Waiting for held-out evaluation.
Numeria Labs / Currency-focused autoresearch engine
CashClaw turns public signal drift into operator-reviewed Kalshi trades.
A live research loop for finding mispriced contracts, stress-testing the thesis, and routing only the strongest candidates into tiny-risk human-approved execution.
Current posture
Loading readiness
Reading forecast lift, calibration, and live gate.
The trade thesis
Markets move slower than public evidence.
CashClaw looks for the gap between where a Kalshi market is priced and where the evidence points after social velocity, source reliability, orderbook pressure, and reference prices are reduced into one currency-aware view.
Public evidence
Qdrant memory
Autoresearch
Kalshi review
What the current run says
It appears profitable in held-out replay, with strict caveats.
Waiting for optimizer.
Held-out Brier lift.
Adversarial suite.
Benchmarks are used as held-out evaluation and gate evidence. They are not training labels for the live strategy.
How the engine improves
Autoresearch is the operating system, not a one-shot model.
1
Fetch Kalshi markets
Active contracts, resolution windows, orderbooks, fees, spreads.
2
Reduce public evidence
Twitter, Reddit, reference prices, and Qdrant memory become candidate-specific signals.
3
Search strategies
Ensemble, GEPA, adversarial synthetic checks, and journal learning compare algorithms.
4
Gate for action
Only operator-review Kalshi candidates can reach manual live review.
Live-money posture
Trade less. Trade only when the evidence survives the gate.
The system is configured for operator-reviewed, tiny-risk Kalshi manual review. No autonomous broad trading. No benchmark-label training. No order without fresh market revalidation.
5%+tradable edge
55%+confidence
1+source
1+directional evidence
Softwarnings only
Cappedcurrency risk budget
Live demo path
Deck, dashboard, white paper, and API now live together.
Start here for the pitch. Jump to the dashboard to inspect markets and readiness. Open the white paper when someone wants the method.
GET /api/live/status
GET /api/live/orders
POST /api/live/kalshi
Operator handoff
Kalshi execution is ready to connect where the account credentials live.
Run loop
npm run loop:kalshi
Check the account
npm run live:kalshi -- --preflight --auth-check
Submit one operator-review batch
npm run live:kalshi -- --auto --submit --confirm CONFIRM_TINY_LIVE_KALSHI_ORDER
The next outcome
Prove the edge on tiny live capital, then scale only what survives settlement.
CashClaw is built to keep learning, keep rejecting weak trades, and make the operator-review path easy to inspect.