Last Updated: 2026-10-26
The earlier seven-bot ranking did not provide repository URLs, owners, pinned commits or the records behind its claimed returns. It cannot support an identifiable shortlist, so those recommendations are withdrawn. This page is an assessment worksheet, not a substitute list of newly invented projects.
Establish what you are reviewing
Copy the exact owner/repository URL, full commit identifier, licence and relevant source paths. A project name, screenshot or star count is not enough to distinguish a maintained project from a similarly named copy. GitHub documents permanent links to a file at a commit; a moving branch link may show different code later.
Record the runtime, manifest, lockfile, install hooks and external services. GitHub's dependency graph can help inspect dependencies, but a clean alert list is not a security audit. Do not run installation scripts just to find out what they do.
For platform compatibility, start with Polymarket's official documentation, including the separate US documentation linked there. Identify the relevant API, authentication method, order lifecycle and product. A public market-data response does not establish eligibility or permission to trade.
Separate observation from execution
Scroll sideways to see every column. Keyboard: focus the table region and use the arrow keys.
| Mode | Expected access | Evidence required |
|---|---|---|
| Read public market data | Public endpoints only | No signer or order submission path invoked |
| Replay recorded events | Local fixtures | Fixed inputs, timestamps and reproducible output |
| Simulate orders | Synthetic balances and fills | No live transport; explicit fill and fee assumptions |
| Place real orders | Account authority | A separate operational review; not certified by the first three modes |
A “dry-run” flag is only a claim until its call paths are inspected. A small funded order is live trading, not paper trading. Do not supply a wallet seed or production key to an unidentified project.
Test the failures before interpreting a return
Use fake balances and a stub order service. These examples describe expected evidence, not commands to connect an account.
- Unknown order outcome: the stub records an accepted order but drops the response. The next action must reconcile that order, not silently create a second one.
- Partial fill: 100 shares requested, 40 filled, 60 still open. The journal and exposure calculation must represent both portions.
- Stale data: replay an old book update after a newer one. The system must not treat it as a fresh opportunity merely because it arrived last.
- Restart after a halt: trigger the configured stop, then restart the process. A stop that vanishes on restart is not a durable control.
- Cancellation failure: stopping the process does not prove its venue orders were cancelled. The state should remain unresolved until acknowledged or reconciled.
- Related markets: two contracts tied to the same event may concentrate risk even when they have different IDs. Exact resolution rules matter.
For each case keep the input fixture, expected result, actual log and source commit. A green unit test is useful only for the behavior it exercises; it does not establish profitability, security or reliable live execution.
A candidate record you can actually compare
Repository and owner:
Commit and review date:
Licence / runtime / lockfile:
Public-data-only entry point:
Signing and order-submission entry points:
Inventory and unresolved-order accounting:
Persistent halt and restart behavior:
Failure fixtures run / not run:
Costs included / omitted:
Unresolved questions and decision:“Needs review: sends a second order after a timeout” is more useful than an 8.7/10 score. A project with no resolvable identity stays unassessed; an identified project with missing controls is not promoted because it has more stars.
If a performance record is later supplied, require every fill, open position, transfer, fee, infrastructure cost and evaluation period. Distinguish a backtest from forward simulation and live activity. One positive month cannot establish a predictable monthly return. The equity journal shows why closed-trade profits alone are insufficient.
Trading automation can lose funds. No repository here has been installed, connected to an account or endorsed for production. AITR may earn referral commissions elsewhere; this worksheet contains no paid-bot recommendation.