts --apr-binary and --model overrides
- Writes canonical evidence/ship-XXX-full-discharge/discharge-evidence-v1.json
  matching the format used by SHIP-001/003/004 (already DISCHARGED)
- Exits 0 on Pass, 1 on Fail; preflight rejects bad apr-binary / missing jq
- Strict shell hygiene: set -euo pipefail, quoted vars, mktemp with EXIT trap

.bashrsignore updated with audited SEC001 suppression — false positive on
the literal substring "eval" in `apr eval` (apr-cli HumanEval subcommand,
not the bash `eval` builtin).

Includes top-level README.md documenting the dispatch matrix, operator
workflow, and prerequisites.

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

---------

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