A common denominator inverts a published model ordering
The premise of different denominators failed: they were identical. On a 654-key shared grid, production and legacy had 0 holes, the other two 654 of 654; the published metric ranked legacy below typed-only (93.58% vs 94.86%).
Prediction failedHow we measureWorking note
- Designed for
- Built to test how the guard prototype's four rule models compare on one common denominator (anchor, real provider object, the object's own amount), and how much of it the published enumeration covers.
What was tested
A grid coordinate in the deterministic probe crossed anchor, real provider object and the object’s own amount across three synthetic domains, 654 keys, and scored four rule models of the guard’s prototype on that one denominator. The plan assumed the published per-model figures had different denominators.
What we observed
The premise was false: 545, 510, 517 and 168 were numerators over identical denominators. On the grid, the production and legacy models had 0 holes in 654, typed-only and foreign 654 of 654. On invoices the published metric ranked legacy (93.58%) below typed-only (94.86%), while the grid gave 0 of 536 against 536 of 536 holes. The published enumeration covered 73%, 76%, 0% and 4.10% of the grid; 97.6% of its keys did not discriminate models.
What this means for Heimdall
A metric without a common denominator can invert a ranking.
Limits
Synthetic deterministic grid; legacy ties the production model; a later census found 87.5% of grid cells unreachable by an agent; intervals on deterministic counts are not sampling uncertainty; the file is undated, so the date is the latest bound.
What remains open
Report the reachable share beside any grid count.