Skip to content

DR-115: Authored decision surfaces are not machine serialisations

DR-115: Authored decision surfaces are not machine serialisations

Section titled “DR-115: Authored decision surfaces are not machine serialisations”

DR-112 tiered the docs-site mine and put *.html in its T4 “machine serialisations anywhere in the docs tree” bucket, beside *.json, *.jsonl, *.yaml and *.yml.

That grouping is a category error. The lavish boards are authored argument surfaces, not machine output: corpus-reframe-review.html is named in reference/platform-context.md as THE corpus-model authority, and the reports/ boards are among the densest ratification records the project produces.

The error was latent while the boards lived in canonical/.lavish (gitignored, mined only via --include-ignored). It became live at S521, when twelve boards were rehomed to the docs-site: rehoming moved them from a surface that mined them to one that excluded them, so they fell out of the palace entirely. The S517 constraint — “MemPalace = what was argued” — makes that the wrong outcome.

The docs-site *.html exclusion carries two narrow re-includes, for src/content/docs/reports/*.html and src/content/docs/**/okf-platform/*.html. Genuine machine or build HTML anywhere else in the tree stays excluded.

Consequently, rehoming a gitignored working artefact is a three-step operation in a fixed order: rehome, re-mine, then prune the old drawers. Pruning before the re-mine destroys the only copy in the palace.

  • Drop the *.html exclusion entirely. Simplest, but re-admits any build output or generated HTML that later lands in the tree — the thing T4 exists to stop.
  • Keep the exclusion and retain the canonical/.lavish drawers as the palace’s only copy. Preserves content but leaves 379 drawers permanently pointing at deleted paths, which every future orphan sweep re-flags and which makes id-411 {411.1c}‘s steady-state cadence unable to ever run clean.
  • Prune and accept the loss. Rejected against the S517 constraint.
  • Twelve boards (506 drawers) are recallable from their docs-site home; the 379 canonical/.lavish orphans were pruned after the re-mine verified 1:1.
  • The negation works here where S519’s did not, and the distinction matters for future edits to this block: the whitelist-idiom warning at its head concerns directory exclusions, where gitignore never descends into an excluded directory and the negation is unreachable. These are file patterns, so no directory is excluded and the negation is reachable. Verified against GitignoreMatcher before landing, both directions.
  • The /handoff scratch gate (Step 2d) is extended to .lavish/ and now states the rehome → re-mine → prune ordering.