How PAT scores and ranks data

1.What a score is

A score is arithmetic on your answers. Every scored item is answered on one scale, 0 to 5, where 0 means not in place. A module score is the plain average of those answers placed on a 0 to 100 line. It is rounded once, to a whole number, at the moment it is displayed. Nothing is weighted by question count, response count or anything else, and no surface shows or implies decimals. These are informational readings for product and operating decisions. Where evidence is thin the surface says so, and where a peer benchmark cannot be published responsibly you see an insufficient peer data state instead of a number. Nothing here is professional advice, and no figure is a weighted, percentile-adjusted or significance-tested statistic beyond what this page states; there is no claimed statistical basis for any label.

2.The five modules and the alignment index

The firm assessment has five modules: Strategy, Operations, Automation, Integration, Governance. Each produces one 0 to 100 score from its own answers, so the five carry equal weight by construction. The alignment index is the equal-weight mean of the module scores you have completed, rounded once. It needs at least 3 final modules. With fewer, your surfaces read Partial and show the module scores on their own.

3.The product assessment and the firm review

A vendor assesses its own product on the scored questions for the utilities it declares. That reading is the vendor assertion. Firms that use the product review it on the same scored questions, utility by utility. That reading is firm-observed. The two are computed, stored and labelled separately and are never blended into one number, and a vendor's own assertion never enters the firm-observed average it is measured against. One firm, one vote: a firm's latest final review of a product is its vote. Earlier finals stay stored and readable as history and do not count. A review counts for 12 months; after that it stays readable on the product page and leaves every average.

4.The peer benchmark

Firms are compared inside a cohort: the same vertical, and the same size band once enough firms allow it. Demo and synthetic rows carry their own boundary and never enter a customer cohort. A cohort publishes only above the evidence floor (five contributing firms, no firm above 25 percent, structured consistent with the former 1996 DOJ/FTC safety zone, withdrawn in 2023). Below the floor you see an insufficient peer data state, never a number. Each calendar quarter has a cutoff, the last day of the quarter. Only final assessments completed on or before the cutoff enter that quarter's run. The run is computed from those rows and frozen; every surface reads the stored run, so a later submission lands in the next quarter and never moves a published number. Your percentile is the share of the cohort at or below your score, counted, never estimated. What you see depends on how many firms are in the cohort:

• Under 5 contributing firms: nothing is published.
• 5 to 9: the cohort median and your band.
• 10 to 29: the quartiles and your band. Your exact rank opens at 10.
• 30 and up: your percentile, with the quartiles and your rank.

5.Confidence labels

Every reading carries a confidence label that says how much evidence stands behind it. The label is about volume only, never quality, and it is a convention for keeping thin evidence visibly qualified. It never implies a p-value, a confidence interval or a significance test. One table applies everywhere; what counts as a data point is stated on each surface (firm reviews of a product, completed modules, reviewed products). Stale evidence lowers the label one step: a Grounded reading on stale evidence reads Early signal.

6.Freshness labels

Every score carries the age of the newest evidence behind it: the most recent module submission, product assessment or firm review that fed it. Age changes the label, never the number. The figure stays exactly as computed and the freshness label sits beside it. A missing evidence date reads as unknown, not as stale.

7.Divergence

When a vendor's assertion and the firm-observed reading differ, the gap is read item for item, and only once at least three distinct firms have reviewed the product. Below that floor the gap is shown as early signal and labelled too thin to read. Above the floor one pair of thresholds applies everywhere a gap is labelled: within 5 points the two readings are aligned; a gap of 10 points or more is flagged as a calibration gap; between the two it is watched, not asserted.

8.Knowledge banks

Each knowledge bank holds 90 items. A sitting serves 30 of them, drawn across the bank's difficulty tiers, with the anchor items the bank's charter names served in every sitting. The sitting score is the share answered correctly, shown as a whole-number percent with its band. Sittings are not benchmarked across firms.

9.Reproducibility

Every published number can be recomputed from the stored raw rows: the answers and their scale, the final flag and timestamp of each submission, and the cohort each firm belongs to. Nothing is published that cannot be rebuilt from those rows, and the run that produced a benchmark is stored with its cutoff so the same cut can be reproduced later.

10.The methodology version

This is methodology version 1.5. Material changes are announced in the changelog below before they take effect, and the version moves with them so a reading can always be traced to the rules it was computed under.