PaperBook · the honesty engine

Your backtest is lying to you.We tell you by how much.

PaperBook is a paper-money central limit order book for crypto up/down binary markets (BTC, ETH, SOL, XRP, BNB, DOGE × 5m and 15m windows). It speaks Polymarket's CLOB wire format — the same headers, the same L2 HMAC, the same order envelope — so py-clob-client and existing Polymarket bots work unchanged. Registering grants $10,000 of paper money. No real funds, ever.

The simulator replays a real recorded Polymarket tape (captured continuously since 2026-05-29) with honest fills: your taker orders cross the book that actually existed, after measured latency. At the end you get a verdict, not a vibe — the edge your backtest promises versus the edge the tape would actually fill, with a data-quality certificate and a reproducibility hash.

The blog is a strategy graveyard: engineering post-mortems of our own trading strategies, each with the exact statistical reason it died — adverse selection, pseudo-replication, lookahead, corrupted ground truth. We built it, we tested it live, and we published the result even when the result was a loss. Built for humans and for AI agents: every surface ships a paste-ready brief an agent can build a bot from.

— latest from the blog: Be the House, They Said. Full graveyard at /blog.

Supported markets

Every coin trades both windows — six coins, two windows, twelve live markets at any moment, continuously replayed from a real Polymarket day.

Coin5m window15m window
BTCYesYes
ETHYesYes
SOLYesYes
XRPYesYes
BNBYesYes
DOGEYesYes

Who's behind this

PaperBook is built and operated by Sofyt, a small software studio — see sofyt.mx for the rest of what our engineering team builds, including LLCrawler, the LLM-visibility auditor that generated the fixes on this page. Source and integration tooling: github.com/SofytMX/paperclob-skill. Questions or integration help: [email protected].

FAQ

Is this real money?

No. PaperBook is a paper-money environment — registering grants $10,000 of simulated balance. No real funds are ever involved, and no position ever touches a real wallet.

Is PaperBook affiliated with Polymarket?

No. According to Polymarket's own CLOB API documentation, PaperBook is wire-compatible with it — same headers, same L2 HMAC, same order envelope — so existing bots, including py-clob-client, work unchanged. It is an independent, unaffiliated simulator.

What does the simulator actually grade?

It replays a real recorded Polymarket tape and fills your taker orders against the book that actually existed, after measured latency. The report compares the paper edge a naive backtest would show against the fillable edge the tape would actually pay, and calls the gap between them the ghost_gap.

Can an AI agent use this without a human in the loop?

Yes. /llms.txt and /llms-full.txt contain paste-ready specs — auth, endpoints, order shapes, the simulator's virtual clock, and a validation checklist an agent can run to prove its integration works.

Who builds this, and why publish your own losses?

PaperBook is built by Sofyt's engineering team, a small studio that also built LLCrawler, the LLM-visibility auditor. Every strategy in our own trading catalog has graded phantom or no_edge on the honest tape — we publish the post-mortem anyway, because the graveyard is where the actual lessons live.