01 / TEST DIRECTION
Evidence selection
What is admitted, excluded or left unresolved?
Evaluation / Internal synthetic work
Measure limits. Disclose them.
Self-testing explores how a system selects evidence, handles contradictions and withholds unsupported conclusions. It is not independently verified product performance.
Internal-prepublication · Evidence attachment required
A test is a question, not an endorsement.
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
Evaluate the methodology, not the headline.
Independent researchers can help define a scoped, reproducible review. Missing or unresolved evidence never counts as a pass.