Failure replays ยท the record, institution by institution
One page per institution: how each lens read it before it failed, with leads in months, and where a lens saw nothing, a miss shown as a miss. Hazard panel behind the headline: 28 institutions, 206 panel rows, 10 failure events.
| Institution | Type | Failed | Action-zone lead (mo) | Funding lead (mo) | Fraud-masked |
|---|---|---|---|---|---|
| Abhyudaya Co-operative Bank | ucb | 2023-11-24 | 32 | miss | no |
| Altico | nbfc | 2019-09-12 | miss | 30 | no |
| CKP Co-operative Bank | ucb | 2020-04-30 | 1 | miss | no |
| First Leasing Company of India | nbfc | 2013-12-02 | miss | miss | yes |
| Global Trust Bank | bank | 2004-07-24 | 28 | miss | yes |
| IL&FS | nbfc | 2018-08-28 | miss | 41 | yes |
| Madhavpura Mercantile Co-operative Bank | ucb | 2012-06-04 | miss | miss | yes |
| Nedungadi Bank | bank | 2002-11-02 | 44 | miss | no |
| New India Co-operative Bank | ucb | 2025-02-13 | 47 | miss | yes |
| PMC Bank | ucb | 2019-09-23 | miss | miss | yes |
| Religare Finvest | nbfc | 2018-01-18 | miss | 22 | yes |
| Rupee Co-operative Bank | ucb | 2022-09-22 | miss | miss | no |
| Sambandh Finserve | mfi | 2020-10-13 | miss | 55 | yes |
| Sri Guru Raghavendra Sahakara Bank | ucb | 2020-01-10 | miss | miss | yes |
| United Western Bank | bank | 2006-09-02 | 66 | miss | no |
the four engines catch different failure physics on the same corpus: PCA/score tripwires catch ratio-visible deterioration (GTB 28 months early, Abhyudaya 32); the funding lens catches rollover/run failures the ratios miss (Altico 30 months early on CP reliance, IL&FS 41, Sambandh 55); the forensic screen owns fabricated reporting; the market layer reprices daily between filings for listed names
Hazard method, as served by the payload: Discrete-time hazard (pooled multi-period logit, Shumway 2001) on the vetted dossier panel; outcome = failure within 24 months (annual-filer-safe window; 12m headline via survival scaling); ridge fixed a priori; every reported number is leave-one-institution-out or past/future split - never in-sample.