Completed-R input precision and export integrity
Keep numeric validation, sample status and original-risk source records attached to an exported series.
A decimal value has a declared representation
The retained completed-R analyser accepts one ordinary decimal value per nonblank line, at most six decimal places, absolute magnitude no more than one million R and no more than five thousand completed outcomes. Its character bound is also explicit. Scientific notation, comma-separated lists and arbitrary annotation are not accepted as a numeric series. That is a software contract, not a statement that a financial dataset has six-decimal economic accuracy. Rounded reporting values can differ from a transaction ledger. Preserve the original cash and risk records if a later audit needs more information than the entered decimal representation supplies.
Zero and missing are not interchangeable
A line containing zero is a completed flat result under the chosen input classification. Blank lines are ignored by the parser, so they cannot document open, unfilled, cancelled or missing positions. Those statuses need another inventory. An incomplete record must not be converted to zero just to pass a completed-series validator. Likewise, a negative value includes a loss rather than an import error, while a text token is invalid rather than automatically removed. The original analyser clears metrics and disables export on invalid input, so an earlier valid total cannot quietly survive as the apparent result of a changed series.
Export identifies what was entered
The JSON record distinguishes the built-in hypothetical example from user-entered unverified outcomes. It reports values, counts, the cumulative closed-R path and bounded diagnostics without claiming portfolio return. The publication does not upload the series or place it in a shared URL. A download remains the reader's local copy, not an independently certified statement. A user can reproduce the numeric summary while the full eligible population and original-risk denominators remain unverified. The ten-field evidence record is a separate note format; importing it is not a way to certify the numeric analyser's financial meaning.
Integrity and statistical evidence have different gates
Integer-scaled arithmetic protects sums and prior-peak declines within the parser's bounds. A canvas-pixel and rendering check establishes that the resulting visual is nonblank, not that its path is account equity. An actual empirical claim additionally needs source identities, closure rule, costs, original risk and complete exclusions. Values with different original cash risk can have cash consequences that the R total cannot show. Keep the actual event order rather than optimising its drawdown appearance. The new reference workbook supplies alternative invented positions, not an observed sequence to paste into the analyser and label as vendor performance.
Completed invented payoff case
The fictional input has seven completed net values: 2, negative 1, 0, 1.5, negative 1, negative 1 and 2. It sums to 2.5R, with three winners, three losers and one flat. A malformed token clears the result and cannot export the former total. The built-in example carries hypothetical classification. No original cash-risk or account-value ledger is supplied, so the 2.5R total is not promoted to a funded percentage return or audited strategy result.
| Object | Record | Bounded finding |
|---|---|---|
| Completed values | Seven valid decimal lines | 2.5R total |
| Classification | 3 positive / 3 negative / 1 zero | Position-level input |
| Malformed token | Not numeric | Clear; export disabled |
| Account return | No account ledger | Not calculated |

Primary references and scope
Expanded source review: 13 October 2026. Documentation supports the stated mechanism, not a tested account, hypothetical fill or provider outcome.