Research & evidence¶
Find evidence by research question · Search and filter the W3 evidence map · Full R76–R95 study chronology
Open science with its failure cases intact. This area explains what the development programme tested, what original archives and independent auditors reproduced, and which conclusions remain unsupported. It is separate from the stable 1.2.0 product.
Research ≠ a release qualification
1.2.0 is published, but these research claims remain unqualified. Simulation and source audits do not establish general nominal-90% simultaneous functional-band calibration or identify outcome-dependent gaze loss. Ten W1–W4 module-scoped workflows are stable engineering APIs; E1–E4 helpers remain experimental. Blocker #289 and #296 remain open.
- Functional inference under stress — W3
Leverage, cluster dependence, temporal and cross-coordinate covariance, heavy tails, contamination and observation support.
Read the W3 study map · Why calibration fails
- Missing gaze and audit return
R82–R95: MCAR/MAR/MNAR loss, selective secondary validation, known invitation, sharp bounds and externally assumed completion direction.
Identification explainer · Original/independent evidence
- Synthetic scientific visualizations
Inspect original figures with provenance and interpretation limits—not decorative performance claims.
Figure atlas · Research source figures
- Methods, numerical replication and provenance
Find simulation protocols, immutable source archives, independent replay and negative results.
New synthesis: which external information actually sharpens a missing-gaze identified set? compares R93, R94 and R95 on the same observed return law without treating assumptions as observed facts.
Interpretation rules¶
A successfully executed experiment proves that its algorithm ran under its specified data-generating law. An independently reproduced source archive provides reproducibility evidence; it does not convert a failed or underpowered inference procedure into a valid one. The qualification boundaries distinguish numerical reproducibility, statistical calibration, identifiability and software-release readiness.
Recent W3 results¶
| Study | What it identifies | What it does not identify |
|---|---|---|
| R93 | Sharp binary outcome sets for an observed return law | Unreturned outcomes from invitation alone |
| R94 | Conditional sharp sets under externally assumed completion floors | Whether those floors hold in field gaze |
| R95 | Conditional bounds if outcome-specific return direction is externally known | Direction of return from observed return data alone |