58 finding rules, every one in the open.

RCPAQAP flags a result that is more than the APS from the expected result for review; |D| equal to the APS is within. The APS is evaluated at the target. RCPAQAP's own flag is authoritative; EQAWarden recalculates it to check concordance.

The warning zone 2/3 < |A| <= 1 is ISO 13528's 'questionable' band 2 < |z| < 3 with σpt = APS/3. It is not an RCPAQAP flag.

Censored results are included in RCPAQAP's statistics with the sign stripped but are not assessed against the APS.

RCPAQAP's assessment is authoritative. EQAWarden reproduces it to check concordance; a difference beyond rounding usually means a different APS table version, target or unit.

Red: outside the APS of the all-result target (and the peer group). Amber: outside one and within the other (introduced about October 2024).

RCPAQAP calculates group statistics only on groups of 6 or more (3 values give a median only, 2 or fewer no statistics); a small group is sensitive to one mis-coded participant and to self-inclusion.

u(T) = 1.25 s/sqrt(p). ISO 13528 treats u(T) <= 0.3 σpt as negligible; above 0.7 (about the IUPAC upper bound) EQAWarden does not interpret z at all.

Without a reference system, or while routine methods are not yet aligned to one (status partial), methods legitimately differ; an all-method median then mixes method groups. While it holds, the calibration weight of findings that restate a persistent offset is counted for material.commutability (RULES 9).

Tested only when the targets are far enough apart to tell (|T_B - T_A| > APS(T_A) + APS(T_B)), at least one result is outside the APS and both would be within it once exchanged. log10 LR = 9(Q - Q')/(2 ln 10) under a normal model with σ = APS/3. Never alters a result.

Two or more swap candidates in one survey, with every other discriminable analyte fitting better exchanged, indicate a specimen swap; a single analyte indicates a result-entry swap.

Each factor of the analyte (units.factors; divalent ions also mEq/L; HbA1c by the IFCC-NGSP master equation) is tried in both directions; factors within 10% of 1 are skipped. Candidates are ranked by |A|.

Powers of ten from -3 to +3 are tried; a candidate is kept only if it lands within the APS while the reported result does not.

Adjacent digits are transposed (decimal point fixed) or a digit is dropped or duplicated. At low concentrations such changes can land within the APS by chance, so 'likely' wording needs |A_candidate| <= 0.5.

Only tested when the two targets are more than one APS apart and share the unit.

Youden rotation of the two samples' z_APS: S = (z_A + z_B)/sqrt(2) is the common component; |S| >= 3 is unusual (p about 0.003) under independent normal errors.

R = (z_B - z_A)/sqrt(2) is the differential component; a swap produces a large |R| with the two points mirrored about the target, so the swap test is quoted.

Inverse-variance pooled log-ratio across the vial's dilution-linear analytes with Cochran's Q for homogeneity; fires when |Z| >= 3, p(Q) > 0.05, at least 5 analytes, |Z| >= 3 still holds with any one analyte left out (the shift does not rest on one analyte), and the other sample does not show it.

Reconstitution affects one vial; a shared analytical cause usually shifts both samples and IQC.

Four or more analytes beyond 2/3 of the APS in one direction on one instrument, with a median |A| <= 2/3 for the same analytes elsewhere.

A fault in a shared part of the analyser or the material moves a known set of analytes together in a known direction (for example every enzyme activity with reaction temperature, or sodium, potassium and chloride with the ISE module). The signature library says which analytes, which direction and what the finding supports.

At least 2 of the last 6 scored results outside the APS, including the current one.

Mean of the survey summaries zbar_s with a t interval (df m - 1), which does not assume the laboratory's SD equals σpt and absorbs the within-survey correlation. Fires when p < 0.05 and the mean APS score magnitude is at least 0.25.

A current run of at least 6 same-sign survey summaries (p = 0.5^(k-1)); also the Westgard-style 10x rule.

Rules on z_APS: 2_2z two consecutive beyond 2 on the same side; 4_1z four consecutive beyond 1 on the same side; R_4z within a survey one above +2 and the other below -2. 1_3z is RES-OUT.

Mann-Kendall on zbar_s against time (exact for m <= 10) with a Theil-Sen slope. Step and drift are reported side by side and not chosen automatically when m < 10.

Two-sided tabular CUSUM on per-sample z_APS in time order. With h = 5 a false signal within 24 results occurs about 4% of the time; a 1 σpt shift is detected within 12 results about 73% of the time.

