Forward evidence contract

Rules

Estimate eligible. At least 252 finite non-constant returns from consecutive published daily marks, plus every provenance gate passing.

Target observed. The annualized point Sharpe is at least the governing forward_sharpe_target; this is not statistical establishment.

Target statistically established. At least 756 returns and PSR probability that the true annualized Sharpe exceeds the governing forward_sharpe_target of at least 0.95, plus every provenance gate passing.

Drawdown. Realized live maximum drawdown is always published. A simulation may be published only with p95, exact configuration mapping and explicit failed establishment dimensions. The older 14-sleeve daily study applies a 10% book-level volatility target and is not an estimate of the current four-sleeve fixed-weight composite. The zero-drift current-composition study maps the exact weights and strategic overlay and may be compared with the 11% objective, but its post-2022 common window and omitted constituent instrument, execution-gap and ladder state mean neither expected value nor p95 establishes live expected drawdown.

Diversification. The exact current-composition research study must publish average sleeve correlation, its synchronized block-bootstrap upper 95% bound, the maximum ordinary pair and its upper bound, the stressed design value, effective sleeve and diversification-ratio diagnostics, and fixed-to-cash marginal historical book-Sharpe deltas. Every threshold comes from the sleeve-admission contract. A positive average correlation fails the non-positive point gate even when described as near zero. Research-window evidence never establishes live-forward diversification and never retroactively adjudicates an existing sleeve.

Configuration change. A live-configuration change that re-sizes the traded book starts a new evidence epoch; pre-change and post-change returns cannot be silently pooled. The current epoch starts on the date of the latest config/live_change_contract.json change_log entry whose contaminates_forward_record is true (a coverage extension declares false and starts no epoch); the returns before it are published as a prior epoch beside the current one and never added to it. Realized maximum drawdown stays descriptive over the whole record.

Missing marks. Missing calendar dates are not filled with zero returns. The continuity audit is a separate mandatory gate.

Public chain. The current paper state must exactly equal the canonical payload disclosed in the signed transparency-chain head. Link continuity, every Ed25519 signature, every disclosed payload hash, and the prospective disclosure boundary must validate. Historical opaque commitments remain explicitly incomplete and are not reconstructed.

Crypto position attribution. The latest locally simulated crypto paper cycle must be non-empty, persist complete order-book marks for every open position, independently reproduce each stored market value and unrealized P&L from quantity, mark, and entry price, and reconcile cash plus marked market value to account equity within the frozen tolerances. A flat cycle cannot vacuously certify the mark path. Entry-price fallback marks, non-finite values, non-positive prices, missing prospective fields, position-count mismatch, stale cycle coverage, position-arithmetic residual, or equity-reconciliation residual fail provenance closed. Historical rows are never backfilled.

Crypto position attribution rollout. A hash-bound deployment receipt and the frozen first later non-empty natural cycle must prove the mark-aware runtime operated on the sole-writer Frankfurt paper host without a forced cycle. Before deployment, missing receipt evidence is published deterministically and fails closed without querying Frankfurt. A successful first-cycle receipt is immutable for that exact deployment receipt.

Target change. The governing forward Sharpe target is read from config/owner_goals.json and every change is appended to forward_sharpe_target_history with the record's status on the day it was superseded. A target change is an evaluation change, never a trading change: it restarts no forward-evidence epoch, pools no returns, and regrades no earlier evaluation. The record is always evaluated against the target in force at evaluation time.

Realized drawdown bound. The owner's realized maximum-drawdown bound is published beside the modeled expected-maximum-drawdown objective and never mixed with it. Realized drawdown inside the bound to date is descriptive; it does not establish that future drawdown stays inside the bound, and the mechanism that acts on it is config/drawdown_control_contract.json.