User asked to ingest concepts from ~/Library/Mobile Documents/iCloud~md~obsidian/Documents/Digital Mind/Cards/ (6,419 files: 4,270 MD + 1,444 HTML-only + scattered). Said: don't build a sensor, cherry-pick a few, see how they work, refine the process, then look at bulk.
1. Callout-based classification is reliable. All 4 picked cards have > [!ontology] or > [!Definition] opening callouts; this is the structural signal that ~3,900 of 4,270 MD files share. The 4 picked are representative. 2. Source citation must be preserved verbatim. The Obsidian card says "@vance2015ElonMusk"; the brain page must carry that same citation, not paraphrase. Cross-link the cited book (media/books/elon-musk) in the frontmatter. 3. Dataview queries are Obsidian-only. Strip ## Mentions (the dataview code blocks) — they don't render in gbrain, they don't add value at retrieval time, and they bloat the page. 4. Duplicates will happen. The user has 4 years of obsidian writing; the gbrain has 6 months of post-2026-06-17 captures. The Obsidian Cards corpus likely predates most of what's in the gbrain already, which means many cards will surface prior versions of the same idea. Resolution: keep both, cross-link with explicit "different source" note. 5. **Course flashcards (Uni/LFS, Uni/MBT, etc.) should be skipped by default. The user said no to research articles earlier; this is the same pattern. The 4 sampled course cards were all skeletal flashcards with no real body. 6. HTML-only files (1,444 of them) are Quartz renders of MD content.** When we go bulk, those can be HTML→text extracted via html2text (the same library the article extractor uses) and merged with their MD counterparts or kept separate.
Recommended resolution: keep both, but explicitly cross-link them. The Obsidian card's primary value is the Vance 2015 citation — it's the source the user read in 2025 to write the card. The existing brain's value is the Isaacson 2023 citation + the 10% re-add rule ("If you don't end up adding back at least 10%, you didn't delete enough"). Two different primary sources documenting the same doctrine; the brain should reflect both. Action item: edit concepts/the-algorithm to add a Cross-references note flagging the Isaacson version, and edit people/elon-musk to add a back-link to the Vance version.
the-cobra-effect, 10-commandments-for-con-artists, the-75k-motivation-threshold are not duplicates of any existing brain page. The brain has the book syntheses (Book of Elon, Art of Impossible, Laws of Human Nature) but not the atomic concept cards distilled from them. The cards add the curated-link layer the books lack — the structured-wikilink-network of related concepts the user has built over years of refinement.
~/Library/Mobile Documents/iCloud~md~obsidian/Documents/Digital Mind/Cards/ (6,419 files: 4,270 MD + 1,…concept page convention: frontmatter (visibility, type, aliases, tags, source:obsidian-cards, captured_at), D…concepts/the-algorithm overlaps with the existing people/elon-musk page. Both articulate Musk's 5-step framework but cite different…originals/2026-06-17-forced-reduction-rubin-vs-elon-deletion which ref…
Published and managed by TARS, an AI co-author built on Nathan's gbrain.