Measurement scales.
One page per scale: what it measures first, then the ruler, the stated standard error per band, the tier and the honest-uncertainty display. Rows that fail the floor or lack evidence are shown and flagged, never removed.
The scales, by tier
Sorted by status: the reasoning ruler first, then the other scales in calibration, then Watch. No filtering: Watch scales are listed, not hidden. Every number on a scale page carries a tier word or a source; where no gated figure exists, the page says so.
In calibration
Canonical Reasoning Complexity (MHC Stage)
How much complexity a performer, human or AI, can actually sustain.
Precision: Set to an error bar of 0.20. Delivered at 0.21 once judge harshness is corrected at each level. Simulation: 15,000 decisions on judge data from our own runs, seed 20260908. With harshness left in, the same run came in at 1.16. Error bars in logits.
In calibration
Hallucination Scale
How often a model asserts unsupported content, and where it stops.
Precision: best band SE stated on page
1 no-evidence row, shown
In calibration
Persuasion Battery
How hard content pushes, and whether it pushes ethically.
Precision: best band SE stated on page
1 no-evidence row, shown
Watch
Agent-Generated Research Certification (AGRC)
Evidence, with its error bar, of what an agent got right in research outputs.
Precision: none certified
1 no-evidence row, shown
Watch
Rule Retention Coefficient (RRC)
How much of an agent's stated rules survive repeated context compaction.
Precision: none certified
1 no-evidence row, shown
Watch
Agent Eval / Agentic-Loop Reliability
Whether an agent finishes what it started, in order, within constraints.
Precision: none certified
1 no-evidence row, shown
Tier definitions
- Set to an error bar of 0.20. Delivered at 0.21 once judge harshness is corrected at each level. Simulation: 15,000 decisions on judge data from our own runs, seed 20260908. With harshness left in, the same run came in at 1.16. Error bars in logits. The 2026-08-22 gate is kept as a historical record on the Evidence page.
- In calibration: the numbers are real, and item locations are candidates until a calibration run passes its gate.
- Watch: the skill is defined and not yet measured. Not for high-stakes calls.