Signal Backtests
How Orrery's price-based signals performed historically, replayed on the top 15 markets by volume. Transparent methodology, no cherry-picking.
History
1M CLOB
Universe
Top 15
Cooldown
1h
Cost model
-2pp
Orrery replays the same live rules minute by minute, then measures 1h, 6h, and 24h follow-through. Flow and resolution-risk need event-time history, so they are excluded for now.
Momentum
328 firings across 6 marketsPrice + 1h direction + 24h direction all aligned, with ≥3pp 24h move. Expected to continue in the same direction.
Expected · 24h, after spread
-3.58pp
raw avg − 2pp typical spread
Median · 24h
+0.00pp
robust to outliers
Win rate · 24h
46%
304 measured
Raw avg · 24h
-1.58pp
before spread / fees
Average move by horizon
- 1h-0.05pp
- 6h+0.03pp
- 24h-1.58pp
Centred at zero. Bar length is proportional to the largest of the three windows, so the decay shape is visible without re-scaling.
By category (3 categories fired)
| Category | Firings | Measured | Win 24h | Avg 24h |
|---|---|---|---|---|
| Geopolitics | 192 | 187 | 40% | -2.64pp |
| Politics | 132 | 114 | 57% | +0.02pp |
| Sports | 4 | 3 | — | -3.17pp |
Win rate hidden until ≥10 measured firings — thin samples can flip with one outlier.
Per-market breakdown (6 markets fired)
| Market | Firings | Win 24h | Avg 1h | Avg 24h |
|---|---|---|---|---|
| US x Iran permanent peace deal by June 15, 2026? | 106 | 48% | -0.58pp | -2.72pp |
| Will Roberto Sánchez Palomino win the 2026 Peruvian presidential election? | 81 | 58% | +0.03pp | -0.07pp |
| US x Iran permanent peace deal by December 31, 2026? | 69 | 22% | -0.30pp | -2.14pp |
| Will Keiko Fujimori win the 2026 Peruvian presidential election? | 51 | 55% | -0.36pp | +0.15pp |
| Iran closes its airspace by June 8? | 17 | 67% | +5.17pp | -4.21pp |
| Spurs vs. Knicks | 4 | 0% | -1.17pp | -3.17pp |
Divergence
188 firings across 6 markets1h move running against the 24h trend. Expected to continue short-term in the 1h direction before 24h reasserts.
Expected · 24h, after spread
-1.72pp
raw avg − 2pp typical spread
Median · 24h
-0.50pp
robust to outliers
Win rate · 24h
38%
175 measured
Raw avg · 24h
+0.28pp
before spread / fees
Average move by horizon
- 1h+0.58pp
- 6h+0.93pp
- 24h+0.28pp
Centred at zero. Bar length is proportional to the largest of the three windows, so the decay shape is visible without re-scaling.
By category (3 categories fired)
| Category | Firings | Measured | Win 24h | Avg 24h |
|---|---|---|---|---|
| Geopolitics | 123 | 122 | 43% | +1.51pp |
| Politics | 58 | 46 | 26% | -2.38pp |
| Sports | 7 | 7 | — | +0.86pp |
Win rate hidden until ≥10 measured firings — thin samples can flip with one outlier.
Per-market breakdown (6 markets fired)
| Market | Firings | Win 24h | Avg 1h | Avg 24h |
|---|---|---|---|---|
| US x Iran permanent peace deal by June 15, 2026? | 69 | 40% | -0.50pp | +0.13pp |
| US x Iran permanent peace deal by December 31, 2026? | 48 | 44% | +0.44pp | +1.56pp |
| Will Keiko Fujimori win the 2026 Peruvian presidential election? | 32 | 26% | +1.84pp | -2.91pp |
| Will Roberto Sánchez Palomino win the 2026 Peruvian presidential election? | 26 | 26% | +1.99pp | -1.74pp |
| Spurs vs. Knicks | 7 | 43% | +1.29pp | +0.86pp |
| Iran closes its airspace by June 8? | 6 | 67% | +0.36pp | +16.87pp |
Honest caveats
- Small samples. CLOB returns ~30 days of minute-level history; short-dated markets may have very few firings. Anything below 10 measured firings shows a dash instead of a win rate so we don't overstate confidence.
- Survivorship. This backtest runs on markets currently live — not on markets that already resolved and disappeared. Persistent historical coverage is planned so the full universe can be evaluated once enough live history has accumulated.
- No execution costs. Avg-move figures are mid-price to mid-price. Real entry/exit would pay spread, slippage, and Polymarket fees — so a +1.5pp 24h move is not +1.5pp of profit.
- Signal drift. If we tune the live rule, these numbers change with it. That's the point — the page always reflects the rule that's currently running, not a frozen historical version.