─── DOSSIER :: THESEUSUNCLASSIFIED
~/projects/theseus/README
A knowledge system for monitoring ideological contradiction.
CLAIMPhilosophical debate can be systematized.
METHODExtract first principles, represent them as a graph, and automatically surface contradictions between them.
ROLEA Kim/Codex for the operator-philosopher.

── PRINCIPLE GRAPH ──

12 principles · 14 edges · 6 contradictions
P-001P-002P-014P-015P-022P-031P-040P-052P-061P-070P-071P-080

── NOOSPHERE ──

┌ NOOSPHERE · CONTRADICTION FEED ──
2026-04-18T12:04:02Z · DETECTED
· Principle P-014 ("autonomy is prior to welfare")
· > tension with
· > Principle P-071 ("collective survival overrides individual consent in extremis")
2026-04-18T10:33:11Z · OPEN QUESTION
? Under what conditions does P-014 yield to P-071, and does that
? yield imply a new meta-principle?
2026-04-17T22:17:44Z · REFINED
* P-022 rewritten: "preferences over outcomes are revealed
* through choice under constraint."
⎯ viewing 3 of 3 entries · stream live ↗

── FIRST PRINCIPLES (12) ──

  • P-001Knowledge is provisional; revision is a feature, not a flaw.
  • P-002Evidence is weighed, not counted.
  • P-014Autonomy is prior to welfare.
  • P-015Consent is the operator of legitimacy.
  • P-022Preferences over outcomes are revealed through choice under constraint.
  • P-031Stated preferences and revealed preferences diverge predictably.
  • P-040Power flows toward whatever is measured.
  • P-052Metrics, once made consequential, cease to be informative.
  • P-061Institutions outlive the problems they were designed to solve.
  • P-070Replacement is harder than reform; reform is harder than capture.
  • P-071Collective survival overrides individual consent in extremis.
  • P-080A model that cannot be wrong cannot be useful.

── METHODOLOGY ──

Theseus extracts first principles from discussions, uploaded essays, and interviews using a language model fine-tuned on Michael's own working corpus. Each principle is stored with its provenance — the source text and the moment of extraction.

Principles are linked into a graph of supports and tensions. A second model — Noosphere — walks this graph at intervals, flagging contradictions, asking refining questions, and proposing edits to bring sibling principles into coherence.

The system is not a chatbot. It is a slow, structured collaborator for the operator-philosopher who wants to know, at any given moment, where their stated commitments contradict themselves.

ARTIFACTS