Case · Checks, scores and validators · Internal record

A convincing validator was rejected because it never observed the geometry

A D495 Unreal workflow whose validator claimed source-to-instantiated-geometry correspondence, and Instrument Qualifier v0, a prototype for tracing whether evidence can reach its claim.

The system said: Transform inversion agreed with the source; validation passed.
The evidence showed: Both values came from the source row. Three changes to instantiated geometry were invisible to the validator and visible to an independent readback witness.

§ 1

The claim tested

The validator establishes that the geometry instantiated in Unreal corresponds to the source coordinates it was given.

§ 2

What Omega did

Traced the provenance of both sides of the reported transform comparison instead of accepting the comparison's mathematical form at face value.
Found that the placed transform and its inverse were reconstructed from the same CSV-derived source row, while the runtime relocation path also iterated source rows without geometry readback.
Built a 4,225-sample regression fixture with a clean control and three post-ingestion corruptions that changed only the instantiated geometry: scale x1.1, a 400-vertex zeroed patch and a +5 metre shift.
Compared the original D495-style calculation with an independent witness that read the changed coordinates, then exercised the dependency rule on seven authored instrument graphs.

§ 3

What was found

The source-derived check reported zero error and produced byte-identical output in the clean run and all three corrupted runs.
The independent readback witness reported maximum errors of 4.5 metres, 44 metres and 5 metres for the three corruptions, and zero for the clean control.
The original validator could therefore verify a source-side transform identity, but it could not decide whether instantiated geometry corresponded to that source.
The seven-case qualifier prototype correctly rejected authored circular and disconnected evidence paths, but those cases were hand-authored; recall on real validators is unknown.

§ 4

Verdict

CONTRADICTED: The D495 V0 validator establishes source-to-instantiated-geometry correspondence. Its evidence path never reads instantiated geometry, and three geometry-only corruptions leave its result unchanged.
INCONCLUSIVE: The original instantiated Unreal geometry corresponded to the source. The available validator cannot decide this claim, and no repaired Unreal readback run was performed.
INCONCLUSIVE: Instrument Qualifier v0 generally identifies invalid validators. Seven authored cases test the rule; they do not establish recall on real systems.

§ 5

What this does not establish

One internal workflow and one locally reconstructed dependency pattern. No repaired validator was built in Unreal, no claim is made that the original scene was wrong, and the local fixture is not an independent Unreal reproduction. The qualifier's graphs were hand-authored and trusted; extraction from arbitrary code was not demonstrated.

§ 6

The records

Every field of each finding this case rests on, as it appears on the ledger.

OBSERVED unreal-validator-never-read-instantiated-geometryobserve · observation boundary A convincing-looking Unreal validator was rejected because it reconstructed transforms from the source CSV instead of observing instantiated geometry. Three planted geometry corruptions left its result byte-identical.

Claim: The D495 V0 validation path could not establish correspondence between source values and instantiated Unreal geometry: its transform check calculated both the placed value and the inverse from the same CSV-derived row, and its runtime relocation path continued to iterate source rows without geometry readback. In a 4,225-sample regression fixture, three post-ingestion geometry corruptions changed the independent witness by up to 4.5 metres, 44 metres and 5 metres while the D495-style check continued to report zero error and produced byte-identical output.

Status: OBSERVED Witnessed in one instance. No claim about how often.

Question: Passed is not correct

Subject: An internal D495 Unreal validation workflow and a local regression fixture; the qualifier, planted corruptions and independent readback witness are Omega's. built for the study

Frame: One 4,225-sample fixture reconstructed from the D495 dependency pattern; one clean control and three post-ingestion corruptions: scale x1.1, a 400-vertex zeroed patch and a +5 metre shift; seven authored instrument-dependency cases in the qualifier prototype.

Method: Trace each claimed comparison to the source of both operands, then mutate the claimed subject after ingestion while keeping the source unchanged. Compare the original validator output with a witness that reads the instantiated geometry rather than reconstructing it.

Oracle: Direct readback of the mutated instantiated coordinates in the regression fixture, compared with the frozen source coordinates. The original validator's own transform and inverse are not used as the oracle.

Negative control: Present Three planted geometry corruptions had to be detected by the readback witness and missed by the source-derived check; the clean control had to remain at zero. The witness reported maximum errors of 4.5 metres, 44 metres and 5 metres, while the invalid check remained at zero and byte-identical.

Denominator: One clean control and three planted corruptions over one 4,225-sample fixture; seven authored dependency-graph cases in Instrument Qualifier v0.

Preregistration: None. This was an instrument audit and local regression investigation, not a preregistered experiment.

Limitation: This does not show that the original Unreal scene was geometrically wrong, and no repaired Unreal validator was built or run against the scene. The regression fixture reproduces the data dependency locally; it is not an independent reproduction in Unreal. Instrument Qualifier v0 used seven authored dependency graphs, so its recall on real validators is unknown and no general qualification capability is claimed.

Source: artifact not public

Reproduce: not reproducible from a public clone

Independent reproduction: None known.

This finding on its own page

All cases Bring us a problem