mirror of
https://github.com/lightningnetwork/lnd.git
synced 2026-08-13 12:32:48 +02:00
0a3abd6cfd
In this commit, we wire the stage E section through the evaluator and report the two numbers that say whether it did anything. The section is read after the clock because it replaces a piece of one, and it is refused outright on a file that configures no virtual time. A file that asks for latency gets the makespan too, whether or not it asks for concurrency: how long the batch took is the number this stage is about, and gating it behind the other section would have hidden it from every tier the stage is actually run on. Both new keys are pointers, like stage D's five, so a file with no latency section emits neither and its output stays byte identical to every run before this stage, while a file with one emits both including the zeroes. They are the manipulation check, and this is the fifth stage in a row to need one. A mechanism that never fires and a mechanism that fires and costs nothing score identically: mean_attempt_latency_sec read against the flat attempt_sec the clock would otherwise have charged is what separates them, and the per-attempt latency in the results array is where the differential half is auditable, since a settle and a first-hop failure over the same route read very different numbers. Neither enters the objective. Objective L is an offline re-scoring of archived runs, which is the lead's decision for the whole program. |
||
|---|---|---|
| .. | ||
| commands | ||
| lncli | ||
| lnd | ||
| routesim | ||