# Kepler Bonus extension and matched null control (2026-10-11)

Scope: this applies Codex's own Kepler Bonus pilot code (`evidence/codex/database_expansion/kbonus/pilot.py`, logic unchanged) to more of the same frozen selection. It is a search extension, not a new validated result.

## What was run
- Selection: the same frozen rule as the Codex pilot (eligible background M dwarfs, Kepler primary targets excluded, ranked by Gaia G magnitude before any photometry was seen). Codex searched ranks 1-64. This extension searched ranks 65-576.
- 512 stars selected, 497 searched, 15 skipped for fewer than 6 usable quarters (`stars_searched.csv`).
- Same split rule: last ceil(25%) usable quarters reserved before the period search. 0.5-25 d band. Same first screen.
- A matched inverted (flux = 2 - flux) run was done on the same 497 stars with identical processing (`signals_real_and_inverted.csv`, column `inverted`).
- Scripts: `batch2.py` (real), `inv.py` (inverted). They import `pilot.py` from the Codex pack. Environment here was Python 3.10 with unpinned recent numpy/scipy/astropy/lightkurve, not the Codex lock file.

## Result
| | stars searched | signals passing first screen |
|---|---|---|
| real | 497 | 10 signals on 9 stars |
| inverted null | 497 | 10 signals on 9 stars |

The real run did not produce more first-screen passes than the null. Six of the nine real passing stars have no inverted pass, but two (KIC 7436275 at 0.614 d, KIC 6206697 at 0.567 d) pass in both polarities, and KIC 5790842 (4.87 d) passes in both. Those are variability or systematics, not transits.

Nothing here approaches KIC 5689121 (BLS SNR 26.7, holdout 15.5 in the same pipeline).

## Individual real passes
| KIC | P (d) | depth ppm | holdout SNR | status |
|---|---|---|---|---|
| 9851000 | 8.482 | 3243 | 12.2 | Rejected: matches KOI K04879.01 (KIC 9850966, 8.4803 d, false positive, 58 arcsec away) |
| 6629201 | 2.638 | 697 | 16.0 | Unresolved, weak. Faint (G 17.9), no KOI/TCE match, no localization done |
| 7436275 | 0.614 | 252 | 8.9 | Rejected: also passes inverted |
| 6206697 | 0.567 | 581 | 10.8 | Rejected: also passes inverted |
| 5790842 | 4.867 | 2345 | 6.1 | Rejected: also passes inverted, deep |
| 6604266, 6266847, 9896737, 8758092 | 0.5-1.0 | 390-1380 | 4.2-13 | Ultra-short; same regime where nulls pass; not leads |

## Limits
- No pixel/PRF localization was done on any of these. No Gaia-neighbor, TRICERATOPS or formal false-positive analysis.
- The catalog check for KIC 9851000 and 6629201 used the Exoplanet Archive KOI/TCE tables only.
- The strict-screen analysis (period > 1.5 d, 12.5-14.5 d excluded, holdout SNR >= 5, event-sign and single-event limits) was tuned after seeing these results, so it is descriptive, not a clean test.
- No candidate was named, submitted or announced.

## Second stage: ranks 577-1800 (added 2026-10-11)
Same code, same frozen ranking (`batch2.py --lo 576 --hi 1800`, inverted with `ranks577-1800/inv3.py`). Tables in `ranks577-1800/`.

- 1224 stars selected: 1175 searched, 49 not searched (48 with fewer than 6 usable quarters, 1 MAST read timeout, not retried).
- Matched inverted run on the same 1175 stars.
- First-screen passes: real 19 signals on 18 stars, inverted null 22 signals on 19 stars. No excess over the null.
- With a stricter screen (P > 1.5 d, outside 12.5-14.5 d, holdout SNR >= 5): real 6 signals on 5 stars, inverted 7 signals on 6 stars. This screen was chosen after seeing the earlier batch, so it is descriptive, not a clean test.

Real strict passes and what checks say (NASA Exoplanet Archive TCE/KOI cone, ~108 arcsec, same period or 2x/0.5x):
| KIC | P (d) | depth ppm | result |
|---|---|---|---|
| 4544570 | 2.189 | ~5200-6950 | Rejected: four KOI false positives at the same 2.189 d within 85 arcsec |
| 11724208 | 5.746 | ~5600 | Rejected: KOI K06244.01 (KIC 11724210, 5.746 d, false positive, 9.6 arcsec away) |
| 12164160 | 19.42 | 5398 | No catalog match. G 18.4 background star with a 0.5% depth: too deep for a planet-size dip on an M dwarf, so likely an uncataloged blend or eclipsing binary |
| 8743279 | 21.96 | 9917 | Same: G 18.6, ~1% depth, likely a blend/EB |
| 8803342 | 5.010 | 2928 | Same: G 18.8, likely a blend/EB |

No pixel/PRF localization was done on these. None is a candidate, and nothing was named or submitted.

## Combined picture
Across ranks 65-1800 (1,672 stars searched, plus Codex's 64) the real search never produced more first-screen passes than the matched inverted null. The only strong signal remains KIC 5689121 from the original Codex run.
Totals: ranks 65-576: 497 searched; ranks 577-1800: 1175 searched; sum 1,672. One unresolved weak item from the first stage: KIC 6629201 (2.638 d, faint, unlocalized).