NO on High-Confidence Favorites (YES 0.85–0.95)

Buying NO against an EvidenceDataset (949 rows). 47 trades fired over the test window — +59708.16% total return, 4.0% max drawdown.

Equity curve

$10,000 → $5,980,816.14 (59708.16%).

How to readThe line is this strategy’s account value over time, starting from $10,000. Each step is a position resolving at $1 or $0 per share; the horizontal baseline marks the starting capital, so anything below it is a net loss on the window.

Metrics

Avg trade return: +890.25%
Sharpe: 3.04
Max drawdown: 3.96%
Win rate: 80.85%
Trades: 47

All stats (7 more)

Edge vs crowd: +0.00%; no trades fired
Compounded (fixed-fraction): +59708.16%; reflects sizing, not per-trade edge
CAGR: +3098.68%
Sortino: 82.89
Probabilistic Sharpe (PSR): 100.00%; P(true Sharpe > 0)
Min track record (MinTRL): 18.4131; daily observations needed for 95% significance
Expected Calibration Error: 72.18%; lower is better; 0% = perfectly calibrated

DrawdownPeak-to-trough 3.96%.

Drawdown

Peak-to-trough 3.96%.

How to readDrawdown is how far below its previous peak the account sits at each point; it touches 0% at every new high and dips during losing stretches. The deepest dip is the worst loss you’d have had to sit through.

Drawdown distribution

Worst drawdown across 1,000 reshuffles of the trade order — the range of bad luck the same trades could have produced.

How to readThe band is the p5–p95 range of worst-drawdown-so-far at each step; the dashed line is your drawdown tolerance. A band that stays below the line means the strategy rarely breaches it.

Median worst drawdown · shaded bands = 25–75% and 5–95% of resampled paths

Robustness

Robustness checks

Sharpe across a grid of entry threshold × sizing fraction. The outlined cell is this run; a broad favourable neighbourhood means the result is not pinned to one lucky setting.

How to readEach cell re-runs the strategy with that entry threshold and sizing fraction against the same dataset and costs. Warmer = higher Sharpe; blank cells fired no trades.

		Sizing fraction →
		0.01	0.01	0.01	0.02	0.03	0.04	0.05
Entry threshold	0.02	3.4642	3.4642	3.4642	3.355	3.2778	3.2195	3.1735
	0.03	3.2701	3.2701	3.2701	3.202	3.1458	3.0989	3.0591
	0.04	3.273	3.273	3.273	3.2454	3.2144	3.1832	3.153
	0.05	3.056	3.056	3.056	3.0393	3.0171	2.9932	2.9696
	0.06	2.9121	2.9121	2.9121	2.899	2.8778	2.8545	2.8311
	0.07	2.7778	2.7778	2.7778	2.7594	2.7404	2.7216	2.7035
	0.08	2.4735	2.4735	2.4735	2.4706	2.4667	2.4621	2.4572

WeakerStronger Sharpe this run’s setting

Best Sharpe (49 configs tried): 3.46
Expected best by luck alone: 0.04
Deflated Sharpe (DSR): 100%

Trying 49 configurations would produce a best Sharpe of ~0.04by chance alone. The deflated Sharpe is the probability this run's best cell beats that bar — an edge that survives the search, not one created by it.

Trade log47 trades · 38 won · 9 lost

Trade log

Decision	Resolution	Entry price	Settle	PnL	Days held	Reason
2024-10-23	2024-10-30	0.0701	0	−$200.00	7 days	hold_to_resolution
2024-11-09	2024-11-16	0.1251	1	$1,370	7 days	hold_to_resolution
2024-12-18	2024-12-25	0.0746	0	−$223.39	7 days	hold_to_resolution
2025-01-20	2025-01-27	0.0936	1	$2,119	7 days	hold_to_resolution
2025-04-23	2025-04-30	0.0751	1	$3,218	7 days	hold_to_resolution

Assumptions

How this was run i

Data source i: Pancake Polymarket
Granularity i: 86,400 sec
Period i: 2024-07-15 → 2026-05-19
Bars i: 1,000
Gaps i: 0 (0 sec)
Price source i: mid
Slippage i: 10 bps
Fees i: 5 bps
Resolution i: realized
Starting capital i: $10,000
Sizing i: fixed_fraction: 0.02
Engine i: batter@0.10.3 Ran on batter
Compiler i: pancake-evidence-compiler@0.1.0
IR i: 0.1
Spec i: 0.1

Resolution settles at $1 or $0 per share with no slippage or fee. On-chain costs and UMA dispute risk are not modeled.
Benchmark = crowd-priced YES at the first bar, held to resolution. Not sizing-matched to your strategy.
Sharpe is annualized by √252 (trading-day convention) even though prediction markets run 24/7.

Spec

pancake-evidence-spec · rawv0.1

// pancake-evidence-spec v0.1 — spec_hash 297f32957a9b…
{
  "name": "NO on High-Confidence Favorites (YES 0.85–0.95)",
  "costs": {
    "fee_bps": 5,
    "slippage_bps": 10
  },
  "strategy": {
    "side": "NO",
    "entry": {
      "when": {
        "all_of": [
          {
            "gte": 0.05,
            "feature": "entry_price"
          },
          {
            "lte": 0.15,
            "feature": "entry_price"
          }
        ]
      }
    },
    "sizing": {
      "mode": "fixed_fraction",
      "value": 0.02
    },
    "yes_payoff": {
      "when": {
        "eq": 1,
        "feature": "resolved_outcome_numeric"
      }
    }
  },
  "spec_family": "pancake-evidence-spec",
  "spec_version": "0.1",
  "compiler_version": "pancake-evidence-compiler@0.1.0",
  "source_spec_hash": "297f32957a9ba294394fc780346b60a2ec10eac07541f847dc132bd366df19a6",
  "starting_capital": 10000,
  "evidence_dataset_id": "48b641d3-8f0a-4626-bc77-2b8bf1353c42",
  "schema_requirements": {
    "required_columns": [
      {
        "name": "market_link",
        "type": "string",
        "semantic_role": "market_link"
      },
      {
        "name": "decision_time",
        "type": "int",
        "semantic_role": "decision_time"
      },
      {
        "name": "resolution_time",
        "type": "int",
        "semantic_role": "resolution_time"
      },
      {
        "name": "entry_price",
        "type": "number",
        "semantic_role": "entry_price"
      },
      {
        "name": "resolved_outcome_numeric",
        "type": "number",
        "semantic_role": "resolved_outcome_numeric"
      }
    ]
  }
}

Evidence dataset i

Dataset: 48b641d3-8f0a-4626-bc77-2b8bf1353c42
Rows hash i: 55a2d27346b347ad4bc830a91d154cb04b8e4ef3f2650872a9916b749e19df5c
Schema hash i: d04a4c14f4951eb0fa269e86cd52cee1c770afad6c4e05561fc53faea7ef169d
Rows: 1,000
Decision window: 2024-07-15 → 2026-05-17
Resolution window: 2024-07-21 → 2026-05-19

Validation i

Schema match: pass
Look-ahead check: pass
Monotonicity check: pass
Range checks: pass
Future rows i: 0

Verified by the runner i

Structural: Schema match, Lookahead, Monotonicity, Range, Required columns. Runner math: Cash ledger, Fee application, Slippage application, Event ordering.

Accepted as agent-supplied i

Feature columns: none. Entry price source: not declared. Liquidity source: not declared.

Not modeled i

Market impact, Resolution lag, Resolver risk.