The read — trailing window 2026-10-18 to 2026-10-19
The analyzed corpus was thin this window: 36 items, against 67 in the prior two-day window. Broad-market coverage tagged crypto_market leaned bearish (5 items); Bitcoin's own coverage leaned constructive (5 items), as did Solana's (3), while Ethereum's leaned bearish (3).
One new asset entered the book: XRP. Its corpus coverage over the past seven days is three items from two sources (an analyst and an insider-kind source), all three bearish, and its price momentum also points lower. That is the only coin this run where a fresh call cleared the conviction gate and had room in the book.
New calls (3)
| Call | Horizon | Engine probability | Base rate |
|---|---|---|---|
| XRP — leans to reach −2.31% within the next 7 days | 7d | ~84% | ~69% |
| XRP — leans to reach −0.87% within the next day | 1d | ~82% | ~66% |
| XRP vs Bitcoin — leans to finish at or below Bitcoin's return over the next 90 days | 90d | ~86% | ~75% |
A touch call grades on whether price reaches the stated percentage at any point in the window; the complement is "does not reach it", not "moves the other way". Touch base rates sit high by construction, so a ~60–85% figure is a units fact, not a confidence claim.
Correlation caveat. Every standing asset call in the book now leans down (BTC, ETH, SOL, ZEC, XRP). They largely ride one market move, so the book holds far fewer independent reads than its row count suggests — the five hits reported in the previous daily all came from that same move.
What was not issued, and why
- ETH, SOL, ZEC — convicted at 7d, 3d and 1d, but each already holds two standing asset calls (the per-coin limit of two).
- XRP 3-day — convicted, but the experimental 3-day lane is at its cap of three (SOL, ETH, ZEC), and XRP's two per-coin slots are taken by the 7d and 1d calls above (horizon order: 7d, then 1d, then 3d).
- BTC — no directional read at any horizon this run (the engine returned sideways).
- HBAR, DOGE, SUI, ADA, AVAX, BNB and others — larger raw scores, but none cleared the conviction gate.
- Regime — the standing choppy read (issued 2026-10-16) runs until 2026-10-23; one regime at a time.
- Bitcoin new-high question — two reads already stand (the per-question limit of two).
- Altseason — re-scored, not materially different from the standing read, so nothing appended.
- Majors-drawdown question — the scorer did not emit a read.
- SOL vs Bitcoin — re-scored, not materially different from the standing read. ETH vs Bitcoin holds two standing reads (the per-coin limit). ZEC vs Bitcoin was not scored while an open engine question on that read is unresolved.
Receipts — resolved since the last run
| Call | Issued | Matured | Resolved | Outcome |
|---|---|---|---|---|
| Bitcoin (BTC) — leans to reach −0.90% within the next day (we had it at ~62%) | 2026-10-18 17:45Z | 2026-10-19 17:45Z | 2026-10-19 17:47Z | Miss — the −0.90% level was not reached; BTC closed the window +0.96% |
One call resolved, one miss. The five hits from the previous run were reported in the previous daily.
Accuracy — trailing 30 days to 2026-10-19 (grading epoch 5, touch)
- Headline: Brier Skill Score vs the base rate = −0.0034 (n = 10, 7-day claim pool). Our probabilities have not yet beaten simply quoting the historical base rate — roughly level with it, and slightly worse. The sample is below the 25-call floor, so the verdict is insufficient data, not a skill finding.
- Why: reliability 0.0375 (where over-confidence shows up; lower is better) against resolution 0.045 (whether our probabilities discriminate; higher is better) — the two nearly cancel.
- 1-day claim pool: BSS −0.170 on n = 24, after today's miss — clearly worse than the base rate so far.
- Directional-accuracy note (not a skill measure, it ignores the probability we publish): hit rate minus mean base rate on the 7-day pool is −0.0113.
The engine's probabilities are uncalibrated and provisional; the record above is negative and is reported as such.
