New direction — concept for review · static preview, no backend
CRBRL · Rev 2026.07 · Precog Labs
Field notes

Written for the cohort that already knows the jargon.

Ten posts, each aimed at a specific reader — AI infrastructure teams, vector-DB-heavy companies, RAG platforms, agent builders — plus two longer-form pieces. Existing content, restyled into this index rather than rewritten; full posts still live on the current Field Notes page linked below.

§B.1 Series

Field Notes — ten posts, by cohort

#TitleWritten for
01The Byte Budget: The Vector Line Item Nobody ForecastsAI infrastructure teams
02Data-Oblivious Compression: What "No Training, No Codebook" Actually Buys YouAI infrastructure teams
03Compressed-Domain Search: Querying Without Decompressing FirstVector-DB-heavy companies
04Migrating Off pgvector Without Swapping Your DatabaseVector-DB-heavy companies
05Recall You Can Defend: How to Read a 0.98 Fidelity Number HonestlyEnterprise RAG platforms · methodology
06Disk-First vs RAM-First: Where Your RAG Index Should Actually LiveEnterprise RAG platforms
07The Memory Tax on Agents: What Long-Lived Context Costs to StoreHigh-cost LLM / agent builders
08When "Just Add More Nodes" Stops Working: The Scaling Wall at 100M VectorsHigh-cost LLM / agent builders
09A Practitioner's Guide to Benchmarking Vector CompressionCross-cutting · methodology
10The Electricity Cascade: The kWh Hidden in Your Vector IndexCross-cutting · environmental thesis
Links point to the existing Field Notes page — open the standalone HTML from the project folder alongside this local server to read full posts for now.
§B.2 Long-form

Standalone pieces

TitleAuthorFormat
The Machine That Forgets: A Short History of AI's Memory ProblemPrecog Labs, Jul 2026MD
Compressed-Domain SearchPrecog LabsPDF