ASkewMM_VWAP — status
Unguarded: PASSGuarded: PASSNeutralMMgunbot-quant
The same document for both runs of the same measurement — first as the strategy ships, then with the break-even guard. Everything else about the two runs is identical.
As it ships PASS
Exits: the strategy's own exits, stops included
Where it stands
| Correctness verdict | PASS |
| Cells measured | 15 (3 timeframes × 5 postures) |
| Failures | 0 |
| Warnings | 0 |
| Hand audit | not done |
| Single-file build | yes — a single self-contained file, no external modules |
| Measured artefact | the standalone build |
| Funding | 50% coin / 50% cash at the first bar (two-sided engine) |
Done
- Measured across 15 configurations on BTCUSDT, 30 days, after 0.030% fees per side and 2 bps slippage
- Checked against the matrix invariants and the per-cell order-path checks
- Posture settings measured for direction rather than assumed; inert settings named
- Built as a single self-contained file, verified to behave identically to its source
Owed
- A hand audit. Passing the invariants is not a line-by-line read. Presentation — log format, side panel, chart marks, colour contract — is not checked here at all.
- A longer window. 30 days on one pair is one sample of one market.
- Out-of-sample confirmation. Nothing here was measured on a live exchange.
What this measurement cannot tell you
- Spot, long-only, one price per bar, no order book. A strategy whose edge is the spread cannot be measured here at all — those are set aside rather than reported badly (see the audit records).
- Trade statistics count closed round trips only, so a position still open at the end is invisible to win rate and profit factor. The realised/unrealised split on the performance page is what exposes that.
- One window, one pair. A strategy can be fitted to it.
Every number on this page is read from the run record that produced it — the audit pipeline writes the page and the record together, so the two cannot drift apart.
With the break-even guard PASS
Exits: no exit below average entry plus fees, stops included
Where it stands
| Correctness verdict | PASS |
| Cells measured | 15 (3 timeframes × 5 postures) |
| Failures | 0 |
| Warnings | 0 |
| Hand audit | not done |
| Single-file build | yes — a single self-contained file, no external modules |
| Measured artefact | the standalone build |
| Funding | 50% coin / 50% cash at the first bar (two-sided engine) |
Done
- Measured across 15 configurations on BTCUSDT, 30 days, after 0.030% fees per side and 2 bps slippage
- Checked against the matrix invariants and the per-cell order-path checks
- Posture settings measured for direction rather than assumed; inert settings named
- Built as a single self-contained file, verified to behave identically to its source
Owed
- A hand audit. Passing the invariants is not a line-by-line read. Presentation — log format, side panel, chart marks, colour contract — is not checked here at all.
- A longer window. 30 days on one pair is one sample of one market.
- Out-of-sample confirmation. Nothing here was measured on a live exchange.
What this measurement cannot tell you
- Spot, long-only, one price per bar, no order book. A strategy whose edge is the spread cannot be measured here at all — those are set aside rather than reported badly (see the audit records).
- Trade statistics count closed round trips only, so a position still open at the end is invisible to win rate and profit factor. The realised/unrealised split on the performance page is what exposes that.
- One window, one pair. A strategy can be fitted to it.
Every number on this page is read from the run record that produced it — the audit pipeline writes the page and the record together, so the two cannot drift apart.
What the guard is, and what a better number under it means. The break-even guard refuses any exit below average entry plus fees. It cannot improve a strategy's judgement — it can only stop it selling at a loss, which means holding the position instead. Across the pack that turns a strategy that churned into one that buys and waits: trade counts collapse and time in market rises towards 100%. In a window where the asset itself rose, that reads as a much better return while being strictly less of a strategy.
So read the return next to round trips, time in market and edge vs buy and hold. The last of those is the one that matters: not one strategy in this matrix beat simply holding the asset, in either run.
Ignore the win rate and the profit factor under the guard. Both count closed trades only, and the guard closes nothing at a loss — in all 101 guarded runs the number of losing closed trades is exactly zero. So the win rate goes to 100% and the profit factor becomes undefined for arithmetic reasons, while the losses sit in the position that is still open. The unrealised row is where they went.