RCPAQAP's method: least squares of results on targets over at least 6 samples with one target source and method category, one pass of atypical exclusion. Deming is primary when target uncertainty attenuates the slope by more than 1%; Passing-Bablok is a robustness check only.

MPS combines imprecision and bias over the cycle against the APS at the mid-point of the targets; RCPAQAP's summary marks values above 1.0.

Slope CI excluding 1 and an effect at T_mid of at least a quarter of the APS there.

Intercept CI excluding 0 and |intercept| at least a quarter of the APS at the lowest target.

Atypical: more than 3 S(y.x) and more than the APS from the laboratory's regression line.

Quadratic term t-test (n - 3 df) and a runs test on residual signs, each at alpha/2 (Bonferroni) so the rule's false-alarm rate stays at alpha; needs at least 8 targets.

Three-state comparison using the one-sided 95% limits of the CV against APS%/2 (RCPAQAP 'APS = 2 x CVa') and, when configured, the biological-variation desirable CV. Fires on 'does not meet'; sigma is never a flag.

Over at least 6 surveys: systematic share, a t test of mean R (constant or non-linear error shows as consistently signed R) and an F test of the run effect.

Absence of evidence is not evidence of absence, so the bias range the interval still allows is stated.

Survey summaries before and after the change (up to 6 each side, excluding the triggering result when testing persistence); exact permutation p. Fires when p < 0.05 and the step is at least half the APS. With 3 or fewer surveys on a side, a non-significant result is never read as 'no effect'.

One calibrator or reagent group serving several analytes, with same-direction LOT-STEP (or IQC steps) in at least 2 of them after its change.

A reagent or calibrator lot change within 14 days before the analysis date.

A primary lot on the analysis date was past its expiry date as LotWarden recorded it.

Context for reading IQC; never evidence for or against an analytical cause by itself.

A stated limit: LotWarden data or a confirmed analyte map is missing.

At least one level's change has a CI excluding 0 with the sign of D, and |Zc| < 2 at every level with data. Zc compares the IQC change with the EQA deviation in %.

Every level's change CI includes 0 and |Zc| >= 2, on a warning or outside result. A reset IQC target can hide a shift, so the caveat is always shown.

QC-material-specific (non-commutable) response to a reagent lot is common.

RCPAQAP's flowchart: check IQC on the EQA run; a run out of control needs action and documentation.

Calibration, maintenance and reagent events shortly before the run are context for the result.

SDI = (laboratory mean - peer mean) / peer SD. Bio-Rad bands: <= 1.25 acceptable, 1.25 to 1.49 acceptable to marginal, 1.5 to 1.99 marginal, >= 2.0 unacceptable. Concordant when the sign matches the series bias and |SDI| >= 1.25.

Comparing across QC lots confounds the lot's own target with any analytical change.

A stated limit.

An analyte pitfall from the knowledge base whose direction and signature match this result or series.

Method-dependent analytes are fairly compared only within the method group; RCPAQAP's amber flag is the scheme's own signal.

Mean A against the category median includes 0 while mean A against the all-method median does not, with a consistent method-group offset. RCPAQAP's flowchart: if the method bias is due to a matrix effect, no further action is required.

A correlation (post-analytical correction) factor set in the middleware, LIS or analyser multiplies or offsets every result it reaches, including an EQA result that is not excluded. RCPAQAP states that such factors must not be applied to EQA or QC results. IQC reviewed on the analyser before the factor is applied will not show it.

The method group median is compared with the assessment target on the same sample. When the laboratory is beyond the warning zone but its method group sits within a third of the APS of the target, the method and the material are unlikely to explain the deviation.

When at least half of the laboratory's deviation, in the same direction, is also the method group's deviation from the target, a method or material effect is more likely than a fault in the laboratory. RCPAQAP's flowchart: a method bias due to a matrix effect needs no further action; a method bias that also affects patients is a matter for the manufacturer.

z = (result - group median) / group SD, from RCPAQAP's Method Comparison table for the laboratory's own group. |z| 2 to 3 questionable, 3 or more unsatisfactory (ISO 13528). The group SD is the state of the art for the method, not a performance specification; the APS remains the assessment.

With a specific (reference) target, the category median is compared with the target on every sample of the series. The rule fires when the group's mean offset is at least a third of the APS with a 95% CI that excludes 0, and the laboratory's mean APS score against its own group's median includes 0.

The ranking is RCPAQAP's, printed in the survey report's Precision and Accuracy block for the cycle's regression. The bands are from RCPAQAP's legacy end-of-cycle flowchart and are EQAWarden defaults.