# Eye check, 2026-10-21 (Tyler)

## Part 2: 28 blind model decisions (1870s fall window, judged from excerpts)

| | Count |
|---|---|
| Agree with model | 22 |
| Disagree | 2 |
| Tyler "can't tell" | 4 (3 are garbled ad or theatre listings the model rejected; 1 is a one-line "hissing hot" Centennial item the model accepted) |

- Agreement on decided items: **22/24 (92%; 95% Wilson CI about 74-98%)**.
- The 2 disagreements:
  - **D16, model wrong:** the Peshtigo fire of 1871, with "balls of fire fell like meteors". A figure of speech the model
    accepted as a sighting (a false positive, which costs reading time, not recall).
  - **D18, excerpt problem:** the page excerpt cut off at "A meteoric stone weighing…". The full article reports the
    5,000-lb stone said to have fallen near Mourzouk, Fezzan, which our rules count. The fix is to excerpt by sentence
    boundaries, not characters.
- **Misses:** Tyler found none among the 14 items the model rejected (11 agreed no, 3 can't tell).

## Part 1: 12 strongest fall leads (Tyler checked the scans)
- 11 judged credible on reading.
- **Wakefield, Mass., 1911** judged doubtful: reads like fiction; headline "House Struck by Meteorite".
- Flags raised:
  - **New London, Ohio, 1873:** the stone was "quite hot" 15-20 minutes later. Freshly fallen meteorites are usually
    not hot, so this is a known folklore marker. Treat it as a caution signal, not a disqualifier.
  - **South Bend, Ind., 1876:** the headline "What became of that meteor" implies earlier coverage. Search backward
    for it.
  - **Jonesboro, Ill., 1914:** thin, with little surrounding detail.

Credible-on-reading is not proof (see the credibility signals in `results/falls/leads.md` and the follow-up plan).