Arc Reactor

1. SourcesChatGPT, posts, media, notes, future imports
2. Corpusnormalized items, chunks, dates, provenance
3. Evidencesearch, relationships, candidate memories
4. Reviewhuman approval and public-context decisions
5. Publishverbatim excerpts, refs, static pages

origin / 2026-04-09

best way to turn this personal website into like lowkey ai version of me? i want it to be user facing and internal.

twenty minutes later

i think this should serve as a 2nd stomach or 2nd brain combined with jarvis/alfred/operator/cortana/etc and also a dashboard as well like apalantir style monitor the situation thingy.

first important failure / 2026-04-11

one problem rn is its interpreting my retweets as my own actual tweets which is lowk an oopsy

The mistake is structural, not cosmetic: something Aaron retained, liked, reposted, or argued with cannot silently become something he said. Authorship and interaction type become first-class fields.

prototype rule / 2026-04-11

i mean idgaf about ugliness rn. thats how laser focused i am. so like thats just cosmetic and idgaf and if its not necessary its probably done lazily

what exists / 2026-07-29

• Source imports and normalized corpus itemsrunning
• Keyword, vector, and hybrid retrievalrunning
• Evidence neighborhoods, threads, media, and annotationsrunning
• Candidate → approved memory lineagerunning
• Period portraits and memory deskrunning
• Scoped Agent API with read-only and proposal-only authorityrunning

publication boundary

The browser never receives an Arc credential and never queries the private API. The checked-in registry contains only selected excerpts, stable evidence references, dates, authorship and fidelity labels, and the context required to publish them without pretending a retrieval model wrote Aaron’s life.