Series

AI Orchestration · Model Risk

How models mark, hedge, and spend under Clearance rights

A long-form series on building, gating, and operating AI on a stock-trading desk - control plane, model-build, ops harness, then equity-desk math under the same refuse culture.

Two parts Long-form

How to read Start with Part I (Ep00, then Ep01-Ep02). Skim by plane if you already know the map. Part II is equity-desk math under the same refuse culture - after Part I.

Part I - Control · Build · Harness

Rulers that refuse midpoint peace, as-of honesty, powered eval, book coupling, and shadow / canary / sized rights.

Ep00The CEH-001 situationRead
Ep01Bertrand's Paradox and AI model riskRead
Ep02Fixing AI model risk after Bertrand: one solution stack that actually shipsRead
Ep03Training can succeed and still leave you without a unique decisionRead
Ep04After underspecification: one stack that makes the model crowd visibleRead
Ep05A number between zero and one can still be theaterRead

Open part Start reading

All 36 essays →

Part II - Equity desk math

Discount curves, portfolio programs, stock trees, measure change, and Black-Scholes as contrast. Read after Part I.

Read after Part I.

S2-01Silent rate conventions and the two marks nobody reconciledRead
S2-02Cashflow contracts as as-of hashesRead
S2-03Tomorrow's 10y is already in the tableRead
S2-04Typed jobs before spendRead
S2-05KKT as the proposal auditRead
S2-06Diversification that never hits zeroRead

Open part Start reading

All 20 essays →