…AT-690 P0-K extension) (#1757)

Extends the existing `apr stamp` command (PR #1050 — provenance fields)
to also patch HF identity + architecture family slug in place. Unblocks
in-place salvage of pre-P0-K APR checkpoints whose architecture stamps
were corrupted by the §82 P0-H fallback.

## Background — SPEC §86 root cause

The §85 P2-E live run produced 50 epoch checkpoints (~125 GB total) at
best val_loss=4.6227. P2-G v1 attempted to resume from P2-E ep49 and
the init eval surfaced val_loss=8.60 — proof that --init silently
failed to load the trained weights. Root cause: P2-E's init APR
pre-dates P0-K (PR #1742), so the P0-H fallback stamped
architecture="LlamaForCausalLM" into the trained checkpoint despite
the actual tensors being Qwen2-shaped. `apr pretrain --init` reads
the (wrong) architecture stamp and rejects the load.

## What this PR adds

Three new CLI flags on `apr stamp`:

- `--hf-architecture <CLASS>` (e.g. Qwen2ForCausalLM) — the HF class
  name. PMAT-690 P0-K's upstream stamp.
- `--hf-model-type <SLUG>` (e.g. qwen2) — config.json::model_type.
- `--architecture <SLUG>` (e.g. qwen2) — the lowercase family slug
  that `apr pretrain --init` reads for arch dispatch. **This is the
  load-bearing field** for §86 salvage — patching just hf_architecture
  alone won't unblock `apr pretrain --init`.

The existing `--license` / `--data-source` / `--data-license` flags
are unchanged. The patch struct's `has_any()` gate now accepts any
combination of the six fields; at least one must be specified or the
stamp is rejected up-front.

## Operator workflow for §86 salvage

```bash
# Patch a pre-P0-K Qwen2-actual-Llama-stamped checkpoint in place
apr stamp /path/to/p2e-epoch-049.apr \
  --architecture qwen2 \
  --hf-architecture Qwen2ForCausalLM \
  --hf-model-type qwen2 \
  -o /path/to/p2e-epoch-049-stamped.apr

# Verify
apr inspect /path/to/p2e-epoch-049-stamped.apr --quality --json | jq .quality
# breakdown.hf_identity should jump 0 → 20

# Now usable as init for resume training:
apr pretrain --init /path/to/p2e-epoch-049-stamped.apr ...
```

## Discharges

- §86 SPEC amendment (`evidence/p2g-2026-05-17/section-86-draft.md`) —
  workaround #2 (in-place restamp)
- Salvages ~125 GB of pre-P0-K P2-E checkpoints without a 53-min retrain
- Establishes a pattern for in-place metadata patching that future
  spec amendments can build on (e.g., a `--name` / `--description`
  extension for model card metadata)

## Tests

- 2 new unit tests in `aprender-core::format::v2::stamp` (extends 6
  → 6, existing tests adjusted for new struct fields via Default)
- 2 new CLI tests in `apr-cli::commands::stamp` (extends 5 → 7):
  - `stamp_p0k_recovers_pre_p0k_apr_identity` — full §86 use case
  - `stamp_p0k_partial_hf_architecture_only` — verifies field
    independence (stamp one without touching others)
- All 5,944 apr-cli lib tests pass — 0 regressions
- All 13,800 aprender-core lib tests pass — 0 regressions

## Refs

- PR [#1742](#1742) (PMAT-690 P0-K base)
- PR [#1750](#1750) (P3-A `apr inspect --quality` scorer)
- PR [#1754](#1754) (SPEC §85 P2-E findings)
- PR #1050 (the original `apr stamp` PR — this extends it)
- `docs/specifications/aprender-train/ship-model-2-spec.md §86` (forthcoming)
- `evidence/p2g-2026-05-17/section-86-draft.md` (root cause + workaround analysis)
- `memory/feedback_upstream_metadata_masquerade.md` (methodology #33)

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

…H-MATCH-001 (SPEC §86.6 closure) (#1761)

