Current Understanding
Snapshot, not history. The changelog shows how explanations got corrected over time — this is just the current best answer to each real question, so nobody has to replay the correction chain to know what's true right now. Reference material other pages link into by row, not a finding on its own.
| # | Question | Current best answer | Confidence |
|---|---|---|---|
| 1 | Does the full 15-estimator pipeline run at all? | Yes | High |
| 2 | Are all 6 regimes reachable? | Yes, verified on held-out seeds | High |
| 3 | Does the Mind Pool produce genuine diversity? | Yes, real non-zero disagreement confirmed | High |
| 4 | Is tilt a spectrum or binary? | Spectrum, real propagation confirmed pre- and post-consolidation | High |
| 5 | Do Kalman and baseline filters actually differ? | Yes, confirmed under correct sequential testing (Fix 8) | High |
| 6 | Does FractalBrain have a real effect? | Yes, context-dependent, direction varies, small magnitude | Low-Moderate |
| 7 | Does HierarchicalConsolidation matter? | Large observed effect (0.55 vs 0.79 confidence) | Low-Moderate, one data regime, n=30 |
| 8 | Are the 6 previously-duplicate estimators now genuinely diverse? | Numerically yes (15 to 1 near-duplicate pairs) | High for diversity, Low for prediction improvement |
| 9 | Did decorrelating those estimators improve real predictions? | No evidence it did; weak evidence it didn't (r=0.09 old vs r=-0.05 new, n=30, not significant) | Low |
| 10 | Does the linear+four-channel alternating architecture work mechanically? | Yes, full pipeline runs, postmortem trace works | High |
| 11 | Does alternating representations outperform a single-paradigm pipeline? | Not yet tested (Pipeline A/B/C comparison proposed, not run) | Untested |
What this table is not
Not a claim that everything below High is wrong — most of it is a real, honest, currently-open question, not a failure. The point is separating "we know this" from "we don't know this yet" without reading the whole changelog to find out which is which.
Real next priorities, in order
- Run the Pipeline A/B/C comparison (row 11) against real outcomes — the biggest genuinely untested claim on this list.
- For any estimator kept specifically because it was decorrelated (row 8, row 9), test it against real outcomes before trusting it, not just against the other estimators.
- HierarchicalConsolidation's large effect (row 7) needs replication on a second real dataset before it graduates past Low-Moderate.