Summary
Within the divergence-state cohort (firm-years where A > 0 and G < 0 — the "healthy-looking earnings, weak balance-sheet ground" pattern), the divergence / masking geometry posts AUROC 0.705 for predicting future structural deterioration — beating structural distress (0.666) and joint-collapse severity (0.468 chance). This is first internal evidence for the divergence regime hypothesis. It is internal future-state, not external.
Anchor definitions (predeclared)
Divergence-state firm-year (at FY t)
A firm-year qualifies as divergence-state if, at year t, both:
- A_score(t) > 0 (above-median operating appearance), AND
- G_score(t) < 0 (below-median structural ground).
Future-deterioration positive (at FY t)
Positive if, at FY t+1 or t+2, EITHER:
- G drops by ≥ 1 z-score-unit from its FY-t value, OR
- The firm enters anchored pre-petition state.
Negative if forward observation exists (the firm has at least one row at t+1 or t+2) and neither condition holds.
Cohort sizes
- Phase-18 panel rows (post-gate)
- 36,041
- FY 2019–2023 rows (observation window)
- 13,582
- Divergence-state rows
- 2,626
- With ≥ 1 forward observation
- 2,573
- Future-deterioration positives
- 280 (10.9%)
Within the divergence-state cohort
| Geometry | n_rows | n_pos | Base rate | AUROC | AP | Lift @ 10% |
|---|---|---|---|---|---|---|
| Structural distress | 2,573 | 280 | 0.109 | 0.666 | 0.183 | 1.86× |
| Joint-collapse severity | 2,573 | 280 | 0.109 | 0.468 (chance) | 0.126 | 1.79× |
| Divergence / masking | 2,573 | 280 | 0.109 | 0.705 | 0.186 | 1.82× |
Comparison to all-comers cohort
| Geometry | Divergence-state AUROC | All-comers AUROC |
|---|---|---|
| Structural distress | 0.666 | 0.465 (mean reversion) |
| Joint-collapse severity | 0.468 (chance) | 0.552 |
| Divergence / masking | 0.705 | 0.342 (mean reversion) |
Headline reads
- The divergence / masking geometry leads on the divergence-state cohort by 4 pp over structural distress.
- The 36-pp swing for the divergence geometry between cohorts (0.342 broad → 0.705 in divergence-state) is too large to attribute to noise.
- This is the first piece of evidence that the canonical R geometry behaves as the Phase-13 doctrine predicted — useful for the divergence regime, not for broad Chapter 11 / Item 1.03 distress screening.
- Joint-collapse severity is at chance on the divergence-state cohort, consistent with its role as a near-petition severity geometry rather than a transition predictor.
Caveats
- Internal anchor. The future-deterioration label is constructed from the panel itself, not from an external bankruptcy or restatement event source.
- Predeclared but limited. The divergence-state thresholds (A > 0, G < 0) and the deterioration threshold (G drop ≥ 1 z) are predeclared, but their sensitivity has not been swept in this study.
- External validation pending. A future-phase study using restatements, going-concern warnings, or material-weakness disclosures would convert this into external validation.
Wording boundary
- ✓ "First internal evidence supporting canonical R for the divergence regime."
- ✓ "Internal future-state anchor; external validation pending."
- ✗ "Canonical R is validated."
- ✗ "Divergence regime is established."
- ✗ "Canonical R predicts bankruptcy."
CORPORATE_PHASE19_DIVERGENCE_ANCHOR_STUDY.md
and processed/phase19_divergence_anchor_results.csv
in the project repository.
Back to Validation · Evidence · Documents index.