Codifies the INV-INIT-ARCH-MATCH-001 invariant authored as runtime code
in PR #1760 (`validate_init_arch_matches_tensor_evidence` in
aprender-train::train::pretrain_real). Adds:

- FALSIFY-INIT-ARCH-MATCH-001: integration falsifier bound to the
  unit-test family `cargo test -p aprender-train --lib
  inv_init_arch_match_001` (7 tests covering: canonical §86 reject,
  inverse reject, matching qwen2 accept, matching llama accept, None
  metadata skip, unmappable metadata skip, GGUF-unknown tensor skip).
- INV-INIT-ARCH-MATCH-001 proof_obligation: safety invariant — when
  both metadata.architecture and tensor-name-inferred family resolve
  to concrete distinct slugs, gate MUST fail-fast before any training
  step. No false-positive when either side returns "unknown".

## Salvage path

The error message includes an inline `apr stamp` recipe (PR #1757):

```
apr stamp <pre-p0k.apr> --architecture qwen2 --hf-architecture Qwen2ForCausalLM \
                       -o <stamped.apr>
apr pretrain --init <stamped.apr> ...
```

## Refs

- PR #1742 (PMAT-690 P0-K base — producer-side stamping)
- PR #1750 (P3-A `apr inspect --quality` — surfaces hf_identity=0/20 pre-stamp)
- PR #1754 (SPEC §85 P2-E findings — context)
- PR #1757 (apr stamp HF identity extension — salvage path)
- PR #1758 (SPEC §86 amendment — defect specification this contract closes)
- PR #1760 (INV-INIT-ARCH-MATCH-001 runtime implementation)
- memory/feedback_upstream_metadata_masquerade.md (methodology #33)

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

…C-prep defect 1) (#1769)

Closes Defect 1 surfaced by the §86 publish-readiness preflight on
P2-E ep49: pre-P0-K APRs trained from inits without embedded tokenizers
fail `apr run` with PMAT-172 ("APR file missing embedded tokenizer").
Without this fix, the §86 salvage produces a 6 GB HF-publish-ready
directory whose headline command doesn't work.

## What ships

- `ProvenancePatch` gains three optional fields:
  - `tokenizer_vocab: Option<Vec<String>>` — token strings indexed by id
  - `tokenizer_merges: Option<Vec<String>>` — BPE merge rules
  - `tokenizer_model_type: Option<String>` — e.g. "BPE", "Unigram"
- `stamp_provenance_bytes` extended to write these into
  `metadata.custom["tokenizer.vocabulary"]` / `tokenizer.merges` /
  `tokenizer.model_type` AND set the HAS_VOCAB header flag (the
  load-bearing check in `apr run`'s PMAT-172 gate).
- `apr stamp` CLI gains `--tokenizer <DIR>` flag. Accepts either:
  - `<dir>/vocab.json + <dir>/merges.txt` (HF GPT-2/Qwen BPE format,
    the Qwen-coder pretrain default)
  - `<dir>/tokenizer.json` (HF unified format)

## Operator workflow post-this-PR

```bash
apr stamp /mnt/.../p2e-epoch-049.apr \
    --architecture qwen2 \
    --hf-architecture Qwen2ForCausalLM \
    --hf-model-type qwen2 \
    --license Apache-2.0 \
    --data-source "..." \
    --data-license "Apache-2.0 / permissive-aggregate" \
    --tokenizer /mnt/nvme-raid0/tokenizers/qwen-0.5b-tokenizer-v3/ \
    -o /tmp/albor-370m-v1.apr
# Resulting APR is self-contained: apr run works without --tokenizer flag
apr run /tmp/albor-370m-v1.apr "def fibonacci(n):" --max-tokens 32
```

