Summary

§50.4 cascade is INTEGRATION-COMPLETE on main following PR #1494 (5f.4 CLI wireup) merging at 2026-05-05T01:48:14Z (commit 9afca16). The apr pretrain --init <PATH> flow is now end-to-end functional on CPU; the legacy "not yet wired" Err is RETIRED; step 5g LIVE is the only remaining gate before MODEL-2 ship-% can move from 57% → ≥58%.

This PR records the cascade-completion milestone in the spec (§53) and bumps the contract from PARTIAL_ALGORITHM_LEVEL to FUNCTIONAL.

Spec amendment §53

• Updated falsifier scoreboard: 6/8 INTEGRATION (001/002/003/005/006/007 via live CLI dispatch); 2/8 PARTIAL_ALGORITHM_LEVEL (004 forward-pass smoke + 008 contract validation are inherently algorithm-level).
• Step roadmap: 5a-5f.4 ✅ MERGED; 5f.5 (CUDA wireup) NOT YET STARTED; 5g (LIVE 500-step fine-tune) operator-dispatchable on RTX 4090.
• Cascade ships statistics: 11 PRs over 2 days (feat(apr-cli): wire apr pretrain --init <model.apr> — §49 step 4 #1471/spec(ship-two-models): v2.94.0 → v2.95.0 — §50 MODEL-2 architecture-coupling finding #1472/contract(apr-pretrain-arch-polymorphic-v1): v1.0.0 PROPOSED — §50.4 step 5a #1473/fix(aprender-train): qwen2_0_5b tie_word_embeddings true — §50.4 step 5b + DEFECT FIX #1474/feat(aprender-train): build_transformer_config polymorphic dispatch — §50.4 step 5c #1475/feat(apr-cli): polymorphic preflight_tokenizer_vocab_matches_target — §50.4 step 5d #1476/test(aprender-train): GQA-7:1 forward-pass smoke test — §50.4 step 5e #1478/feat(aprender-train): validate_pretrain_init_arch_compatible — §50.4 step 5f.1 #1479/feat(aprender-train): load_init_tensors_from_apr — §50.4 step 5f.2 #1481/contract(apr-pretrain-arch-polymorphic-v1): v1.0.0 → v1.1.0 PARTIAL_ALGORITHM_LEVEL #1482/feat(aprender-train): populate_trainer_from_init_tensors — §50.4 step 5f.3 #1483/spec(ship-two-models): v2.96.0 → v2.97.0 — §52 cascade ALGORITHM-COMPLETE + 5f.4 wireup gap #1486/feat(apr-cli + aprender-train): apr pretrain --init wireup — §50.4 step 5f.4 #1494).
• MODEL-1 ship % unchanged at 91%; MODEL-2 ship % unchanged at 57% (gated on 5g empirical val_loss < 9.38 evidence).
• 3 CI andon classes documented as feedback memories during cascade (workspace-test missing-binary, trueno SIGSEGV-on-cleanup, auto-merge behind-state).

Contract bump v1.1.0 → v1.2.0 FUNCTIONAL

• All 8 falsifiers PASS on main; 6/8 reach INTEGRATION via the user-facing apr pretrain --init flow.
• verification_summary updated: tested 7 → 8; status partial → functional.
• Added §52 + §53 references.
• Promotion to DISCHARGED still requires §50.4 step 5g LIVE empirical 500-step fine-tune on canonical Qwen2.5-Coder-0.5B-Instruct.apr producing val_loss < 9.38.

Five Whys

1. Why is §53 a separate amendment from §52? §52 identified the wireup gap; §53 records its closure. Same-day spec hygiene per feedback_falsifier_first_cascade_pattern.md.
2. Why bump to FUNCTIONAL rather than DISCHARGED? FUNCTIONAL means "all falsifiers pass and the integration path is live"; DISCHARGED requires LIVE evidence on the canonical model+corpus combination (step 5g).
3. Why call out 6/8 INTEGRATION rather than 8/8? Two falsifiers are inherently algorithm-level: FALSIFY-004 (forward-pass smoke is a unit test, not a CLI flow) and FALSIFY-008 (contract validation is a pv smoke).
4. Why didn't §52 include the FUNCTIONAL bump? §52 was authored before 5f.4 landed. The contract was at v1.1.0 PARTIAL because 5f.3 was the last merge at that point.
5. Why is the cascade 11 PRs and not 1 mega-PR? Per feedback_falsifier_first_cascade_pattern.md: one PR ≈ one falsifier discharge or one author-step. The cascade discipline keeps each merge auditable and bisectable.

Test plan

• pv validate contracts/apr-pretrain-arch-polymorphic-v1.yaml exits 0
• PMAT pre-commit quality gates pass
• CI gate green (workspace-test, ci/gate)
• Auto-merge fires on green CI

🤖 Generated with Claude Code