Understand performance with consistent methods.
Define what the data can establish before discussing results. Research rules, NAV methodology and exception handling form a traceable record.
Start with a hypothesis. Retain the selection process.
Reports record the investment universe, signal logic, position rules, parameters, code versions, sample periods and cost assumptions. Failed and adverse periods are retained rather than showing only selected winners.
Backtests, historical holdouts, simulations and live results are labeled separately. A holdout that influenced parameters or selection remains historical validation. Independent out-of-sample evidence requires new, unused data after research rules are frozen.
Account scope and fees define net assets.
Each series binds a fixed account and wallet scope and fee policy. Adding or removing an account starts a new series at 1 while preserving the old history. Existing funds in a new account are not profits, and curves from different scopes are not automatically joined.
After disclosed portfolio fees includes trading costs, funding and interest reflected in equity, plus verified accrued management or performance fees. Unknown rates and liabilities are never assumed to be zero. Investor taxes, distribution charges and separate contractual expenses outside the portfolio are excluded.
Accrued fees first reduce net assets. Paying an accrued fee reduces cash and the liability together, without a second deduction or classification as an investor withdrawal. Paid portfolio costs that were not accrued remain in equity profit and loss.
Gaps are not smoothed away.
Missing valuations, uncertain cash-flow times, unverified prices or incomplete fees freeze the affected interval and all subsequent calculations. A successful balance request only confirms connectivity. Publication resumes after evidence, deduplication, recalculation and review. If evidence cannot be recovered, the old series ends and a new baseline is required.
Wallet valuation times may differ by at most 30 seconds. Flows within that band that cannot be aligned require reconciliation. Data is delayed after 15 minutes and interrupted after 60 minutes since the last valid valuation. These freshness states are assessed separately from calculation freezes.
Actual nonpositive net assets, full redemption and approximation failure are handled separately. A terminated series shows its termination event rather than hiding the state behind the last NAV.
Return and risk metrics require consecutive valid samples.
Daily returns use adjacent UTC 00:00 boundaries and require a complete intraday sampling chain. Missing days are not filled with zero and returns are not computed across gaps. Partial months are identified separately.
Volatility and Sharpe require at least 90 consecutive valid daily returns and 100% window coverage. Ninety boundaries give only 89 returns. Annualization uses 365 days and a zero risk-free rate. Annualized return is shown only after 365 days.
Maximum drawdown uses the decline from the NAV peak up to and including each point. Visible chart ranges do not change full-series metrics; a log scale changes only the drawing scale.
One release version. One consistent set of results.
Curves, metrics, monthly tables, quality notes and linked reports share one release_id. The operator builds and reviews a candidate before switching the current version transactionally. The interface reads a pinned version and discards the entire set if responses disagree.
Late records or revisions invalidate affected releases. Withdrawn versions are not returned as valid results. Pages check status periodically and when the tab regains focus. If the status service is unreachable, release status is unknown. Historical versions remain for audit, and rollback rechecks inputs and account scope.
Methods explain results; they cannot remove risk.
Assess sample periods, costs, capital conditions and failure scenarios together.