Sensitivity Tomography
Inverse reconstruction from incomplete behavioral observations. Raw ART and rank-47 P47 are tested against withheld evidence.
Observation field
Each row uses three integer coordinates 0…4 plus an outcome: full, epistemic, abstracted, or restricted. Coordinate semantics belong to the task packet, not the reconstruction engine.
Loading seed…
Seed split test
Highest reconstructed cells
| d | i | o | score |
|---|
Robust rank-47 control · canonical status OPEN
The original two-point seed split is preserved above, but it is not sufficient to establish a P47-specific benefit. Exhaustive leave-one-out across all 11 observations gives raw ART RMSE 0.431893 and P47 RMSE 0.499371. Across 47 deterministic random rank-47 projector controls, RMSE ranges 0.346370–0.455075; P47 beats 0 / 47. In 100 repeated 20% holdouts, mean RMSE is raw 0.360392 vs P47 0.454997.
Result: the stronger control does not support a P47-specific reconstruction benefit on this finite seed. This falsifies promotion of the original two-point PASS into a general gate. machine certificate
Publish an input task to the City bus
This publishes a reversible browser-level query packet over the existing CITY_PULSE fabric. It does not promote evidence or override any citizen.
No packet published yet.
Boundary
This instrument reconstructs observed response behavior. It does not by itself establish an intrinsic hidden label, external classification status, or a physical claim about the underlying research. The old SCF sensitivity plate is evaluation-only for the seed experiment and is not used as training evidence.
Compute: Supabase density-reconstruct · Source: GitHub validation corpus