wingtrade
Quiet is discipline
The options terminal that shows you its losing trades.
An automated ETF-options bot that enforces its risk rules in code and publishes every decision it makes — the exits that lost money, and the thousands of setups it refused to touch. Most tools show you the wins. This one shows you the log.
Both ends are known before it trades
Every position is a spread with a capped, known-at-entry maximum loss. There is no configuration in which a single trade can take an unbounded bite out of the account, because the shape of the position does not permit it.
Above that sit ladders of limits — per trade, per correlated cluster, per day, per week — and they only ever tighten. A limit in this system has no path that loosens it under pressure, which is precisely when a human would.
Discipline, with receipts
Signals, refusals, fills, slippage, exits — recorded as they happen, with the rule that fired. A quiet week is the system working, and the log is how you can tell that from a system that broke.
Code guardrails, not prompt guardrails
The risk limits are not instructions we hope a model follows. They are breakers compiled into the system, on a path that can only reduce exposure. A model cannot talk its way past a check it does not run.
Paper first, gates before money
It runs on a paper account and graduates only by clearing defined evidence gates — one of which asks how its own P&L was priced, because a modelled exit and a broker fill are not the same evidence. No gates cleared, no graduation, and we publish that too.
What it actually did
lifetime paper record · robo ledgerrisk
expected_value
quality
trade_desk
strategy
Where these come from. Counts are the rejection rows in the bot's own decision log, taken from the final database export and grouped by the gate that rejected the candidate. Cheap tests run first, so most candidates never reach the expensive ones — which is what the shape above is. Cycle-level stand-downs are deliberately excluded: a cycle that paused before reaching a setup did not refuse one, and counting it would grow this number with uptime rather than with work. The full evaluated / taken / refused figures publish alongside the SHA-256 of the export they were derived from, so any number here can be traced to the exact file — and we would rather show you five sourced counts than one impressive unsourced one.
What would change our mind
A refusal record is only worth something if it can be wrong, so the standard is stated in advance: if the setups this system declines go on to outperform the ones it accepts, the gates are miscalibrated and the logic is wrong — not unlucky. The log is kept in a form that can show that, and it gets published either way.
A note for the other kind of reader
Some people who get this far are not shopping for a trading system. They are looking at the layer underneath it: gates that fail closed rather than open, a hash-chained decision log, per-rule attribution on every refusal above, and a provenance stamp on every number saying whether it came from a model or a fill.
That layer does not know it is trading options. It is a way of making an automated decision defensible after the fact, and the same shape applies wherever a machine acts and someone later has to show why. We built it because we needed it, and we are aware that is not the same as having built a product.
If that is the part you came for — tell us that instead, and say what you would need it to do. We would rather hear it early than guess.
Follow the build
A weekly desk note: what the machine did, what it refused, and what it got wrong. No signals, no tips, nothing to buy — and the boring weeks go out too, because those are the ones that tell you anything.
One email a week. Unsubscribe in one click. We do not sell the list.