Proof samples and selected-winner bias
Keep sample selection, whole-period coverage and successful check counts distinct before making a complete-history claim.
Evidence methods, not a provider certification. Authored examples and user-entered check states are not independently verified trading results.
A convincing example is not a population
A provider can show one technically strong proof without supplying all original calls for the period. That example may be useful, but its scope must remain one record. A complete-history claim needs the original-ID population and check map. The reviewer should identify how examples were selected, whether selection occurred before outcomes and which records were inaccessible.
The practice packet lets a reader compare the full eight-ID roster with a smaller declared-pass subset. Publishing only those three labelled rows would conceal pending, failure, unperformed and missing states. The three rows do not represent an observed provider sample. They are an authored illustration of why proof success count, sample size and original population are separate denominators.
Predeclare the selection rule
A sample can be selected for a stated purpose, such as testing byte-format reproduction across versions. That is different from estimating full-period proof coverage. Keep the selection rule, source roster, sampling time and limitations. Convenient examples from the latest website may omit abandoned channels, earlier format versions or unsuccessful check states. A clean sample cannot answer questions its inclusion policy excluded.
Do not invent a statistical confidence interval for a non-random convenience sample. This publication has no genuine provider proof panel and no measured population success rate. If a real review uses sampling, its methods and evidence need to justify the resulting inference. Where that basis is absent, report the inspected sample directly instead of upgrading it into a verified full-history statement.
Coverage should retain unsuccessful and unperformed states
Pending receipts, malformed inputs, missing files and unperformed checks have different meanings. A failure count cannot be inferred by subtracting displayed winners from all originals. Conversely, unperformed records cannot be counted as successful merely because the proof protocol is capable of verifying them. Preserve each state and the exact requested artifact.
In the packet, three declared_pass labels coexist with one pending, one declared_fail, one unperformed and two missing. These are invented labels, not actual success or failure transcripts. The ID checker retains them separately and verifies only their schema consistency and identity membership. It cannot convert them into independently established events. A public table should keep this status language visible in downloads and machine discovery too.
A proof sample is not a strategy-validation sample
Software correctness checks and genuine financial observations answer different questions. Reproducing a digest for an authored file shows a hashing operation; it does not validate a trading entry, test a model out of sample or demonstrate subscriber profitability. Even genuine proof coverage would need separate quantities, fills, costs and account accounting to evaluate performance.
A proof attached to a winning display should therefore be reviewed for its committed scope and chronology, not assumed to certify the displayed profit. Omitted fields and research-reference prices remain separate. A selected set of target touches cannot establish quantity-weighted net outcomes. The absence of a complete ledger should remain an evidence request rather than be replaced by a favourable simulated account path.
Publish a bounded sample conclusion
A careful summary identifies the selection policy, inspected original IDs, file identities, check states and limits on extrapolation. Link unresolved records rather than hiding them behind an overall star rating. If a later review expands coverage, preserve the earlier sample and new cutoff. The changed denominator is part of the history, not an embarrassing detail to overwrite.
The sample exercise can be exported alongside the full packet to demonstrate the lost information. It remains fictional. Actual review conclusions require real retained inputs and performed checks. The key discipline is maintaining a distinction between an inspectable example, a reconciled declared population and a genuinely completed whole-period verification. None alone supplies an economic result or validates a provider recommendation.
Mechanism and source references
Mechanism references are not provider certifications. The new editorial examples are authored exercises; no independent financial review was performed.
Inspect the working packet
Eight originals, four discrepancy variants and six chronology cases