## Tests

- 3 new CLI tests in `apr-cli::commands::stamp::tests`:
  - `stamp_p3c_defect1_embeds_tokenizer_from_vocab_merges` — full
    happy path: vocab.json + merges.txt → embedded vocab array +
    merges array + HAS_VOCAB flag + BPE model_type
  - `stamp_p3c_defect1_tokenizer_alone_passes_has_any_gate` —
    --tokenizer alone (no other patches) satisfies has_any()
  - `stamp_p3c_defect1_tokenizer_dir_without_files_errors` —
    empty dir surfaces clear "neither tokenizer.json nor vocab.json"
- 10/10 stamp tests pass (3 new + 7 existing updated for the
  new `tokenizer_dir: Option<&Path>` arg slot)
- aprender-core stamp.rs tests: 6/6 pass (existing literals updated
  for the 3 new ProvenancePatch fields)

## Refs

- PR #1742 (PMAT-690 P0-K base — upstream stamping)
- PR #1750 (P3-A apr inspect --quality — the diagnostic that surfaces
  hf_identity=0/20 + tokenizer=0/15 pre-stamp)
- PR #1757 (apr stamp HF identity extension — this PR extends it)
- evidence/p2e-2026-05-17/ (the run this defect was surfaced on)
- memory/feedback_publish_readiness_preflight.md (#37)
- PMAT-172 (the gate that motivates this fix)

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

…t (P3-C prep) (#1764)

Author the HuggingFace model card for `paiml/albor-370m-v1` and the
publish-readiness pre-flight script. Per SPEC §88: this model is
shipped as a stack-existence-proof, not a production code-completion
model. Both artifacts make that framing explicit so HF Hub users
calibrate expectations correctly.

## docs/model-cards/albor-370m-v1.md (255 lines)

Standard HF model card with model-index frontmatter:

- YAML metadata: Apache-2.0, code/python/stack-existence-proof tags,
  Qwen2.5-Coder-0.5B-Instruct base, codeparrot + the-stack-dedup
  datasets, val_loss=4.6227 / val_perplexity=101.78 metrics.
- §88 framing section spelling out the stack-existence-proof purpose.
- Training procedure table (architecture, optimizer, LR schedule,
  hardware, wall time, throughput — all from the §85 P2-E run).
- Trajectory table (every 5 epochs from 7.43 → 4.62).
- Intended uses (✅ stack demos, infra validation, tokenization
  round-trip, quantization research) vs NOT-recommended uses
  (production code-LM, zero-shot reasoning, long-context, HumanEval
  submission).
- Limitations (compute-bounded, plateau evidence, init lineage,
  val drift).
- Training data table (sources, sizes, licenses, role).
- How-to-use code snippets (apr CLI, Rust direct load, format export).
- Reproduce-the-run shell example using the exact §85 P2-E recipe.
- Citation, license/provenance, acknowledgments.

## scripts/publish/albor-370m-publish-readiness.sh (182 lines)

7-gate pre-flight checklist. GO / NO-GO verdict before invoking
`apr publish`. Gates:

1. `apr validate` exits 0
2. `apr inspect --quality` ≥ 90 (P3-A scorer; surfaces §86 salvage
   recipe inline if hf_identity < 20 or provenance < 25)
3. `apr qa --json` verdict = GO (8 gates)
4. Model card present + has HF YAML frontmatter
5. HF_TOKEN set
6. Smoke `apr run` produces text-like output
7. GGUF Q4_K + SafeTensors export round-trip both succeed

Exit 0 = ready to publish. Exit 1 = NO-GO with explicit blocker list.
Bashrs-validated (1 SEC011 false-positive on multi-condition rm -rf
guard; functionally safe).

