Aggregate only after feedback
Measure the record.
Measure the record.
Keep the claim small.
Preregister controlled comparisons, assign conditions without choosing them, and read effect estimates with their uncertainty. The analysis excludes sessions you marked compromised but never deletes them from the ledger.
0running
0completed
Evidence checkpoint / 40 valid trials
Still collecting a stable baseline
40 more valid trials needed before the 40-trial review checkpoint.
Chance expectation10 hits in 40
Reference threshold16 hits → p 0.026
One fewer hit15 hits → p 0.054
What this supports: No performance claim yet. Keep the protocol stable and retain every raw transcript, miss, and scorecard.
Longitudinal signal
25% lineCumulative blind-match rate
◎
Two valid reviewed trials are needed to draw a trajectory.
Interpretation boundary: this curve assumes independent 1-of-4 trials with equally plausible packets. Reused targets, memorable pool composition, subjective packet quality, optional stopping, and self-judging can violate that model.
Forecast calibration
Brier —Did certainty mean anything?
Judging confidence is locked after choosing a packet but before reveal. Lower Brier scores are better; use the buckets only after enough trials accumulate.
forecastnmeanobserved
25–49%0——
50–74%0——
75–100%0——
Mean session confidence —Mean correspondence —
Locked scorecard audit
0 scoredWhich descriptors separate the target?
The descriptor audit begins after the new locked scorecard is used; older sessions stay in the ledger but cannot be retro-scored.
Actual packet mean—
Strongest decoy mean—
Target advantage—
Selected / runner-up gap—
DescriptornActualBest decoyAdvantage
Geometry / shape
0
—
—
—
Natural vs artificial
0
—
—
—
Texture
0
—
—
—
Color / brightness
0
—
—
—
Temperature
0
—
—
—
Motion
0
—
—
—
Scale
0
—
—
—
Spatial arrangement
0
—
—
—
Distinctive objects
0
—
—
—
Overall gestalt
0
—
—
—
Preregistered comparisons
New experiment +
Experiment registry
◌
No experiments yet
First collect a stable baseline. Then compare one preparation variable with a frozen A/B plan.
Draft the first comparison ↗Reproducibility packet
Take the record with you.
Exports include closed sessions, immutable timestamps, condition snapshots, outcomes, confidence, and integrity flags. Unresolved assignments are deliberately omitted.