deal-v0.1 pre-checks (IDEA-014 M1)
Counts and minimum detectable AUC only. No score exists yet and no evidence-to-label relationship was computed. Produced by PYTHONPATH=src ~/.venvs/diaspore-sk122/bin/python -m diaspore.acquire.prechecks on 2026-11-03, fixed data end 2026-03-31.
Minimum detectable AUC: smallest true AUC separable from 0.5, alpha 0.05 two-sided, power 0.80, null Hanley-McNeil variance, one row per company.
(a) Comparable cells: priced deals dated on or before each as-of, stage read the day before each deal
266 live acquired_public rows; 179 have a low edge. All value edges have basis consideration_x_shares.
| as-of | no trials | ph0-1 | ph2 | ph3-4 | stage cells at n >= 10 |
|---|---|---|---|---|---|
| 2018-12-31 | 38 | 7 | 2 | 10 | 1 of 3 |
| 2020-12-31 | 50 | 8 | 4 | 18 | 1 of 3 |
| 2022-12-31 | 65 | 10 | 8 | 26 | 2 of 3 |
| 2024-12-31 | 84 | 15 | 15 | 38 | 3 of 3 |
These differ from the Obstacle map's cells because the map read stage on the deal date and did not apply the pre-declared merges. These counts govern (director, decisions.md "IDEA-014 D5 refinement").
(b) R2 hindcast label: later trial at a phase above the pre-deal maximum
Targets with pre-deal trials only. Deals whose horizon passes the data end are censored.
| horizon | uncensored | positives | censored | minimum detectable AUC |
|---|---|---|---|---|
| 36 months | 69 | 4 | 55 | 0.920 |
| 60 months | 48 | 2 | 76 | 1.000 |
Both exceed the registered 0.85, so M5 is descriptive and reads "cannot test" (plan rule M1(b)).
(c) R3 pools at T, positive = acquired_public row dated in (T, T+36 months]
| pool | T=2020 pos/neg | min AUC | T=2022 pos/neg | min AUC | use |
|---|---|---|---|---|---|
| trial-bearing, listed at T, not terminal at T | 33 / 373 | 0.647 | 52 / 438 | 0.619 | primary (ruling) |
| trial-bearing, any | 36 / 1,683 | 0.636 | 54 / 1,947 | 0.612 | descriptive secondary |
| all companies, any | 75 / 15,291 | 0.594 | 98 / 15,228 | 0.582 | descriptive secondary |
| trial-bearing, no prior ledger row of any kind | 3 / 1,269 | 0.968 | 2 / 1,459 | 1.000 | dropped: degenerate |
The ledger label can only fire for a listed company, so a pool with private companies largely measures "was public". All 266 targets have another ledger row, which is why the literal "no other exit" pool collapses.
(d) Targets with no trials
142 of the 266 targets have no trial row on or before the data end. They carry no stage evidence and leave the primary R3 pool.