Strategy grammar

Every TinyCorp Signal strategy is a JSON config. The Builder lets you assemble it visually, but the raw shape is documented here so power users can fork-and-edit. The engine runs your config against live market data every 5 minutes and logs simulated fills.

Overall shape

A strategy config is a JSON object with five top-level keys:

{
  "platforms": ["kalshi", "poly", "manifold"],
  "categories": ["weather", "sports", "crypto", ...],
  "entry": { /* all filter fields below go here */ },
  "exit":  { "at": "expiry" },
  "sizing": { "type": "flat", "amount_cents": 500 },
  "side":   "YES"
}

Every cycle (every 5 minutes), the engine fetches markets matching platforms + categories, then applies the entry filter rules in order. Matching markets get a simulated fill at the current ask.

platforms + categories

Price filters

Liquidity filters

Timing filters

Momentum signals

Price-delta filters that require a lookback window. Engine reads from market_price_snapshots (cached every 1-3 min per platform).

// "Buy when ya dropped 5+¢ in the last 30 min"
{
  "yes_ask_min": 0.40,
  "yes_ask_max": 0.85,
  "ya_change_max_cents": -5,
  "ya_change_window_min": 30
}

Volume-spike signals

Calendar gates

Short-circuit filters that fire before any market query. Use for strategies that should only run during specific times.

External anchors

Signals that reference external data sources we poll.

Exit + sizing

Backtesting

The backtester replays a strategy config against captured market-price history (market_price_snapshots, recorded ~every 5 minutes) and settles each simulated trade with the same exit math the live engine uses. It answers: "how would these rules have done over the data we've recorded so far?"

The snapshot window grows continuously as history accrues. Delta signals (ya_change_*, vol_spike_*) are replayed with a sliding per-ticker buffer that mirrors the live engine's lookback, so a backtest verdict matches what the cron-paced runner would actually have done. max_concurrent is applied as a post-pass over the time-sorted fills.

Signals the backtester faithfully replays

Signals not yet replayable

These work in the live engine but aren't backtested yet — they need historical series we don't capture retroactively. When a config uses one, the backtester flags it as unsupported rather than silently skipping it (so a backtest never quietly ignores part of your rules):

How we measure performance

Every number on a strategy card, the leaderboard, and the detail page is computed the same way, in engine.perf_stats. All P&L is paper-traded at a flat $5 stake per trade — so "+$6" means +$6 on $5 bets, not on $50. Only closed trades (settled at expiry against the real market outcome) count toward P&L, win rate, EV, Sharpe, and drawdown; open positions are not marked-to-market. Headline figures are lifetime (every closed trade, across all versions of the strategy) unless a surface explicitly says "30d".

Full examples

Fade weather brackets in the late window:

{
  "platforms": ["kalshi"],
  "categories": ["weather"],
  "entry": {
    "yes_ask_min": 0.85,
    "yes_ask_max": 0.95,
    "hours_to_close_max": 18
  },
  "exit": { "at": "expiry" },
  "sizing": { "type": "flat", "amount_cents": 500 },
  "side": "YES"
}

Multi-signal: weekday US morning Poly mover with vol spike:

{
  "platforms": ["poly"],
  "categories": ["other", "politics", "economics"],
  "entry": {
    "yes_ask_min": 0.25,
    "yes_ask_max": 0.75,
    "vol_24h_min": 10000,
    "vol_spike_ratio_min": 1.3,
    "ya_change_window_min": 60,
    "hour_of_day_min": 13,
    "hour_of_day_max": 17,
    "day_of_week_in": [0,1,2,3,4]
  },
  "exit": { "at": "expiry" },
  "sizing": { "type": "flat", "amount_cents": 500 },
  "side": "YES"
}

API reference

House strategies using each signal

Live snapshot from /api/quants/strategies. Click any strategy name to see its full config + live performance.