The System Atlas

WHAT I've built and how to decide + build the next — the next project chosen by a demand × supply brainstorm, then registered, reviewed, mined for reuse tiers, and scaffolded from them.

Anything interesting becomes durable knowledge — capture one #inbox line from any session, triage to a route, and end every route in the retrieval tail. Three rules: attention is all I need · output is the only thing that matters · retrieval IS the studying. The chief lends the loop four thin commands: /triage · /aim · /card-check · /next — judgment work de-automated from the frozen learn-to-ship agent.

SPEAKING English well enough that language is not the reason I lose an interview — a placed daily rep plus a paid human counterparty who corrects, keeps a cross-session error ledger, and holds the count. The 16th row, minted 2026-08-07 when Felix vetoed the forensics record's "attach, don't mint" default and ruled English a standalone project: deliberately NOT career-owned, because subordinating it to the job hunt is exactly what the 2026-07-11 demotion to "lifestyle slots" produced, ten days before the language rejection. owner_repo: null on the review-cadence / wiki-management precedent — the doctrine lives in the vault, the drill in career, the cards in Anki; no new repo and no new doctrine file, Felix's own scope choice. The 90-day deliverable is one number between 0 and 12: tutor sessions attended, held by the tutor and never by Felix. Wanted, not running — no diagram until it does.

The relationships that carry career, money and emotional life — built and maintained deliberately rather than left to whoever happens to still be in touch. The 17th row, opened 2026-08-09 at triage from Felix's own 2026-08-06 capture, because the sweep found no row covering relationships anywhere in this inventory: a capture that fits no row is the signal the row is missing, not the signal the capture is noise. A deliberate stub — owner_repo: null, entry_points: [] — because the substance is Felix's to supply and no framing interview has happened yet; filling it from the chief's guesses would repeat the failure the English record named, a document written ABOUT him rather than FROM him. Open until that interview: what "connection" means here, whether the money and emotion halves are one row or two wearing one name, what the current state actually is, and whether this is a build project or a maintenance cadence. Not career-owned, on purpose — the same trap English fell into.

Evidence of what I've done — reviewed projects become case studies, distilled into master.md, the only fact source resumes draw from.

WHO I am, in every session — one interviewed master profile, distilled into the CC snippet and its Drive mirror; career facts enter the career repo only via master.md, so no second profile exists to drift.

GETTING the job — scout, match, tailor, and interview against real JDs. Scouting is Felix-run manual (the scheduled agent was retired — scouting is judgment work).

Prompts as versioned assets — engineered once, kept in one library, deployed via thin loaders, never pasted copies. Since 2026-07-20 last-tested is audited, not trusted: the checker warns when a prompt's newest commit postdates its last real session.

Close each day, week, and month deliberately — each tier feeding the next. All three tiers are chief commands now: the evening /daily-review (migrated from Cowork 2026-08-04; the agent-kit prompt retired — one home, no pasted copy), the Saturday /weekly-review (graduated from blog 2026-07-12), the monthly /monthly-review (first ran 2026-08-02/03). Felix-run throughout: reflection is judgment work, de-automated 2026-07-07 — and inside the reviews Felix, never the AI, assigns every day and clock time.

HOW the vault stays coherent — page naming, journal format, the capture pipeline, content ownership. Conventions, not content: a standard, not a step loop, so it runs without a flow diagram. The chief audits from outside (check_page_h1); the fix stays Librarian-run inside Logseq.

WHAT I say in public — drafts refined into native-English posts, curated into the project-led site, one language per page. The site leads with case-study pages (ragx · learn-to-ship) and a curated /writing; published posts are dated snapshots, never derived views.

HOW I come across — a deliberate public image and charisma, built from profile facts, presented across social platforms, and audited outside-in on a cadence. Running since 2026-07-07: the first BUILD session shipped positioning, image pillars, and per-platform bios into brand-records.md. The brand check rides the monthly review (2026-08-03).

WHERE the whole thing is going — every life path (employ, freelance, build, study, relocate) held as falsifiable dated checkpoints that get scored, with stances set by Felix alone. Running since 2026-07-20; the map holds 7 paths and its re-scores are honest: the Toptal screen scored wrong (rejected 2026-07-21, parked default executed), fte-human-contact scored right 119 days early. The layer renamed its own home 2026-08-03: job_hunting → career.

WHAT my free time is for — favorites and reading data distilled into a taste map, tested against unforced-hours evidence in a quarterly audit, and allocated as weekly hours, with verdicts set by Felix alone. The third decision layer, so it shares the anatomy: an options inventory the chief never edits, a Felix-only verdict enum, a self-kill criterion. Running since 2026-07-23: 4,616 items across 7 surfaces, a 15-topic taste map, and the first quarterly audit — 10 verdicts, invest workout 5.5h + english 5h.

WHAT keeps me solvent — money facts owned in one file, every outflow over the threshold decided by memo, and a monthly review that re-asks what "rich" means and appends the dated answer. Life-scope, not work; no daily expense tracking and no alpha. Running since 2026-08-02: runway trio filled, three dated memos decided, monthly review #1 executed.

WHERE files live — one placement policy across Drive, GitHub, and local/external volumes. No entry points, no diagram until designed.

HOW I work with AI — the meta-loop that keeps every flow inventoried, fact-checked, and judged. Listed in its own inventory, because a behavior can only be audited if it is listed; the mechanical checker adopted the shared audit-kernel 2026-08-03. Since 2026-08-04 the row also carries /council — deliberation, on demand: seats with a doctrine and an owned fact base are spawned blind (facts attend as witnesses; only judgment sits), one cross-examination round, then a dissent-preserving synthesis — a decision packet with graded close, flip log, and reopen conditions, never a verdict. First convening 2026-08-04 (own-vs-access); Felix alone decides, and the decision routes to its owning ritual.