Skip to content

Summary

The EV contract operationalizes the council-amended owner rules from the charter: every expected-value claim in a playbook must link an Attested Computation, use real-world (not risk-neutral) probabilities with a named model, include explicit costs, report a baseline comparator, and state the regime context at evaluation time. A playbook without a passing EV run is not tradable — full stop. The books justify the edge qualitatively (e.g., the variance risk premium in premium-selling strategies); the EV contract makes each edge claim quantitatively testable before capital is exposed.

The Five Clauses

# Clause Requirement Failure mode it prevents
1 Attested computation EV arithmetic runs the blessed code linked from an Attested Computation concept in /85-computations/index.md (runtime: python); no improvised EV in prose LLM arithmetic, non-reproducible numbers
2 Real-world probabilities Probabilities are physical (objective), not risk-neutral, and the probability model is named (historical conditional, GARCH/HAR forecast, …) Silently assuming the risk-neutral measure is a pricing measure, not a forecast
3 Explicit cost model Commissions, fees, slippage/bid-ask spread, and borrow/assignment costs stated numerically — the books ignore these entirely Paper edges eaten by transaction costs
4 Baseline comparator Report the IV−RV (variance risk premium) spread and a GARCH/HAR-class vol forecast; positive EV without beating the baseline is not edge Claiming as edge what is simply mean vol exposure
5 Regime context State the regime label at evaluation time and link /60-regimes/index.md concepts Applying a playbook outside its regime of validity

How a Run Works

  1. A playbook draft references its EV concept; the EV concept references the attested code and input data slice (options bars per data infrastructure).
  2. The deterministic stage of the Hermes loop executes the code and records an executor receipt. The attester (references/attesters/ev-binding.py) re-executes the computation on the receipt's inputs and requires the recorded outputs to match — a receipt with fabricated numbers fails attestation.
  3. Output: EV per contract, cost-adjusted EV, baseline comparison (IV−RV spread, GARCH/HAR forecast), regime tag, and a pass/fail verdict.
  4. Only pass verdicts make the playbook tradable in the current regime; a regime change re-opens the question.

Worked Skeleton

playbook:      45-dte SPX iron condor (7150/7275 P · 8075/8200 C)  [ILLUSTRATIVE — real rehearsal receipt]
EV concept:    /85-computations/ev-iron-condor.md
receipt:       /references/receipts/2026-09-09-condor-dress-rehearsal.json (attestation: pass)
P model:       GARCH placeholder forecast 11.0% vs RV window 9.52%   [PLACEHOLDER — not fitted]
Costs:         $0.65/contract/leg fees; 15% of credit slippage
Baselines:     baseline_ev_rv = 18.48 ; baseline_vrp = +6.33 pts ; edge_vs_rv = −5.75
Regime:        calm-contango-vrp+ (/60-regimes/index.md)
Verdict:       EV gate FAIL — edge_vs_rv < 0 → candidate declined (see 96-journal/2026-09-09.md)

The example is the actual day-1 rehearsal run: a positive ev_net (+12.73) that the edge test still declines. Note both placeholder inputs — a real fitted forecast replaces them at V1.

Links

References

  • Open Knowledge Format v0.2 specification — EV contract and §10 attested computations: ../OKF_SPEC_v0.2.md
  • Carr & Wu (2009), "Variance Risk Premiums," Review of Financial Studies 22(3) — the IV−RV variance risk premium as baseline comparator.
  • Bollerslev (1986), "Generalized Autoregressive Conditional Heteroskedasticity," Journal of Econometrics 31 — GARCH-class volatility forecasts (clause 2/4).
  • Corsi (2009), "A Simple Approximate Long-Memory Model of Realized Volatility," Journal of Financial Econometrics 7(2) — HAR-RV forecast model.

Source Notes

Amendment Log

Version Amendment Status
v0 edge_vs_rv (baseline comparator) as the EV gate retired 2026-09-17 — algebraically vacuous
v1 Amendment 2026-09-17/01 — layered gate: L1 ev_net > 0 (never-rule, ev_gate in attester verdict) in force; L2 VRP-size ≥ θ pending calibration; L3 forecast-SE guard pending se_forecast; L4 conditional-vs-unconditional pending backtest harness L1 active (human:yasu, 2026-09-17); L2-L4 pending

  1. OKF v0.2 spec, EV contract section (sources[]: okf-spec). ↩