Timing
Week 5 — Prepare, then POST
Dublin was already up. What I still have from this night is a named clock: build the order, POST, stop when the API answers.
| StratB | 52 rows, 00:44:50–04:54:53. Median 35.35 ms. p95 298.0 ms. Max 521.4 ms. 51 resting, 1 matched in the log. |
|---|---|
| UC | 22 rows, 00:44:50–01:59:55. Median 33.0 ms. p95 289.2 ms. Max 293.0 ms. All 22 resting. |
| Inside the total | Build-order medians 1.7 ms / 1.9 ms. POST medians 32.85 ms / 29.45 ms. Component medians do not add to the median total. |
| Not in the total | Feed delay, earlier decision time, matching, settlement. The “matched” row was not verified as a fill through the Data API. |
An older bot report’s ~27-second oracle update is a different metric. It is not a before/after for these milliseconds. Hashes of the private logs sit in EVIDENCE.md. The raw files stay private because they contain order identifiers.
35.35 ms is how long this stack took to hear back. It is not how long it took to get filled, and it is not why the wallet later lost money.
← Blog