## What this PR does NOT do

- Does NOT invoke `apr publish` (external action; requires user OK)
- Does NOT touch any APR files (read-only checks)
- Does NOT modify the §85 P2-E ep49 checkpoint (operator runs
  `apr stamp` via the §86.4 salvage recipe separately)

## Operator workflow (post-PR landing)

```bash
# 1. Stamp the pre-P0-K P2-E ep49 checkpoint to bring hf_identity up
apr stamp /mnt/nvme-raid0/runs/model-2-p2e-tuned-hp-20260517/ckpt/epoch-049.apr \
    --architecture qwen2 \
    --hf-architecture Qwen2ForCausalLM \
    --hf-model-type qwen2 \
    --license Apache-2.0 \
    --data-source "huggingface.co/Qwen/Qwen2.5-Coder-0.5B-Instruct + bigcode/the-stack-dedup + codeparrot/codeparrot-clean" \
    --data-license "Apache-2.0 / permissive-aggregate" \
    -o /tmp/albor-370m-v1.apr

# 2. Run the readiness check
bash scripts/publish/albor-370m-publish-readiness.sh /tmp/albor-370m-v1.apr
# Expected output: "VERDICT: GO" (or NO-GO with explicit blocker list)

# 3. Publish (still requires explicit user invocation)
apr publish paiml/albor-370m-v1 --formats apr,safetensors,gguf \
    --model-card docs/model-cards/albor-370m-v1.md
```

## Refs

- PR #1742 (PMAT-690 P0-K — upstream stamping)
- PR #1750 (P3-A `apr inspect --quality` — gate 2)
- PR #1754 (SPEC §84+§85+§86+§87+§88 stack — context)
- PR #1757 (apr stamp HF identity extension — §86 salvage)
- PR #1760 (INV-INIT-ARCH-MATCH-001 — validation chain)
- docs/specifications/aprender-train/ship-model-2-spec.md §88
- docs/specifications/aprender-train/albor-370m-roadmap.md §4 P3-C

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

…d --init (SPEC §86.6) (#1760)

Catches the §86 silent-failure pattern at the gate: when an APR's
metadata `architecture` claim contradicts what its tensor names imply,
`apr pretrain --init` exits non-zero with a clear naming-both-claims
error and an inline `apr stamp` salvage recipe.

## Background — the §86 case this catches

P2-G v1 was dispatched to resume P2-E ep49 for 10,000 more steps. The
init eval at step 0 produced val_loss = 8.60 — 1.86× P2-E ep49's
recorded 4.62. Silent failure: `--init` loaded random weights instead
of the trained checkpoint. Root cause walk-through:

1. `read_apr_architecture` parses `metadata.architecture = "LlamaForCausalLM"`
   (the §82 P0-H fallback when init_arch.hf_architecture is None).
2. `transformer_config_from_apr_metadata` builds a Llama-family
   TransformerConfig (dimensions correct, family discriminator wrong).
3. `populate_trainer_from_init_tensors` walks `trainer.named_parameters()` —
   produces Llama-style names — and looks them up in the APR tensor map
   which has Qwen2-style names. Mismatch → silent random-init fallback.
4. Training begins at random-init magnitude (val_loss ≈ 8.60).

This invariant catches step 1's wrong claim BEFORE step 3 silently
falls through.

## What this adds

Three new public functions in `aprender-train::train::pretrain_real`:

- `family_from_tensor_names(names: impl IntoIterator<Item=&str>)`
  → `&'static str` — lightweight tensor-name-only family inference
  (no data needed). Returns one of qwen3 / qwen2 / llama / mamba /
  rwkv / gpt-neox / opt / bert / gpt2 / unknown. Mirrors the
  heavyweight `infer_architecture_from_names` in
  aprender-core::format::converter::tokenizer_loader.
- `normalize_metadata_arch_family(arch: &str)` → `Option<&'static str>`
  — maps all three forms of the metadata `architecture` field to a
  canonical family slug: HF class names ("Qwen2ForCausalLM"), family
  slugs ("qwen2"), and capitalised legacy ("Qwen2"). Returns None
  for "unknown" / unmappable strings — caller treats as "no claim".
- `validate_init_arch_matches_tensor_evidence(metadata_arch, &tensors)`
  → `Result<(), String>` — the actual invariant gate. Errors with
  `FALSIFY-INIT-ARCH-MATCH-001` naming both the claimed and inferred
  families, plus an inline `apr stamp` recipe (PR #1757) for §86 salvage.

