Theseus Codex · Documentation
How the system works
These pages describe Theseus Codex at the level of methodology, algorithms, interfaces, and operator workflow. They paraphrase the firm’s internal guides for outside readers, sanitized of private corpus material, internal credentials, and unreleased work. For the methods themselves and the calibration record, the companion routes are /methodology and /principles.
This documentation describes Theseus Codex's infrastructure and methodology. It does not expose private firm materials, uploaded source documents, or unreleased internal records.
The path material takes
Source → claims → principles → algorithms → public surfaces
Every public artifact on this site reduces to one of the stages below. The vocabulary is worth pinning down before reading the individual guides: an evidence claim is an atomic sentence with provenance, a principle is the durable reusable rule, and an algorithm is the repeatable reasoning function that applies principles to inputs.
- 01Source material
A written document, transcript, or recorded conversation enters the system. Source material is the raw input; nothing is published from this stage alone.
- 02Evidence claims
The source is segmented into one-sentence atomic assertions, each carrying its speaker, span pointer, type, and embedding. (Historically these rows were also called "conclusions" inside the workshop — a legacy term for an evidence claim, not a finished public answer.)
- 03Principles
Clusters of related claims are distilled into reusable principles. A principle is a durable, third-person statement carrying a kind (rule, mechanism, heuristic, and others), a domain of applicability, falsifiable proxies, and a verbatim source anchor.
- 04Algorithms
Repeatable reasoning functions that apply principles to inputs. An algorithm names what it observes, the condition that fires it, the principles it draws from, and the structured output it produces.
- 05Public surfaces
Reviewed methodology pages, Currents opinions, Forecasts, and articles. Each public artifact carries its citation chain back to the principles and claims it depends on.
The seven guides
Documentation index
- 01 · Guide 1 — Quick StartOverviewWhat Theseus Codex is, and how the parts fit together
Theseus Codex is the website face of a research and investment firm whose product is its recorded reasoning. Three programs work together: a desktop recorder that turns spoken conversation into structured transcripts, a background workshop that turns recorded material into a navigable library of claims and principles, and the Codex itself — the website you are reading now, which exposes a reviewed subset of that work to the public.
Read → - 02 · Guide 2 — Knowledge and PrinciplesKnowledge, Evidence Claims, and PrinciplesHow recorded material becomes durable, reusable belief
Everything the firm reads or records is broken into atomic evidence claims, embedded, clustered by meaning, distilled into principles, and — when the firm is willing to commit — promoted into vetted positions that carry a falsifiability layer. The corpus is two passes: one builds the principle library, the other builds the principle-shaped conclusions that cite them.
Read → - 03 · Guide 3 — The OracleThe Oracle and Ask InterfaceCitation-grounded question-answering, with deliberate abstention
The Oracle is the firm's question-answering surface. It does not run a free chat against a foundation model; it answers from the firm's corpus and refuses, or warns, when the corpus is too thin. Every quoted span must appear verbatim in the cited source — if it does not, the answer is rejected before it reaches the reader.
Read → - 04 · Guide 4 — CurrentsCurrentsLive commentary, source-grounded or abstained
Currents is the firm's live commentary surface. A scheduler pulls recent public posts in, runs each through a significance floor and a relevance gate, and either writes a short citation-grounded opinion or abstains. Every published opinion satisfies eight structural invariants designed to prevent free-floating assertion and to keep private material out of public output.
Read → - 05 · Guide 5 — Forecasts and PortfolioForecasts and PortfolioPrediction-market opinions, the decision trace, and the eight gates
The Forecasts surface tracks markets on Polymarket and Kalshi, retrieves relevant firm conclusions, and builds a deterministic decision trace for each market. Live trading requires eight successive human gates; paper mode is the default. After resolution, every prediction is Brier-scored, log-loss-scored, and added to the public calibration manifest.
Read → - 06 · Guide 6 — Operator ConsoleOperator Console and Safety GatesPublication review, signing, kill switches, deletion versus retraction
A small number of actions concentrate in the operator role: publishing to the public site, signing the quarterly methodology review, flipping kill switches, authorizing real-money trades, removing material the firm has cited publicly, and managing roles. The split exists because these actions have real-world consequences, require cryptographic signing, or carry concentrated risk.
Read → - 07 · Guide 7 — What ChangedArchitecture After Round 19Principles instead of summaries, algorithms on top, one contradiction engine
The system was rebuilt around a different idea about what it should store. Earlier, the extractor produced first-person summaries of what an author seemed to be saying. The firm cannot act on that. The pipeline was reshaped so the stored object is a third-person, generalizable principle — and a new layer of algorithms was added on top to turn principles into structured reasoning when their conditions match an observed input.
Read →
What these docs do not contain
The guides describe the platform’s public-facing methodology and architecture. They deliberately omit the firm’s private corpus rows, uploaded source material, internal release URLs, machine credentials, environment configuration, unreleased research memos, and any organization-specific content. The firm’s reviewed conclusions and reasoning live at /methodology and /principles; the operator surfaces they describe are accessible only to authorized roles inside the firm.