Fool: FOOL-SEED-RULE.md §2a corrected two days before the beacon — the heading
claimed a ruling scope the recorded ruling could not reach (d6377af, pushed).
Obsidian: the paper bridge tested and working, first notebook transcription
filed under its own entry date; vault-links.py built and twice caught failing
while its selftest read green; Paradigm + Marshall read in full; the Archive
Convention and Re-encounter placed in the vault; the vault's CLAUDE.md
rewritten after fifteen months stale.
Files: session record + notebook-provenance reference memory + 6 KG lines
(3 drift-patterns, 2 preventions, 1 measured-state); MEMORY.md rotated with
the prior Active Session demoted verbatim to MEMORY-reference.md;
scripts/vault-links.py archive-path fix (substring -> path component, 3 new
controls, 26/26).
No new PENDING items — today's vault work is steward-direct and touches no
governed artifact.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01JQKeKY9T9d95KpvHwwok8T
An Obsidian link census is only worth acting on if it can be trusted, and twice
on 2026-08-23 we were one step from acting on numbers that were wrong: a regex
census that read 52 path-form links as broken, and a fuzzy matcher that offered
to rewrite [[2025-05-30]] as [[2025-08-30]] — different days, not aliases.
Implements Obsidian's resolution order (exact path, basename, alias, then
case-insensitive; ambiguous basenames to the shortest path) over the files
themselves. Not the app's index: that is a 43 MB LevelDB cache which, measured
this afternoon, was 78 minutes stale and held ZERO entries for a note created
the same day (control: 29 entries for a note months old). It also carries a
LOCK — freshest while Obsidian runs, readable only while it does not.
NEVER fuzzy-matches. Date- and week-shaped targets get no candidate at all,
with a positive control proving that is a refusal and not a broken function.
The first run was PASS-BUT-FALSELY and the log records it: selftest green at
18/18 while the live vault reported 8,296 dead targets, of which 7,830 were web
paths and app URLs the tool had no business judging. The fixture contained no
web links, so the selftest certified its author's blind spot. Now 23/23 with
two positive controls proving the new filter did not silence real breakage.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01JQKeKY9T9d95KpvHwwok8T
