PRACTICE MATERIAL; ALL RECORDS INVENTED
Practice core outcome-review rules, version 19.4, 12 November 2026.

1. Match samples, summaries and outcomes by sample identifier exactly as written, in both directions. Each expected sample must have exactly one summary and at least one outcome. A missing summary or missing outcomes gives missing input and needs a look under this rule, never invented zero reads. Any identifier in summaries or outcomes but absent from the sample list needs a look under this rule. A repeated summary identifier or repeated full change for the same sample gives unusable input and needs a look under this rule, even when the repeated rows are identical. Keep every source row; never choose, merge or discard a repeated record.

2. All read counts are nonnegative whole numbers and aligned reads must be positive. Input reads equal aligned plus excluded reads. Where outcome rows exist, outcome reads sum exactly to aligned reads; unclassified reads equal the sum of outcomes classified unknown and are included in aligned reads. For expected samples, order, batch, reference, annotation and settings in every summary agree with the sample list. Check unexpected samples for every count problem too. Every field in the three data files is required; only the control field on control records may be blank. Allowed classifications are edited, unmodified and unknown; allowed frame annotations are frameshift, in_frame, unmodified, unknown, ambiguous and noncoding. A missing required field in a supplied row, a label outside those allowed values, contradictory summary values, metadata inconsistent with the sample list in those five fields, an invalid count, a zero aligned denominator or a count discrepancy gives unusable input under this rule. Contradictory values are a separate fact from the existence of repeated records. Check each supplied summary separately; if an outcome count is invalid, say the outcome total cannot be checked rather than treating that count as a valid number. Never replace a blank or invalid value with zero or an unknown label.

3. For usable input, an outcome's percent is 100 times its reads divided by its sample's aligned reads. List an outcome at 5 percent or more, including exactly 5 percent, comparing integer counts before rounding. Show the numerator and denominator, an under-floor subtotal, and every under-floor outcome separately for inspection. Show excluded counts outside aligned reads and unclassified counts within aligned reads. Never rescale the displayed outcomes to sum to 100 percent. Listing or retaining an outcome under this rule adds no needs-a-look mark.

4. Any supplied summary row recording a positive whole aligned-read count under 1,000 makes that sample need a look for depth, even when its input is unusable or its identifier is unexpected. Check every supplied summary row; show each triggering count and the 1,000-read limit. Exactly 1,000 does not trigger this rule. Zero is outside this depth rule. Any shares shown for a low-depth sample say depth concern beside them and are arithmetic on those reads only.

5. A clone recorded as having two expected copies needs a look when more than two outcomes are listed. The expected-copy entry is the order record's premise, not a conclusion from reads. A clone with exactly one listed outcome carries the note that this table cannot establish that every copy was represented. This single-outcome note applies to clone records only; controls and founders are outside it. The note is not itself a needs-a-look mark. For a founder, state that the clone rules do not apply and no inheritance conclusion is made.

6. For a clone with requested_review recorded as frame review, a listed outcome whose frame annotation is in_frame or unmodified needs a look. This is a sequence-record check, not evidence of loss of function. This rule does not apply to founders or controls.

7. Count every valid control read classified edited before using the floor; if any edited count is invalid, do not invent an edited total. A positive edited control count makes the control need a look under this rule. A control's review is unresolved when it has a finding under rule 1, 2, 4, 7 or 9; retain the finding's own rule number on the control, without adding a second citation for that fact. Every expected sample linked to a control with unresolved review is itself marked needs a look under rule 7, with that control's identifier and unresolved concern as the reason, even when no other rule marks the sample. The link is a separate fact from the finding on the control. Never accept a control, subtract its reads or decide release from this list.

8. Two or more clones with usable input in the same order, batch, reference, annotation and settings whose listed changes form the same set each need a look. Compare the full recorded change, including positions and inserted bases and any listed unmodified outcome, never size alone; ignore read shares when comparing the sets. List the partners and their changes. Matching sets do not establish common origin. Do not compare samples with missing or unusable input, controls or founders.

9. Any supplied outcome with unknown classification or an unknown, ambiguous or noncoding frame annotation makes its sample need a look, regardless of count or listing floor, including missing or unusable input, controls and unexpected identifiers. Preserve all supplied classifications, frame annotations and sizes; never infer a replacement or calculate a frame from size.

10. Keep input status and review marks visible together: missing input or unusable input does not replace needs a look under any applicable rule. Apply every rule that can be checked from supplied rows or links, even when another rule makes the input missing or unusable. If both input statuses apply, show missing input first and retain every unusable-input reason. Say no additional mark under these rules only for usable input with no needs-a-look reason. Give each distinct fact exactly one citation, to the rule that defines it, with relevant supplied counts; several different facts on one sample may have different rule numbers. This presentation rule adds no second citation or review mark to those facts. Show no calculated shares or outcome-set comparisons for missing or unusable input. A sample with no additional mark has no release decision. Never label a sample, clone or founder confirmed, correct, good, best, safe, passed or approved, and never rank candidates or call a knockout.

11. Head every list with its date, input file names and checksums, rows read from each file, and these rules with their version. Keep each sample's source-result pointer and the original row numbers beside its entries. A pointer names the source record; it does not mean the program has read that record. Any unreadable row is listed with its reason.

PRACTICE MATERIAL; ALL RECORDS INVENTED
