2026-07-26 Atlas Notes Convention Audit and Triage
What I set out to do
Audit all 145 notes in Atlas/Notes/ against the atomic-note convention (README 0017: claim-as-API titles, 50-250 word prose, 0-3 theme tags) and triage the ~60% that didn’t fit.
What I actually did
Full audit, then triage by bucket. Roughly 68 notes left Atlas/Notes/:
- 52 to
Atlas/References/: the CSS/design study pack, the ten dash-titled algorithm study notes, the Citadel interview set, GSSAPI/Negotiate research, personal specs (clothing, color, parking, insurance, choreography), and the oldGraph Algorithmsnote, renamed to LeetCode Tree Problem Error Patterns since its atomic insights had already been extracted in 2025. - 5 to
Chronicles/: restaurant reviews (2024-11-08 Hojokban, 2024-11-10 John’s of Bleecker St), 2024-09-15 Experience Using UV, the cryptic 2024-11-21 TS Impactening meeting log, and 2026-04-01 Claude Code Spend Q1 2026. - 3 to
Atlas/Sources/: two podcast voice-memo notes and PyData NYC 2024. - 5 seedlings back to
Inbox/, 3 empty/derivable stubs deleted (CSP Concepts,Jupyter,Nix - Home Manager). - Kept with repair: whisky trio de-headered, Mega Backdoor Roth recased, React Components Are JavaScript Modules and Notebook V7 Rebuilt the Classic Experience on JupyterLab retitled to claim form, Alpha in Finance rewritten from encyclopedia headers to prose.
- New Graph Algorithms MOC weaving the five extracted search-strategy notes plus the study references; the dangling
[[Graph Algorithms]]link in Dijkstra Algorithm now resolves to it correctly.
All moved notes got normalized frontmatter (type-tag first, ≤3 tags, dates filled). Backlinks were checked before every rename; only three needed patching.
What was striking
Pleasant surprises cut both ways: Dimensions of LLM Quality and Embedding Models looked like topic dumps from their titles but were excellent prose notes, while Graph Algorithms looked like a MOC candidate but turned out to be an already-mined husk. Title-level auditing misclassifies in both directions; reading before disposing was worth it. Also: case-only renames on APFS need a two-step move.
Follow-ups
Lightbulb Standards(References) self-describes as unverified ChatGPT output; verify or delete.- The four term-titled CSS notes kept in
Atlas/Notes/(Custom Properties, Stacking Context, Design Tokens, Feature Detection) share the study-pack body style; prose-ify opportunistically. - Possible merge: LLM Pairwise Preference Judging and LLM as a Judge for Preference Annotation overlap.
Related
2026-07-26 Coffee Brewing Notes Restructure, Graph Algorithms, LeetCode Tree Problem Error Patterns