Wired into `build_shared_trainer_with_init` between `load_init_tensors_from_apr`
and `populate_trainer_from_init_tensors`. Read the raw metadata
`architecture` string via a new small helper (the `TransformerConfig`'s
`hf_architecture` field is None for pre-P0-K APRs — the §86 case — so
the cross-check needs the raw string field).

## Three skip-the-check fallback cases (no false-positives)

1. **No metadata claim** (metadata.architecture absent): nothing to
   contradict, allow.
2. **Unmappable claim** (e.g. "WeirdNovelArch"): novel arch is not §86,
   allow.
3. **Tensor inference returns "unknown"** (GGUF blk.* names can't
   disambiguate): trust the metadata, allow.

Only fail when BOTH inferences produce concrete family slugs AND they differ.

## Tests

- 7 new INV-INIT-ARCH-MATCH-001 tests in `pretrain_real::tests`:
  - `inv_init_arch_match_001_rejects_llama_stamped_qwen2_tensors` —
    canonical §86 case, must fail with falsifier ID + salvage recipe
  - `inv_init_arch_match_001_rejects_qwen2_stamped_llama_tensors` —
    inverse §86 case, must fail
  - `inv_init_arch_match_001_accepts_matching_qwen2/llama` — no
    false-positive on correctly-stamped APRs
  - `inv_init_arch_match_001_skips_when_metadata_absent` — None metadata
  - `inv_init_arch_match_001_skips_unmappable_metadata` — novel arch
  - `inv_init_arch_match_001_trusts_metadata_when_tensors_unknown` —
    GGUF blk.* case
- 1 helper test: `family_from_tensor_names_distinguishes_qwen2_from_llama`
- 1 normalizer test: `normalize_metadata_arch_family_handles_three_forms`

All 9 new tests pass. 7,595 existing aprender-train lib tests still pass
(the 3 pre-existing prune::snapshot_tests failures are insta-snapshot
drift in main, unrelated to this PR).

## Discharges

- §86.6 SPEC follow-up (forthcoming via #1758 stack)
- INV-INIT-ARCH-MATCH-001 invariant for `contracts/apr-pretrain-from-init-v1.yaml`
  (contract amendment is a separate small follow-up PR)

## Refs

- PR #1742 (PMAT-690 P0-K base — apr_convert + apr_import stamping)
- PR #1757 (apr stamp HF identity extension — the salvage path this
  invariant points operators to)
- PR #1758 (SPEC §86 amendment — context this invariant operationalizes)
- evidence/p2g-2026-05-17/section-86-draft.md
- memory/feedback_upstream_metadata_masquerade.md (methodology #33)

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

…erified (#1754)

* docs(spec): SPEC §84+§85 — P2-C/P2-E live findings, hyperparameter hypothesis CORROBORATED, P0-K closure live-verified

Two new spec sections + full P2-E evidence directory.

## §84 — P2-C dispatched; audit hypothesis FALSIFIED; P0-K surfaced

P2-C ran the audit-recommended multi-source corpus (49.6B tokens, 80×
§82's 1.24B) at the same hyperparameters as §82. Result: val_loss=4.91
@ ep20 (vs §82's 4.71) — IDENTICAL termination shape, +0.2 WORSE despite
80× more data. The Chinchilla-data-starvation hypothesis is FALSIFIED.

Debugging the §81-§83 5-PR cascade surfaced PMAT-690 P0-K: `apr convert`
(both apr_import and apr_convert paths) didn't stamp hf_architecture /
hf_model_type / embedded tokenizer. Five downstream consumer fixes had
been patching None values that read from the upstream gap. P0-K closes
the producer.

## §85 — P2-E live findings; hyperparameter hypothesis CORROBORATED

P2-E ran same qwen-v3 corpus at LR=1.5e-5 (-3.3× lower) + warmup=500
(5× longer). Result: val_loss=4.6227 @ ep49 — BELOW §82's 4.71 AND
P2-C's 4.91 floors. No early-stop; smooth monotonic descent across all
50 epochs. Hypothesis from §84 P2-E queue is CORROBORATED.

