R12 EFC Quotient-Fisher Fresh-Oracle Result
Decision
qfcr_mechanics_no_go
The fresh externally seeded audit does not authorize quotient-Fisher causal retraction as a neural architecture or optimizer mechanism. The treatment failed its preregistered every-world mechanics requirement and did not establish the required advantage over an equal-step Euclidean control.
This result does not weaken the source/process custody mechanics. Those gates passed. It closes the proposed geometry.
Frozen Provenance
- source commit:
1e60a19c942ce0bd02156832f0324db68ae394d3 - authorization commit:
11e1dc9e45c871083e62f85669b4101459a3518f - NIST Beacon chain:
2 - exact pulse:
1,874,057 - pulse timestamp:
2026-07-24T21:37:00.000Z - minimum source-to-pulse delay:
21,600seconds - official report:
artifacts/r12/qfcr_fresh_oracle_20260724.json - report payload SHA-256:
71ae528e431e7aff4418446420a9c3cdbb1a197e72a3a7f272965e24ac6d30e6 - full report SHA-256:
7fc9a73717788bc423044d67b33193b5634ca298d379adbe89d2c586cc09d97b
Both consecutive NIST signatures, pulse output hashes, the previous-output
link, and the previous precommitment reveal verified. The source tree remained
clean and identical to public origin/main while the detached auditor ran.
Audit Coverage
- 200 fresh worlds
- 83 eligible worlds
- 666 eligible faults
- 133,200 expected evidence identities
- 133,200 observed evidence identities
- source closure: pass
- pre/post environment identity: pass
- board counts: pass
- recoding decisions: pass
- output reservation and publication: pass
Arm Results
| Margin | QFCR exact | Euclidean exact | QFCR minimum world | QFCR median first exact | Euclidean median first exact |
|---|---|---|---|---|---|
| 0.05 | 100.0000% | 100.0000% | 100.0000% | 2 | 2 |
| 0.10 | 100.0000% | 100.0000% | 100.0000% | 2 | 2 |
| 0.20 | 99.6997% | 100.0000% | 66.6667% | 2 | 2 |
| 0.40 | 98.1982% | 97.2973% | 0.0000% | 2 | 3 |
| 0.80 | 96.6967% | 92.9429% | 0.0000% | 3 | 3 |
The one-step and small-step QFCR controls recover 0% at every margin. This confirms that the full update carries useful mechanics, but it does not rescue the claim: the treatment must recover every represented world at every margin before any stress advantage can be attributed to quotient-Fisher geometry.
Interpretation
At small margins, update scale is sufficient and QFCR is no earlier than Euclidean. At margin 0.20, QFCR loses exact recovery while Euclidean remains perfect. At higher stress, QFCR has a modest aggregate advantage, but both methods fail complete worlds. This is not the preregistered signature of a reusable causal geometry.
Do not:
- tune the frozen margins after seeing this result;
- rescore the consumed pulse;
- use aggregate stress improvement to override the every-world gate;
- authorize a QFCR neural fit; or
- report this as a reasoning gain.
Any future signed, second-order, quotient-algebra, or other replacement must be a separately named mechanism with a new board, controls, source freeze, and decision rule.