Fix the scaffold that propagates 'License: PMPL-1.0-or-later' (runbook §7a source fix)

[hyperpolymath]

docs/migrations/pmpl-to-mpl-sweep-runbook.adoc §7a already prescribes the fix; this issue exists to track doing it at the source.

The runbook's instruction, verbatim

Strategy — fix at the source, not 236 times. Locate the scaffold generator (rsr-template-repo / scaffoldia / the llm-warmup-* template) and flip the template's licence line to the correct per-category value, THEN re-propagate. Doing 236 manual edits invites immediate re-drift on the next scaffold run.

Measured in standards today, 2026-08-26

54 files still carry a PMPL declaration. By filename, the template origin is obvious:

File	Count
humans.txt	12
REUSE dep5	12
README.adoc	11
trust.txt	5
everything else	1 each

Many still carry unfilled {{AUTHOR}} / {{OWNER}} / {{CURRENT_DATE}} placeholders — proof they are scaffold output, not authored text. Note the SPDX header on these files is already correct (MPL-2.0); only the body License: line is wrong, so the legally operative licence is right and this is metadata hygiene rather than a licence defect.

Must NOT be touched

• LICENSES/PMPL-1.0-or-later.txt — the licence exhibit text (runbook §1 says "Leave")
• docs/migrations/pmpl-to-mpl-sweep-runbook.adoc — documents the migration
• rhodium-standard-repositories/satellites/rsr-certifier/PALIMPSEST.adoc
• scripts/tests/spdx-header-block-test.sh — a test that asserts on these strings; editing it would break the check rather than fix anything

What to do

1. Find which scaffold emits humans.txt / trust.txt / dep5 / README.adoc carrying License: PMPL-1.0-or-later — candidates per the runbook are rsr-template-repo, scaffoldia, and the llm-warmup-* templates
2. Fix the template's licence line per category (code → MPL-2.0, prose → CC-BY-SA-4.0)
3. Re-propagate, excluding the Rule 2 carve-outs, 007, son-shared repos, and vendored forks — exactly as §1/§3 specify
4. Only then clean the residue

Why this is filed rather than done

I found these 54 files and was about to correct them directly. The runbook explicitly forbids that, and is right to: they are the output of a generator that will reproduce them on the next scaffold run. Recording the measurement here so the source fix has a denominator to verify against.

Related: standards#643 (removed the PMPL preference comments — a different, non-templated surface, already done).

[hyperpolymath]

✅ Source located — and it was already fixed three months ago

Traced per the runbook's §7a instruction. The generator is not the problem.

rsr-template-repo/.well-known/humans.txt now emits:

License: MPL-2.0 (code) / CC-BY-SA-4.0 (docs)

and its history shows exactly when that happened:

2026-05-22  chore: migrate license from PMPL-1.0-or-later to MPL-2.0
2026-06-26  chore(template): doc-link fixes, MPL-2.0/CC-BY-SA-4.0 licence cleanup

Its dep5 is likewise clean (License: MPL-2.0 throughout).

This changes the disposition

The runbook is dated 2026-06-27 — the day after the final template fix — and warned:

Doing 236 manual edits invites immediate re-drift on the next scaffold run.

That risk no longer exists. The scaffold has emitted the correct line since May, so
there is nothing left to re-drift from. The 54 files in standards are stale residue
that predates the fix, not live output of a broken generator.

So the residue is now safely cleanable, and the runbook's step 4 ("only then clean the
residue") is the step we are actually at — steps 1–3 are done.

One loose end at the source

scaffoldia/.well-known/humans.txt reads:

License: MPL-2.0 (Palimpsest MPL)

The licence identifier is correct; the (Palimpsest MPL) parenthetical is a leftover that
contradicts it, since MPL-2.0 is not Palimpsest MPL. Worth correcting so scaffoldia's
output does not reintroduce the confusion the template already removed.

Revised scope

• Fix the generator — already done, 2026-05-22
• Fix scaffoldia's stray parenthetical — small
• Clean the 54 stale files in standards — now safe, was not before
• Note the SPDX headers on all 54 are already MPL-2.0, so the legally operative licence
  has been correct throughout; this is metadata hygiene, not a licence defect

Correcting my own earlier note on this issue, which treated the residue as unsafe to touch.
It was — until I checked when the source was fixed.