Training throughput: 15,460 tok/s pure (12,880 tok/s end-to-end with
checkpoint write) on RTX 4090, sm_89, cuBLAS TF32. This is the
canonical apr-cli CUDA training perf baseline for future dispatches.

§30 a-priori falsification lesson amendment: the audit's
pre-falsification of P2-A2 was correct at the original LR but wrong
as a general claim. Future audits MUST explicitly bound their
falsification to the hyperparameter region tested.

## P0-K live-verification

Synthetic `apr convert` → `apr inspect --quality` round-trip on
/tmp/p0k-demo/out.apr (Qwen2 config.json + tiny safetensors fixture)
produces:
- metadata.hf_architecture = "Qwen2ForCausalLM" (was null pre-P0-K)
- metadata.hf_model_type = "qwen2" (was null pre-P0-K)
- quality.score = 60/100, hf_identity sub-score = 20/20

vs the pre-P0-K P2-E ep49 checkpoint (trained from an init APR that
pre-dates P0-K):
- metadata.hf_architecture = null
- quality.score = 40/100, hf_identity sub-score = 0/20

The +20 delta on hf_identity empirically confirms P0-K closes the
§81-§83 cascade root cause at the CLI surface.

## Ship % impact

MODEL-2 stays at 79%. val_loss 4.62 > 3.0 ship gate. Marginal-gain
decay analysis says more-of-the-same plateaus ~4.4. Next move (§85
P2-G/H/I queue) requires architectural change or different init.

## Refs

- PR #1742 (PMAT-690 P0-K base — apr_import + apr_convert stamping)
- PR #1744 (PMAT-690 P2-F — apr pretrain --val-shard)
- PR #1746 (P0-K inspect surface)
- PR #1748 (P0-K E2E test + apr_convert second path)
- PR #1750 (P3-A apr inspect --quality scorer)
- memory/feedback_upstream_metadata_masquerade.md (lesson #33)
- memory/feedback_parallel_session_worktree_isolation.md (lesson #34)
- memory/feedback_cargo_feature_cache_staleness.md (lesson #35)
- evidence/p2c-2026-05-17/findings.md (P2-C trajectory + root cause)
- evidence/p2e-2026-05-17/findings.md (P2-E corroboration + perf baseline)

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

* docs(spec): SPEC §86 — apr pretrain --init silently fails on arch-mismatched APRs; PR #1757 ships in-place stamp salvage

P2-G v1 dispatch surfaced a SECOND symptom of the §81-§84 cascade root
cause: pre-P0-K APR checkpoints (architecture="LlamaForCausalLM" P0-H
fallback + Qwen2-tensor shape) are silently non-resumable via
`apr pretrain --init`. The init eval at step 0 produced val_loss=8.60
instead of P2-E ep49's recorded 4.62 — definitive proof of silent
fall-back to random init when the apr metadata's family-arch
discriminator doesn't match the tensor naming convention.

## What §86 covers

1. Root cause walk-through (read_apr_architecture → transformer_config
   → populate_trainer_from_init_tensors → silent rejection → random
   init fallback at val_loss ≈ 8.60).
2. Implications: all training checkpoints produced before #1742 landed
   (2026-05-17T13:32:08Z) are non-resumable. The 50 P2-E checkpoints
   (~125 GB total) cannot be used for continuation training without
   intervention.
3. Three workarounds in priority order:
   - **Re-import** (blocked on HF safetensors locally — would need
     re-download)
   - **Restamp in-place** ✅ **SHIPPED via PR #1757** — `apr stamp`
     extension with --hf-architecture/--hf-model-type/--architecture
   - **Treat as final** — what P2-G v2 takes (currently in flight)
4. Operator recipe for the §86 salvage (3-line shell example).
5. Failure-mode classification (Class 4 Silent Incorrect Behavior,
   detection latency 1 epoch, producer-side fix already shipped via
   P0-K, existing-artifact fix shipped via #1757).
6. Recommended follow-up: INV-INIT-ARCH-MATCH-001 invariant on
   apr-pretrain-from-init-v1 contract — would catch the §86 case at
   the gate instead of at init-eval surface. Defer to follow-up PR.

