Final Freeze Note — Corporate A/G Research Extension

Summary

The Corporate A/G Research Extension is a research-grade application of A/G-style structural decomposition to public SEC company filings. It studies structural distress, joint-collapse severity, and divergence / masking regimes among eligible non-financial operating companies. After a five-cohort validation chain (Phases 15b → 16 → 17 → 18) and a five-track Phase-19 hardening round, the corporate branch is locked as a research artifact.

What it is

A reproducible firm-year panel built from public SEC EDGAR data only.
Per-firm-year scores for three risk-up geometries computed from a locked Phase-13 spec.
An evidence chain from raw data through validation metrics, with documented attenuation as the cohort moves from enriched to realistic.
An honest record of where the signal survives and where it is matched or beaten by simple primitives.

What it is not

Not a bankruptcy prediction product.
Not a credit rating.
Not investment advice.
Not a production model, regulatory-grade screen, or automated decision system.
Not a replacement for credit analysis.
Not a production risk model or automated decision system.

Main result

Phase-18 full-population pooled metrics on the FY 2019 – FY 2025 test fold (19,410 rows; 231 strict positives; base rate 1.19%):

Structural distress AUROC 0.711 strict / 0.683 anchored — the strongest A/G composite screen.
Joint-collapse severity AUROC 0.677 / 0.642 — near-petition severity geometry.
Divergence / masking AUROC 0.489 / 0.510 — chance for broad Chapter 11 / Item 1.03 distress screening; retained for the divergence regime.

Phase-19 baseline comparison: two single primitives — operating cash flow / assets and interest coverage — slightly outperform the composite on AUROC (0.736 and 0.726 vs 0.711). The composite’s value is robustness and regime separation, not maximum raw discrimination.

Geometry roles

Structural distress — primary broad screen.
Joint-collapse severity — near-petition severity / joint deterioration geometry.
Divergence / masking — diagnostic for the high-A, low-G regime; first internal evidence in Phase 19; not a Ch.11 screen.

Why corporate uses its own geometry

SEC / XBRL data is heterogeneous (tag variation, restatements, late filings, small-cap shells).
Chapter 11 / Item 1.03 distress is low base rate (~ 0.5 – 1.0% per year).
State-vector geometry keeps G and A separate so distinct regimes (joint collapse vs divergence / masking) can be diagnosed rather than averaged out.
Anchor coverage is imperfect (Item 1.03 8-Ks only).
Sector heterogeneity is large; pooled metrics hide cell-to-cell variation.

Allowed claims

"Research-grade preliminary full-population screening evidence."
"Eligible SEC non-financial operating-company universe under project filters."
"Ranked monitoring / triage primitive."
"A/G-style structural decomposition adapted to SEC public-company filings."
"Research-grade extension subject to SEC/XBRL constraints."
"Not a bankruptcy prediction product."
"Canonical R retained for divergence / masking; first internal evidence; external validation pending."

Forbidden claims

"Predicts bankruptcy."
"Validated commercial bankruptcy predictor."
"Production model" / "regulatory-grade."
"Replaces credit analysis."
"Investment advice" / "credit rating."
"Top bankruptcy risks" / "highest-risk companies" / "likely to fail."
"The A/G composite improves on simple financial primitives" (it does not, on this data).

Final classification

Corporate branch status: LOCKED / research-grade SEC public-company extension / not production / worth preserving.

Source basis: this is a public summary of CORPORATE_FINAL_FREEZE_NOTE.md in the project repository. The full text records allowed / forbidden wording in detail and cites the underlying phase notes.

Back to Evidence · Documents index.