Advisory review: "Opening the Door" — field log corrections, retitle, and portable enforcement

[mparrett]

Advisory PR — review only, do not merge. Per the convention from #1/#5/#6/#7/#8, the base branch pre-revival-log is deliberately frozen at the prior phase's tip (bf3c4e5) so this phase reads as one diff. Merging would fast-forward the pin and destroy the "before" snapshot. Close after review.

The arc

Norman Nunley opened nnunley/fmpl#1 against the published field log, correcting two factual claims — as a PR against our words rather than a note asking us to change them. This phase applies both, and takes the log apart to find out what else was wrong with it.

1. The two corrections (9bc8959), in his framing.

   • §2 said "months later, nobody has needed anything back from the archive." It was days — and agent-era feature work off the same lineage (the ITER-0005 persistence arc: durable TupleSpace, content-addressed recovery, bytecode persistence) has since been grafted onto main. The scaffolding was disposable; the code beneath it was not.
   • §7 said the loop "never had a reviewer, so nothing distinguished roadmap from rot." Wrong: review and roadmap lived inside the loop's own process layer — PAR adversarial audits, a dated iteration log, a behavior-scenario corpus — illegible to a reader who never opened it. The surface-legibility point survives; the claim doesn't.

2. Retitle and re-tone. "The Revival, Revived" → "Opening the Door." The rehab framing read as a verdict on someone else's project, which was never the point — a fork arriving at a working repo isn't a rescue. The file moves to docs/fmpl-revival-log.html; fmpl-rehab-log.html stays as a redirect stub so the published URL keeps working. Cross-links updated in index.html, the tour, field log 2, and README.

3. A field note on the correction (§7). The §7 mistake is the most generalizable thing in the log, so it stays visible rather than being quietly patched: we read the absence of our process as the absence of a process, while holding the evidence — §2 moved those files with our own hands.

4. Lineage fix. ColdStore, not cool++/CoolMUD, is the reference for the MUD-era bio (tour §1, README). Research docs citing ColdMUD as an architectural influence are untouched — that part was never wrong.

5. Portable invariant enforcement (10c733c, specs/portable-invariant-enforcement.md). The archaeology behind the §7 correction surfaced a real forward-looking cost: invariants the loop enforced in code survive here as prose in AGENTS.md. The spec audits what is actually enforced and concludes the baseline is much stronger than AGENTS.md implies — CI plus build.rs cover most of it. One conversion worth doing (parser-epoch bump detection), two one-line fixes, and an explicit list of what no repo-resident mechanism can enforce. Git hooks are rejected as a tier: jj skips them.

Verification

• Stale counts corrected against the tree: 74 ignored tests (the log and known-gaps.md still said ~185/~75), seven workspace crates (fmpl-wasm was missing from DEV.md and the tour), pattern bucket 54
• Every internal href across docs/*.html resolves; no reference to the old filename survives anywhere in the repo
• just test green; the only Rust changes are two comments

[mparrett]

Added a sixth item to this phase: docs/fmpl-process.html — "The Layer Underneath" (8f878ed).

The §7 correction in the field log names the loop's process layer but doesn't show it. This page does, from primary sources on archive/agent-harness: the requirement lattice (EPIC → STORY → AC → SCENARIO → ITER), the hook-enforced state machine and what each state blocks, PAR as the reviewer of record, the ITER-0004b reconciliation where a "final step of X" commit is checked against X's acceptance criteria and found wanting, and the memory system.

Gate hooks are quoted from 2a6db48 rather than the branch tip, where they are symlinks into /Users/ndn/development/ralph-loop/ — the page says so instead of inviting readers to open files that aren't there.

Counts verified against the tree: 31 epic documents, 100 stories, 331 acceptance criteria (246 tagged seam:integration), 88 scenarios, 22 logged iterations across 2026-05-08–13, 253 commits from 2025-12-21, 1,648 episodic entries yielding 4 graduated lessons.

Linked from the landing page, both field logs, the tour, and README. Validated in-browser: no horizontal overflow at 1200px or 390px, every TOC anchor resolves, zero console errors, light and dark both checked.