01 / TEST DIRECTION
Evidence selection
What is admitted, excluded or left unresolved?
Evaluation / Internal synthetic work
Know when not to answer.
A useful system should recognize when its evidence runs out. Internal synthetic tests explore evidence selection, contradictory explanations and the decision not to conclude. They are not independently verified product performance.
Evidence-governed design / In development
Measure the failure, not just the answer.
Evaluation asks whether evidence was admitted correctly, alternatives survived challenge, and unsupported consequences remained contained.
Internal synthetic tests can expose failures; they cannot certify phone protection or official benchmark standing. AUREX supports adversarial evaluation, while independent witnesses and inspectable artifacts remain separate requirements.
The aim is simple: know the source, test alternative explanations and keep permission separate from persuasion. Independent checks must support consequential claims. Approval must be current, limited in scope and withdrawable before an action—not after it.
EVIDENCE → QUALIFICATION → REALITY → AUTHORITY → EFFECT
- 01
Evidence
What was observed?
- 02
Qualification
Can we rely on it here?
- 03
Reality
What is supported?
- 04
Authority
What is permitted now?
- 05
Effect
What happened within bounds?
History is not permission.
A past observation stays in the record. A present action needs current, scoped authorization. Missing, stale or unverified evidence never counts as PASS. UNKNOWN ≠ SAFE.
A system must never be its own sole witness.
Independent support must be demonstrated, not assumed from agreement between models. Unresolved checks mean HOLD → human review. ORB MOTION AUTHORITY = NONE; colors, wheels and zoom provide presentation only.
DESIGN INTENT / DEVELOPMENTAL · This public explanation does not establish operational gates, independently verified product performance or autonomous authority.
Internal-prepublication · Evidence attachment required
A score needs a record behind it.
Prior internal holdout and hard-mock reports, including an HLE-shaped extreme mock exercise, lack attached protocols, item-level records and scoring artifacts here. Numerical results are withheld from this public presentation pending inspectable supporting evidence.
02 / TEST DIRECTION
Verification & abstention
When should the system decline to conclude?
03 / TEST DIRECTION
Contradiction handling
Do alternate accounts and failures stay visible?
04 / TEST DIRECTION
Authority containment
Does a proposal remain separate from permission and effect?
HLE-shaped. Not official HLE.
INTERNAL MOCK / SYNTHETIC HARNESS EVALUATION / EVIDENCE ATTACHMENT REQUIRED
Not official Humanity’s Last Exam. Not a live XAELI benchmark. Not independently validated. Not directly comparable to official leaderboard scores. Synthetic results can overstate real-world capability.
What a review would require
- 01Frozen dataset and provenance
- 02Harness, version and scoring protocol
- 03Item-level outputs, misses and exclusions
- 04Independent replication and scope limitations
HOLD · Independent qualification not established
AUREX / Experimental lab
Research without a borrowed endorsement.
AUREX explores reproducible questions, adversarial testing, model limits and abstention. Its research is not BCXMET phone security authority or an independently verified benchmark.
Help test the method behind the claim.
Independent researchers can help define a scoped, reproducible review. Missing or unresolved evidence never counts as a pass.