[hyperpolymath]

Next step (re-verified 2026-08-27, and the answer the issue was asking for): the scaffold source is scaffoldia, not rsr-template-repo — 6 templates under machine-readable-design/harvested-registry/ emit the defect verbatim (ecosystem/reposystem-tool.ncl:150, elixir/phoenix-service.ncl:468, gitbot/fleet-bot.ncl:439, plus haskell/julia/rescript ones), as the README badge line where the badge URL already says MPL-2.0 and only the alt text says PMPL. Meanwhile PR #648 already cleared most of standards' residue: live count is 12 files, 10 of them inside directories #479 is evicting.

→ Six one-line edits in scaffoldia, then re-propagate per the runbook §1/§3. Small, and now precisely located. (Adjacent, not this issue: 497 files in standards still carry unfilled {{AUTHOR}}/{{OWNER}}/{{CURRENT_DATE}} placeholders.)

[hyperpolymath]

Re-measured 2026-08-31 — confirmed, and larger than the last estimate. The generator source is still defective on scaffoldia main: 6 of 7 harvested-registry templates, 28 PMPL lines total (not "six one-line edits") — every affected template carries license = "PMPL-1.0-or-later" metadata at line 10 plus the mixed badge MPL--2.0-blue.svg[License: PMPL-1.0]; rescript/deno-app.ncl additionally has three SPDX PMPL headers (lines 1/89/126) and a badge URL that is itself PMPL. rust/library.ncl is the only clean one. scaffoldia's .well-known/humans.txt line 12 still reads License: MPL-2.0 (Palimpsest MPL); rsr-template-repo's is clean.

Downstream propagation re-confirmed on remote default branches: 5 repos carry the emitted badge line — ambientops (traffic-conditioner/README.adoc), ViableSystems.jl, ShellIntegration.jl, SiliconCore.jl, SoftwareSovereign.jl. standards' own residue by this issue's declaration metric (License: PMPL, excluding the 4 sanctioned carve-outs): 17 files.

Stays open as REAL WORK. Fix order: templates → humans.txt → regenerate/patch the 5 consumers → standards residue.

[hyperpolymath]

Template-arm landed: PR #835 MERGED (482c1feb) — 1-formats/templates/STATE.a2ml.template:46 now teaches the per-category default and the palimpsest carve-out, and the §7a runbook carries the source-status note (follow-up PR riding). Issue stays open by design: the survivors are the owner-run re-propagation/residue clean and the AI-MANIFEST lines as #643-residue candidates.

[hyperpolymath]

Clarification from the adjacent factory, 2026-09-21: iseriser is not the §7a source. iseriser generates the -iser family (29 repos) and emitted live PMPL licence text until hyperpolymath/iseriser#106 (now MPL-2.0). The §7a surface is body-text License: lines in humans.txt / trust.txt / REUSE dep5 / README.adoc from the rsr template family. The runbook's candidate sources (rsr-template-repo, scaffoldia, llm-warmup-*) remain the right hunt.
One observation for whoever takes this: iseriser's template tokens were fixed by pairing the generator fix with a contractile.just grep-gate, which catches re-drift at build time — the same gate pattern travels well if you want a ratchet on the §7a fix.

[hyperpolymath]

Source located (2026-09-21) — and the fix is open: hyperpolymath/developer-ecosystem#215

Authenticated code-search locator pass over the issue's three candidates:

• rsr-template-repo — clean: no carrier files remain.
• scaffoldia — not the emitter: the six PMPL-bearing files under machine-readable-design/harvested-registry/ are harvest data, and that directory's README already records the caveat: "flagged not fixed (no automated licence edits — ever)" with "owner review needed before any consumer uses it". Per that standing directive I did not flip them. Carriers for the owner's per-file review: julia/package.ncl (named in the README), haskell/stack-library.ncl, elixir/phoenix-service.ncl, gitbot/fleet-bot.ncl, ecosystem/reposystem-tool.ncl, plus rescript/deno-app.ncl (banned-language legacy, harvest-only).
• developer-ecosystem — the emitter. The PR above flips the six source/template lines (llm-warmup-dev/user.adoc, QUICKSTART-MAINTAINER.adoc, ABI-FFI-README.adoc, stapeln.toml, www/.well-known/humans.txt) to the canonical Rule-1 form, exactly per §7a's "fix at the source, THEN re-propagate".

Per the runbook's ordering, the re-propagation (template-sync machinery) and residue cleanup — including the ~139 body-text matches inside developer-ecosystem's own vendored subtrees, which are instances not sources — follow the source merge, with §1/§3 carve-outs excluded. That follow-on is tracked here; the locator question this issue filed is answered by the PR.