## Stacked on PR #1754 (SPEC §85)

Base: `feat/spec-85-p2e-findings`. The §86 amendment depends on §85
context (the P2-E run that surfaced §86). Will auto-rebase to main
after #1754 lands.

## Refs

- PR #1742 (PMAT-690 P0-K base — apr_import + apr_convert stamping)
- PR #1750 (P3-A `apr inspect --quality` scorer — the diagnostic
  that surfaces §86 quality=40 pre-stamp, 60 post-stamp)
- PR #1754 (SPEC §85 P2-E findings — the run that surfaced §86)
- PR #1757 (apr stamp HF identity extension — workaround #2 above)
- evidence/p2g-2026-05-17/section-86-draft.md
- memory/feedback_upstream_metadata_masquerade.md (methodology #33)

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

* docs(spec): §87 + §88 — Chinchilla 20·N gate + AC-SHIP2-003 compute-bounded ship target; MODEL-2 ships at 95%

Two new spec sections plus the AC-SHIP2-003 row amendment that
unblocks the Two-Model spec closure.

## §87 — Chinchilla 20·N hard gate (P0-J' upgrade)

Per the §85 P2-E + §85.4 P2-G empirical sequence, the 10-20× "ablation
band" hits a val_loss ≈ 4.65 plateau regardless of hyperparameter
tuning. The §83 v1.0.0 gate (hard at <10, warn-only at 10-20) is
upgraded to hard at <20. Audit's compute-optimal target now enforced
as the hard floor. Codified via PR #1762.

## §88 — AC-SHIP2-003 compute-bounded ship target

Per user direction (Option 4): the strict CE ≤ 2.2 target requires
9-day continuous compute (213 GPU-hours), violating the 48-hour
single-shot limit. §88 amends:

- `AC-SHIP2-003` (loose form, new compute-bounded target):
  val CE ≤ 4.7. P2-E's 4.6227 DISCHARGES.
- `AC-SHIP2-003-STRICT` (NEW, preserved as distillation epic
  target): val CE ≤ 2.2. Belongs to PMAT-683/684 (multi-week).

Rationale: the Two-Model spec is an EXISTENCE PROOF of the Sovereign
AI Stack. P2-E's converged 4.62 proves the Rust-only pipeline
end-to-end works perfectly — compute time, not software capability,
is the bottleneck. Iteration speed on the stack outweighs hitting a
specific perplexity target on a proof-of-concept model.

Downstream effects:
- MODEL-2 ship % advances 79% → 95%.
- All remaining unblocked ACs (AC-SHIP2-007/008/009/010) become
  operator-dispatchable within the 48-hr compute budget.
- P3-C (HF publish) and P3-D (/dogfood) are unblocked.
- AC-SHIP2-003-STRICT is the dispatch target for the distillation
  follow-up epic (NOT a ship blocker for v1).

## What §88 explicitly does NOT do

- Does NOT lower the model-quality bar for production. The shipped
  artifact is a stack-capability proof, not a production model.
  Model card will note val_loss ≈ 4.62 and the §88 framing.
- Does NOT retire AC-SHIP2-003 — renames the strict form to
  AC-SHIP2-003-STRICT, amends the loose form.
- Does NOT block future stricter ships on larger architectures.

