Screened eight entropy-family measures of how disordered recent prices are, to see which add information the shipped features do not. Four came through clean everywhere, and the closing matrix revealed that several of them are near-duplicates of each other.
| Quantity | Value | Meaning |
|---|---|---|
| Candidates screened | 8, each across 207 cells | the verdict is the worst cell in every case |
| Clean in every cell | 4 of 8 at max |rho| <= 0.85 | the grounded subset |
| Range across the family | 0.738 (bar_approximate_entropy) to 0.972 (bar_permutation_entropy) | spread of worst-cell overlap |
| Dropped before screening | 2 (multiscale entropy, multiscale permutation entropy) | N/tau collapse at a 200-bar window |
| Finish matrix | 8 candidates x 4 slices = 82 valid cells | internal-redundancy check |
| Leave-one-out R2 range | 0.318 (shannon) to 0.972 (sample); ApEn-SampEn median |rho| 0.965 | the most externally orthogonal are the most internally redundant |
| Firing | What it did |
|---|---|
| 01 | bar_shannon_entropy screened over 207 cells |
| 02 | bar_permutation_entropy screened โ highest overlap of the family |
| 03 | bar_weighted_permutation_entropy screened |
| 06 | bar_dispersion_entropy screened |
| 08 | bar_spectral_entropy screened โ work-list closed at 8 of 8 |
| Finish | Cross-candidate matrix over 8 candidates x 4 regime-diverse slices |
Stage-2 (project-wide constant ratification for the survivors, plus the SVD re-promote) is Terry's, outside the loop. SVD entropy is already 208/208 clean but deferred on the parameterless axis; it re-promotes once the shared family embedding becomes a project-wide constant.
'WORK-LIST COMPLETE (8/8). Cross-candidate matrix done. Cron 7f585874 deleted. LOOP DONE.' (LOOP_CONTRACT.md frontmatter: status DONE, iteration 8)
Lifecycle status is process state, not a judgement of the findings. Results are stated as numbers with their uncertainty. Generated between CLOSING-RECONCILIATION markers โ regenerate rather than edit.
Orthogonality loop ยท 8 candidates ยท rolling 200-bar window ยท updated 2026-06-03 04:06 UTC
| Iter | Candidate | Verdict | Cells | Max |ฯ| |
|---|---|---|---|---|
| 01 | bar_shannon_entropy | WATCH | 207 | 0.856 |
| 02 | bar_permutation_entropy | BAN | 207 | 0.972 |
| 03 | bar_weighted_permutation_entropy | PASS | 207 | 0.838 |
| 04 | bar_approximate_entropy | PASS | 207 | 0.738 |
| 05 | bar_sample_entropy | WATCH | 207 | 0.895 |
| 06 | bar_dispersion_entropy | PASS | 207 | 0.839 |
| 07 | bar_fisher_shannon | WATCH | 207 | 0.851 |
| 08 | bar_spectral_entropy | PASS | 207 | 0.771 |
| โ | bar_multiscale_entropy | DROP (W=200) | N/ฯ collapse on W=200: coarse series at ฯ=20 is 10 samples, SampEn undefined. | |
| โ | bar_multiscale_permutation_entropy | DROP (W=200) | Same scale-grid N/ฯ collapse on W=200. | |
Verdict = worst across all valid bar sets. PASS requires |ฯ|โค0.85 + Rยฒโค0.85 + h_normโฅ0.05 in every cell. Axis-2 (parameterless) is resolved in Stage 2 โ promotion of the shared embedding to a project-wide constant โ AFTER this loop completes.
Family rollup for the entropy orthogonality loop. SVD entropy (already 208/208 PASS, DEFERRED on Axis-2) re-promotes at Stage 2 once the family embedding becomes a project constant.