1. Point-in-Time OOS Replay
Causal multi-asset research replay beginning 2023-01-01. A decision can affect only a subsequent interval. Future information is not admitted into the decision set.
METHODOLOGY · VERSION 1
Performance methodology
Neutral technical conventions governing the reviewed historical replay and the append-only forward record. Strategy logic, PAS formulas, and Authority allocation law are outside this publication layer and are not redesigned here.
2. Frozen Live Forward
Forward-only model portfolio evidence after the public Generation-1 freeze timestamp 2026-09-30T14:35:19Z, anchored to backend main commit 3a25a6644f0d3066c30f2749e2d15e21ec32399a. The earlier 2026-09-09T11:07:56Z causal-policy boundary remains provenance, not the public live-return start. Historical rows are immutable and chained cryptographically.
3. Separation
OOS and Live Forward are different evidence populations. They are not spliced, compounded, or labeled as one continuous live track.
4. Frequency & first return
Publication follows completed weekly Authority/effective intervals. The first live return begins only between two consecutive post-freeze READY weekly snapshots; initial normalized NAV is 1.000000.
5. Benchmark
BTC Buy & Hold uses the same interval endpoints as the model and is normalized at the same starting point.
6. Execution friction
Both gross and a 10 bps turnover-stress path are published. The stress path is a sensitivity, not observed brokerage or client execution cost. It does not prove the absence of overfitting and does not represent all venue-specific slippage, spread, funding, market-impact, or implementation costs.
7. Sharpe & Sortino
Sharpe uses a 0% risk-free assumption. Sortino uses a 0% target. Weekly returns are annualized with the documented weekly convention.
8. Maximum drawdown
Maximum drawdown is the minimum of NAV divided by its running peak minus one, recomputed from the same equity curve exposed to the public renderer.
9. Missing data
Required evidence failures are fail-closed. The affected interval remains DATA_GAP or PUBLICATION_PENDING; no future price, fabricated quote, or silently shifted effective timestamp is substituted.
10. Delisted assets
Frozen asset identity and provider lineage are preserved. A missing required valuation cannot be replaced by a convenient future survivor merely to keep the track continuous.
11. Stablecoins
The canonical Cash / Stablecoin bucket is modeled at zero return in this methodology generation unless a future version explicitly changes that convention.
12. Tokenized metals
Forward valuation follows the canonical sealed provider-identity policy already used by the PAS live-forward evaluator; symbol-only substitution is not permitted.
13. Data revisions
Objective source-data corrections create append-only correction records referencing the original hash. Existing records are not overwritten.
14. Model changes
A material model or strategy change is not a correction. Generation 1 freezes the model SHA-256 and strategy-critical PAS/Authority source fingerprints; drift fails closed and requires a new model generation, freeze manifest and commencement date while prior evidence remains available.
15. Hash-chain verification
Each ledger record is serialized canonically, hashed with SHA-256 and linked to the previous published record hash. The verifier recomputes the chain, timestamps, freeze boundary, interval uniqueness, model generation, frozen model hash and strategy-source fingerprint.
16. Limitations
The forward series is a model portfolio record, not audited fund performance, actual client P&L, realized investor return or evidence of custody/execution. Historical replay remains research evidence.