Pip

What Pip does

Polymarket is a prediction market — people bet real money on whether real-world events will resolve YES or NO. The edge Pip plays is not predicting better than the crowd — it’s reading the resolution criteria more carefully than the crowd reads it.

A market like “Will [politician] win Iowa caucus?” looks simple. But the resolution doc specifies exactly which source counts, what timestamp the market closes at, what counts as a “win” (majority of delegates? plurality? media call?), and what happens if there’s a contested result. Reading those details changes the actual probability of YES vs NO often enough that an attentive reader has an edge.

Pip’s job is to be that attentive reader, every 15 minutes, across the whole market.

How the decision loop works

Four LLM roles, deterministic gates between them:

1. Analyst (Haiku) — scans the market list, filters by liquidity, expiry, category. Cheap fast pass.
2. Researcher (Sonnet) — for each candidate, reads the resolution criteria document carefully, fetches recent news, builds a probability model with uncertainty bounds.
3. Trader (DeepSeek) — given the probability + current market price, decides whether to enter, what size, what limit.
4. Risk (Sonnet, gating) — 14 gate checks before any trade fires (max exposure per market, daily VaR cap, sector concentration, drawdown stop, blacklist screen, etc).

Every gate decision is logged. Every trade has an explanation tied to a specific resolution-criteria insight. The whole loop runs every 15 minutes via launchd.

Where the data lives

• Per-trade decisions in Supabase pip_trades, pip_gates, pip_research_cache
• P/L mark-to-market every 15 min in pip_marks
• Public daily P/L mirror at acridautomation.com/pip (read-only operator-facing)
• Telegram alerts on every fill, every gate rejection, every drawdown stop
• Weekly Sunday sweep recovers stuck positions + posts a weekly P/L digest

What’s getting better over time

Pip’s research cache is the moat. Every market it has analyzed leaves a trail — the resolution-criteria reading, the news fetch, the model assumptions, the eventual outcome. When a similar market opens later, Pip reads the trail before deciding. The longer Pip runs, the better its first-pass screening.

Planned: weekly pattern-mine on which research insights led to wins vs losses. Currently the trail exists but isn’t actively mined.

Live trading status

Paper mode. Capital tracking is virtual. Operator-set HALT flag at agents/pip/HALT stops all activity instantly (currently set since 2026-05-10 while operator finalizes the wallet + KYC pathway for live deployment).

When live trading turns on, Pip starts with ~$200 of real Polymarket capital — about a third of what it earned in compressed paper P/L over the prior week. Conservative ramp.

Why it’s not buyable yet

Pip is a trading agent, not a content agent. Selling Pip means selling either a managed-trading service (we run Pip on your capital, take a fee) or selling the research framework (resolution-criteria reading as a methodology). Both are real paths. Both require regulatory shape (Pip handles money). Neither is shipped.

If it’s interesting, drop your email — first managed-Pip pilot ships after live trading proves out.