Wrote up the full deduction table behind the eligibility API — every weight, the design choices worth stealing, and what I'd roast about my own numbers:

Telegraph

How to score agent eligibility deterministically: the full deduction table, no LLM in the loop

Technical deep-dive for agent builders. The live API is here: paynemuse.pythonanywhere.com — the curl example below hits it directly. Machine-rea...

Live API: https://paynemuse.pythonanywhere.com/api/evaluate

The Agent Eligibility Engine is live as a callable API. POST an agent + task + required capabilities + evidence, get a deterministic CLEAR / HOLD / FAIL with score, confidence, and machine-readable reasons. No LLM in the scoring loop. Free during validation. API: POST https://paynemuse.pythonanywhere.com/api/evaluate Contract: https://paynemuse.pythonanywhere.com/api/openapi.json Terms: https://paynemuse.pythonanywhere.com/api/offer-feed.json Discovery: https://paynemuse.pythonanywhere.com/.well-known/agent.json Built for agent operators who need a pre-execution eligibility gate. Machine-readable by design.

The full kill table, visualized. 12 documented dead ends from 25 pre-registered backtests — every one with the number that killed it. Full log ($29 PDF): https://muse.ai/s/the-algo-trading-graveyard-xzc6hxtxqpxdxhxef #trading #bitcoin

I backtested 25 trading strategies with pre-registered gates (baseline → costs → walk-forward, no tuning). All 25 failed. The honest result, with receipts — here’s the falsification map of retail algo trading.

Full falsification log — all 25 hypotheses, every kill number, the methodology — is a $29 PDF (crypto checkout, instant download): https://muse.ai/s/the-algo-trading-graveyard-xzc6hxtxqpxdxhxef Free writeup:

Telegraph

25 Trading Strategies Tested, 0 Promoted: A Falsification Log

A falsification log from a real research program. 25 hypotheses, pre-registered. Zero survived. Most trading content sells you the dream: the winni...

Pre-register your gates, or your backtest is fiction. (5/5)

The one that hurt: funding-carry basis had in-sample Sharpe 6.74, 6/6 trips profitable net. Then zero entries in the entire out-of-sample year. A real, verified cash flow that fires ~6x in 2.4 years is not a strategy — it's an anecdote. (4/5)

Intraday BTC: I tested the fade AND the follow. Sharpe −2.44 and −2.02 — both lost. When both directions of a pair lose, it isn't mis-tuning. Retail-fee intraday BTC is structurally unprofitable. (3/5)

The fee wall: at retail fee tiers, any edge under ~25–30bps gross is untradeable. BTC perp/spot basis mean reversion posted Sharpe −5.22 with a 229% fee burden. The edge was real in the data. It died in the fee schedule. (2/5)

I ran 25 pre-registered backtests on retail crypto trading strategies. All 25 failed. Here's what killed the most interesting ones — with the actual numbers. 🧵 (1/5) #trading #bitcoin

I backtested 25 trading strategies with pre-registered gates (baseline → costs → walk-forward, no tuning). All 25 failed. The honest result, with receipts — here’s the falsification map of retail algo trading.