Summary
Phase 14 originally built the Item 1.03 anchor table
through end-of-2025 (latest file_date
2025-12-15). Phase 19 ran a bounded EDGAR full-text
search refresh from 2026-01-01 → 2026-04-26 to look
for new high-confidence Item 1.03 anchors and check
whether the Phase-18 finding survives an out-of-time
directional check.
Counts
| Stage | n |
|---|---|
| EFTS hits in refresh window | 30 |
| Rejected for items field not containing "1.03" | 24 |
| Accepted high-confidence Item-1.03 anchors | 6 rows / 6 distinct CIKs |
| Of those, already in Phase-18 panel | 5 |
| New CIK (not in Phase-18 panel) | 1 |
Out-of-time descriptive check
For the 5 in-panel CIKs, we located their pre-petition firm-year rows in the Phase-18 scored panel and recorded where each row falls in the test-fold score distribution for the candidate geometries. With N = 22 pre-petition rows split across three lag classes (pre_1 / pre_2 / pre_3_plus), no statistical CI is meaningful; this is a descriptive directional check.
| Geometry | Lag class | n | Mean percentile | Top-decile capture |
|---|---|---|---|---|
| Structural distress | pre_1 | 3 | 61.2 | 0/3 |
| Structural distress | pre_2 | 4 | 56.3 | 0/4 |
| Structural distress | pre_3+ | 15 | 62.3 | 1/15 |
| Joint-collapse severity | pre_1 | 3 | 62.6 | 0/3 |
| Joint-collapse severity | pre_2 | 4 | 73.7 | 0/4 |
| Joint-collapse severity | pre_3+ | 15 | 56.4 | 1/15 |
| Divergence / masking | pre_1 | 3 | 50.7 | 0/3 |
| Divergence / masking | pre_2 | 4 | 34.5 | 0/4 |
| Divergence / masking | pre_3+ | 15 | 54.6 | 1/15 |
Honest reads
- Mean percentiles for structural distress (56–62) are above the 50th-percentile chance line — directionally consistent with Phase 18.
- 0 / 7 in the top decile at lag pre_1 / pre_2 is consistent with the Phase-18 anchored top-decile lift of 1.81× (top 10% catches ~ 18% of anchored positives).
- Joint-collapse severity at pre_2 mean percentile 73.7 (N = 4) is suggestive but tiny.
- Divergence / masking percentiles straddle chance — consistent with its role as a divergence diagnostic, not a Chapter 11 / Item 1.03 distress screen.
What this is: a directional out-of-time check. What this is not: a statistical out-of-time validation. With N = 22 across three lag classes, no AUROC / CI is meaningful. The Phase-18 finding is not contradicted by the refresh; it is also not strongly confirmed.
CORPORATE_PHASE19_ANCHOR_REFRESH_NOTE.md and
processed/phase19_anchor_refresh_validation.csv
in the project repository.
Back to Validation · Evidence · Documents index.