Causal z-scores and the overshoot problem
Define a prior-window reference, measure absolute recovery and avoid calling every correctly directed move a successful reversion.
Reconcile earlier reference windows, original calls, unresolved positions and costs before accepting a mean-reversion performance claim.
Educational diagnostics, not a signal service. A standardized price displacement is not a calibrated probability; invented examples and user-supplied observations are not verified market performance.
A mean-reversion record might describe a research signal population, a simulated strategy, a founder's account or a copied subscriber account. Begin by naming that object and its period. A statement from one account cannot establish the experience of every subscriber. A collection of successful price recoveries cannot establish a portfolio return without quantities, exits, costs and capital accounting.
Request the original source and its revision history rather than a cropped result image. Identify the instrument, price convention, reference rule and opportunity definition. A seller may use a moving average, a spread or a model-estimated equilibrium; those are different objects. Write the claim in a form another reader could reject if the evidence contradicts it. If the seller will not identify the population or provide lawful access to records, keep the assessment incomplete rather than assigning a confident numerical grade.
Suppose an invented rule uses the prior three completed closes 100, 102 and 98. Their mean is 100. A newly observed close of 94 has displacement -6 against that earlier mean. Including 94 in the reference window changes the object being measured; using later closes changes it again. Save the exact contributing observations, time they became available and reference version for each opportunity.
If the rule standardises displacement, also record variance definition, warmup and zero-variance handling. A zero variance is not permission to report an infinite-strength signal. Missing bars need an explicit continuity policy rather than silently shortening a required window. Compare the reconstructed reference with the original message, not just a later chart indicator. An autoregressive software example can clarify model specification, but it does not independently verify the seller's proprietary reference or historical data availability.
Follow an original event identifier from publication to access by the relevant subscriber tier, then to any amendment, cancellation and exit. Preserve messages in their original order. A later edit to the entry or target must be visible as a changed instruction, not incorporated into the first message. A timestamped commitment has only the field scope of the bytes actually committed.
Distinguish publication evidence from execution reports. A research entry can use a declared observation-window convention while an actual broker fill uses the broker's reports; neither should be relabelled as the other. Record whether a claimed price was tradable, selected within a hypothetical window or merely a reference observation. Where no fill evidence exists, an event trace can verify message history but cannot certify account profit. Review one complete favourable event and one adverse or unresolved event before drawing a population conclusion.
Using the unchanged invented reference mean of 100, a starting close of 94 is six points away. A next close of 98 is two points away and satisfies a strictly-closer test. A next close of 108 moves upward but is eight points away, so it fails that same test. A next close of 106 is equally distant and also fails a strict inequality. Direction alone is not the declared outcome.
Keep this reference test distinct from the trading exit. A price that becomes closer can still produce a loss after spread, costs or a different fill, and an overshoot can have a positive long-trade outcome. Save both labels when the study needs both questions. Do not revise the recovery definition after seeing which overshoots improve a win rate. The three cases are correctness fixtures with invented prices, not market observations proving that a rule mean-reverts.
Count opportunities, issued calls, abstentions, cancellations, unresolved positions and completed exits. State whether simultaneous signals share one position or create several units of exposure. A summary containing only closed trades can conceal losing open positions or repeated calls attached to one market move. Explain how each identifier contributes to the return numerator and capital denominator.
Ask for costs additional to the recorded prices, currency conversion, quantity changes and cash flows. Account equity, closed balance and a summed signal return are not interchangeable. Mark the latest source cutoff and the method used for still-open exposure. If overlapping observations are reported as independent successes, request an uncertainty method that reflects their dependence. No universal trade count or history length makes the evidence reliable automatically; completeness and a frozen convention must be inspected together.
A consumer-facing conclusion can say that the reference is reproducible, publication times are documented or execution evidence is incomplete. These are separate judgments, not one blanket verified badge. Link each conclusion to an inspectable artifact and its date. Specify what could change the assessment: a missing amendment, a different access tier or a corrected accounting convention may matter even if the headline percentage does not change.
The blank verification record is organised around event lineage and population reconciliation. Reproduce the reference and overshoot fixtures, then collect actual records before publishing a performance assessment. Do not infer a provider's private calibration or reconstruct missing trades from favourable charts. This guide supplies a review method rather than a completed paid-account test. An explicit request still awaiting evidence is a useful result for a buyer deciding how much confidence a marketing summary deserves.
| Object | Preserve | Do not infer |
|---|---|---|
| Reference | Earlier window, availability and warmup | A later chart reproduces the original signal |
| Event | Original message and amendment lineage | A timestamp proves every economic field |
| Population | Open, closed, cancelled and missing states | Selected recoveries are account returns |
Download blank review record (CSV)
Headers only. No trades, subscriber outcomes or operating receipts are supplied.
Documentation establishes the stated mechanism, not tested trading performance. Checked 10 October 2026; a later version can differ.
Define a prior-window reference, measure absolute recovery and avoid calling every correctly directed move a successful reversion.
Understand positive AR(1) decay, observation clocks, coefficient sensitivity and why half-life is not a trade deadline.
Design a bounded stationarity experiment without turning a unit-root test, fitted spread or chosen chart into proof of trading profit.
A record should retain events that cross without closeness, remain at equal distance, widen or fail to hit within the horizon. Their reference and operator must match successful rows. Missing source values require a separate unresolved branch. The workbook includes these authored examples to show why a hit-only table is incomplete. It is not a seller's actual ledger or an audit of its figures. Financial verification still requires authentic quantity, native exit, costs, weights and independent scope, with the performance cutoff distinct from a newly revised article date.