🎯 ONE SHOT — L-MR-L3-ALT · Honest matrix repopulation via JSONL extension (Neon-free path)

[gHashTag]

👑 Queen ruling — parallel-strategy ALT for L-MR-L3 (Neon-blocked)

Original L-MR-L3 sub-lane is BLOCKED on Neon connector quota recovery (~2026-06-01). Rehearsal #2 deadline = 2026-05-25 (17 days). Neon path will NOT make Rehearsal. This ALT lane runs in parallel — whichever lands first wins.

Mission

Repopulate assertions/matrix_samples.jsonl from current 20 rows → N≥100 honest rows, without Neon dependency, so the new honest closure_gate.py (merged in #570 → eeaf0db7) can render an R7-compliant verdict on the PhD Coordinate Matrix (#446 erratum context).

Storage decision (Queen-locked)

JSONL extension — the existing assertions/matrix_samples.jsonl artefact stays canonical. Append rows, do NOT migrate to SQLite/CSV.

Why JSONL (variant C):

• ✅ R10-cleanest — single artefact, no new format
• ✅ Git-friendly streaming diff, grep-able
• ✅ closure_gate.py already reads JSONL (no schema migration)
• ❌ SQLite (variant A) — checked-in binary, harder review
• ❌ CSV (variant B) — loses nested fields, requires schema flattening

Schema (fixed, mirror existing 20 rows):

{"format":"<str>","algo":"<str>","seed_phi":<int>,"step":<int>,
 "min_bpb":<float>,"initial_bpb":<float>,"workflow_id":<str>,
 "sha":<str>,"wallclock_ms":<int>,"ts_utc":"<ISO8601>"}

Acceptance (R7-witness)

• N ≥ 100 rows in assertions/matrix_samples.jsonl (current 20 + ≥80 new)
• ≥ 2 distinct seed_phi values per (format, algo) cell (current: only 47)
• MAX(step) ≥ 3000 per cell (current: many at 0)
• wallclock_ms > 0 for all new rows (R5 honest — no synthetic stubs)
• Forbidden seeds {42,43,44,45} excluded — use {47, 89, 144, 123}
• CPU-only training (no GPU) — per project mandate
• Per-row sha pinned to commit on main at training time
• closure_gate.py (HEAD eeaf0db7) re-runs against new file → emit R7-honest verdict comment on 🌻 EPIC — E2E TTT Pipeline O(1) · Ring-Pattern Refactor (3 GOLD × ~19 SR) #446

Sub-lanes (parallel-claimable)

Lane	Target	Owner skill
L-MR-L3-ALT-T1	training driver — CPU-only loop generating 80 new rows across (format, algo, seed) grid	trainer-igla-sot
L-MR-L3-ALT-T2	append + integrity validator (scripts/validate_matrix_samples.py — schema, no-duplicates, threshold pre-check)	coq-runtime-invariants
L-MR-L3-ALT-T3	run new closure_gate.py against repopulated JSONL, post verdict comment on #446	phd-monograph-auditor

T1 → T2 → T3 sequential (T1 produces data, T2 validates, T3 reports). T1 itself can split into format-axis × algo-axis parallel sub-claims.

Anti-strategy (forbidden)

• ❌ Reopen 🌻 EPIC — E2E TTT Pipeline O(1) · Ring-Pattern Refactor (3 GOLD × ~19 SR) #446 before T3 verdict lands — content-less reopen = noise, R10 violation
• ❌ Migrate matrix_samples.jsonl to SQLite/CSV — Queen-locked storage decision
• ❌ Synthetic stubs (wallclock_ms=0, step=0) — R5 dishonest, what we are fixing
• ❌ Reuse forbidden seeds {42, 43, 44, 45}

Parallel relationship with original L-MR-L3 (Neon path)

Original L-MR-L3 stays OPEN (BLOCKED on Neon quota). When either lane reaches T3 verdict, the other auto-supersedes — closer of L-MR-L3-ALT or Neon-recovery wins. No queue, no waiting.

#446 reopen criterion

When L-MR-L3-ALT-T3 lands honest verdict (any direction — victory holds OR victory falsified), THEN queen reopens #446 with content-delta justification. Until then, #446 stays CLOSED with Erratum marker visible.

Anchor

phi^2 + phi^-2 = 3 · DOI 10.5281/zenodo.19227877
R5 honest · R7 witness · R10 atomic · CPU-only · Forbidden seeds {42-45}

[gHashTag]

🚨 Seed canon erratum

This issue body references SEEDS={47, 89, 144} — incomplete canon. Full canon is {47, 89, 144, 123} (2 Lucas + 2 Fibonacci). Original body left unchanged (R10: history immutable).

Resolution tracked in #600 L-SEED-CANON. Running Railway service hot-patched 2026-05-08T16:34:01Z.

Anchor

phi^2 + phi^-2 = 3 · DOI 10.5281/zenodo.19227877

[gHashTag]

Superseded by #588 L-MR-MATRIX-PRIORITY

This duplicate was opened ~1 min after #582 (sibling L-MR-L3-ALT). Strategy evolved to 50-cell priority subset (#588) — smarter than the N≥100 plain plan specified here.

Canonical work lives in:

• #588 L-MR-MATRIX-PRIORITY (DONE via #589)
• #597 L-MR-POSTRUN (retrieval path)
• Railway service trios-mr-priority-runner (running, hot-patched to canon SEEDS=47,89,144,123)

Closing as not-planned per R10 housekeeping. No work lost — strategy superior.

Anchor

phi^2 + phi^-2 = 3 · DOI 10.5281/zenodo.19227877