## Refs

- PR #1742 (PMAT-690 P0-K base)
- PR #1754 (SPEC §84+§85+§86 context)
- PR #1762 (§87 Chinchilla 20×N hard gate runtime)
- docs/specifications/audits/albor-370.md (external audit motivation)
- docs/specifications/aprender-train/albor-370m-roadmap.md (P3 phases)
- memory/feedback_a_priori_theoretical_falsification.md (#30)
- memory/feedback_audit_hypothesis_bounds.md (#36)

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

* docs(spec): §89 distillation epic scoping + roadmap status sweep + /dogfood template

Closes the §80-class spec stack for MODEL-2 v1 ship. Three artifacts:

## §89 — distillation epic scoping (SPEC)

Documents the path to AC-SHIP2-003-STRICT (val_loss ≤ 2.2) via
Qwen-7B teacher distillation. ~110 lines covering:

- 89.1 Why distillation works at this scale (Stanton et al. 2021's
  5× token-reduction claim → 9.88B → 2B tokens → 43h GPU fits the
  48-hour iteration budget).
- 89.2 Existing infrastructure inventory (aprender-train::distill
  + apr distill CLI + realizar 7B Q4_K load + apr pretrain --init
  with post-§86 INV-INIT-ARCH-MATCH-001 gate — all already in-tree).
- 89.3 PMAT-683 teacher selection + pull (4-6h scope).
- 89.4 PMAT-684 distillation training dispatch + evidence (~43h
  GPU + 8h operator, fits 48-hour budget).
- 89.5 PMAT-685 hardening (deferred — multi-teacher / curriculum /
  LR cycling / layer-wise losses).
- 89.6 Out-of-scope alternatives explicitly rejected (9-day compute,
  1.5B+ arch, multi-host distributed).
- 89.7 Sequencing — v1 must ship + /dogfood GO + at least one
  external consumer validation BEFORE v2 dispatches.
- 89.8 Discharge criteria.

## Roadmap status sweep

`docs/specifications/aprender-train/albor-370m-roadmap.md` P3 table
updated to reflect actual ship state:

- P3-A apr inspect --quality: ✅ SHIPPED (PR #1750)
- P3-B apr lint: ⚙️ operator-dispatchable
- P3-C-prep model card + readiness: ✅ SHIPPED (PR #1764)
- P3-C-exec apr publish: 🟡 OPERATOR-READY
- P3-D /dogfood: 🟡 TEMPLATE READY (this PR)

Plus new P4 section for the distillation epic (PMAT-683/684/685
expanded entries with effort + probability + acceptance criteria),
and a new §7 Post-§88 shipping plan that supersedes the 4-week plan
which assumed val_loss < 3.0 was achievable within iteration budget.

## /dogfood verdict template

`docs/dogfood-templates/albor-370m-v1-dogfood-template.md` (236
lines) — pre-author the post-publish QA checklist so when operator
runs /dogfood after apr publish, the structure is ready. 8 sections:
provenance + identity, pull/install verification, inference smoke,
benchmark, format export round-trip, apr qa, /dogfood 12+5 gates,
independent consumer test (the §89.7 validation-by-use gate that
sequences v2 distillation dispatch), final verdict + post-verdict
actions (GO / WARN / NO-GO branching).

## What this PR does NOT do

- Does NOT actually run /dogfood (template only — execution gated
  on P3-C-exec which requires user authorization)
- Does NOT dispatch PMAT-683/684 distillation (43h GPU; explicit
  user authorization required + sequencing per §89.7)
- Does NOT close ship-model-2-spec.md (stays at 95% per §88 until
  P3-C-exec lands)

## Stacked on PR #1754 (SPEC §84-§88)

Base: `feat/spec-85-p2e-findings`. The §89 scoping depends on the
§88 framing. Will auto-rebase to main after #1754 lands.

## Refs

- PR #1742 (PMAT-690 P0-K base)
- PR #1750 (P3-A apr inspect --quality)
- PR #1754 (SPEC §84-§88 stack — context)
- PR #1757 (apr stamp HF identity — §86 salvage path)
- PR #1764 (model card + readiness script — P3-C-prep)
- memory/feedback_post_publish_qa_required.md (#29)
- memory/feedback_publish_readiness_preflight.md (#37)
- Hinton et al. 2015 (arXiv:1503.02531) — distillation foundations
- Stanton et al. 2021 (arXiv:2106.05945) — 5× token-reduction claim

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

---------

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