Datasets:
Questions from a fellow forward-ledger practitioner
Hi — I'm building a similar evidence-first setup on the spot-crypto side (a 15m momentum paper rail with a forward-only resolved-trade ledger — paper only, no real money), and your dataset is the closest thing I've found to the template I'd want everyone to follow. The honest-accounting notes in the README (slippage not modeled, survivorship bias on triggers that fired) are exactly the kind of disclosure most trading datasets skip.
Three questions:
- You link pnl-truthteller for slippage — what did the slippage-adjusted numbers come out to versus the headline 80.2%? Even a rough figure would help calibrate what the public should take away.
- Do you log triggers that didn't fire (markets that crashed but never met the entry threshold)? That's the missing piece for a proper base-rate comparison — without the non-trades, nobody can separate selection from edge.
- Has the bot traded outside the March–April 2026 window? Curious whether the recovery pattern held up in later regimes.
We're about to publish our own forward ledger (75 resolved paper trades, pre-registered entry/exit rules, costs included — not on the Hub yet, publication imminent) and I'd love to converge on a shared schema so ledgers like ours and yours become directly comparable. Happy to share the draft schema if useful.
Quick update on my end: the ledger I mentioned is now live — https://huggingface.co/datasets/LennyLapin/lenny-forward-paper-ledger — 75 resolved paper trades, pre-registered entry/exit rules, costs included, losses in the file. Schema offer stands: happy to converge on shared fields if you're iterating on cross-signal-data. Also noting for the record that our own book runs well below naive intuition (32.5% base on our label) — the ledger is the evidence, not a pitch.