spec(ship-two-models): v3.02.0 → v3.03.0 — §58 v0.32.0 cascade publish + 4 RE defects closed

[noahgift]

Summary

§58 records the release-engineering hygiene track that landed during the §57 drift-sweep wait window:
the v0.32.0 user-facing-crate cascade publish (Issue #1514 CLOSED) and the four hidden defects
the cascade surfaced + closed in their own PRs.

This is the third hygiene amendment in a row (§56 LIVE smoke → §57 drift sweep → §58 publish cascade);
the next §59 will record 5g.1 completion when the manifest emits (currently 62 shards / 16h19m wall, pending).

What's recorded

• Issue v0.32.0 cascade publish: workspace-level path-only dev-dep audit + topological publish #1514 CLOSED at 2026-05-05T16:14:56Z (v0.32.0 cascade publish)
• 4 user-facing crates verified live on crates.io: aprender, aprender-rag, aprender-core, apr-cli (all v0.32.0)
• 4 hidden defects surfaced + closed (each its own PR): feat(aprender-rag): v0.31.2 → v0.32.0 — rename lib trueno_rag → aprender_rag (BREAKING) #1512 (lib-name drift), fix(aprender-orchestrate): cmd_code arg-count drift — pass None for emit_trace #1513 (arg-count drift), fix(aprender-core): path-only dev-deps — break publish-time cycle (Refs #1514) #1515 + fix(aprender-core): permissive version + path on dev-deps — clean-room compat (Refs #1514) #1517 (dev-dep cycle), fix(apr-cli): bring aliases.yaml into crate dir for cargo publish (Refs #1514) #1518 (include_str scope)
• PR feat(pv-lint): add --strict-test-binding flag — catches dangling test refs (Closes #1510) #1511 (pv-lint --strict-test-binding) closes §57.4's foreshadowed prevention rule

Why §58 records release-engineering instead of 5g.1 completion

§57.7 foreshadowed §58 = "(a) 5g.1 full-run completion, or (b) 5g.2 dispatch result." Neither has fired yet.
But Issue #1514 CLOSED is a major user-facing milestone (cargo install aprender@0.32.0 now works on any host).
Recording it in real time avoids conflating two unrelated narratives when 5g.1 fires.
Per §57.4 lesson: 1 amendment ≈ 1 logical event.

Ship-%

• MODEL-1: unchanged at 91% (release-engineering hygiene, not falsifier flip)
• MODEL-2: unchanged at 57% until step 5g.3 produces val_loss < 9.38

Test plan

• Spec MD-only edit (no code changes)
• Verified cargo search aprender returns "0.32.0"
• Verified cargo search aprender-rag returns "0.32.0"
• Verified cargo search aprender-core returns "0.32.0"
• Verified cargo search apr-cli returns "0.32.0"
• Verified Issue v0.32.0 cascade publish: workspace-level path-only dev-dep audit + topological publish #1514 state CLOSED
• §58 cadence preserves §57.7's foreshadowed §59 = 5g.1 completion
• Auto-merge once required checks pass

Refs: #1514

🤖 Generated with Claude Code