The rule this whole map lives by: a claim only gets on the map if it was earned — predicted in advance, scored against a rule written down before the data arrived, and reproduced exactly. Where we could not earn a claim, the map says what we measured and what is missing, with the same care.
The learned structure kept predicting correctly, most of its weight is confirmed, and a deliberately scrambled decoy signal earns nothing. Green cannot be argued onto the map; it can only be earned.
We have real measurements and often a trusted direction — but either confidence hasn’t been earned yet, or a tripwire fired and we won’t claim what we can’t defend.
No measurement exists, so a named public model’s estimate stands in — stamped with the model’s name, version and source, never rendered green, and always on record as a stake: when the measurement arrives, the model gets scored. Where several models fill the same cell, their disagreement is recorded too — model spread is how we price which measurements matter most.
A hole, stated as a hole — always with the specific measurement that would close it. Holes are product here, not embarrassment.
Every analysis behind this page follows the same discipline: hypotheses, thresholds and pass/fail rules are committed to a version-controlled ledger before the data is fetched; raw data is hash-pinned at download; learning runs are replayed independently and must match to four decimal places; decoy signals run alongside real ones as tripwires; and predictions that fail are published at the same volume as the ones that succeed. Several of the most useful findings on this map came from our own staked predictions failing.
The stamps flow both ways. Where measurement-earned structure and a named model’s output coexist, the register grades the model — agrees / corrects / flags / no-opinion-yet — with the confidence of the opinion stated. A certified zero is the sharpest audit there is: it bounds every model that predicts otherwise.
The register is the atlas’s other half: a machine-readable set of YAML files, one per gap — each carrying the place, the size, the list of explanations we tried and refused, and the named external measurement that would close it — plus the predictions we have staked in advance (including one scoreable when the ESCAPADE mission reaches Mars, staked ~2 years ahead). It ships as a citable repository; the DOI/link is stamped here on publication: [gap-register repository — DOI on publication].
Every claim on this page comes from a pre-registered analysis: the hypothesis and its pass/fail rule were committed to a ledger before the data was touched, results were reproduced bit-for-bit, decoy signals ran as tripwires, and refused hypotheses stay on record. The complete gap register — data sources, refused hypotheses, and open staked predictions — is published alongside this atlas.
Nervous Machine