Module stats 

Aggregate statistics over every recorded run.

These tables are a by-product of running the graph, not a benchmark. The seat assignment rotates, the task distribution is whatever the operator happened to ask for, and a model that draws harder tasks looks worse. Read them as “relative performance on my workload”, which is the only claim the data supports.

Structs

AdvisorStats

Design-deliberation record for one agent.

AgentStats

Implementation record for one agent.

E2eStats

What real-machine verification caught that static review did not.

NodeDuration

Per-node duration breakdown, aggregated across every loaded run.

ReleaseBumpStats

What the post-merge release-bump step did, across every merged run.

RepoStats

One repository’s Stats, as grouped by by_repo.

ReviewerStats

Review record for one agent.

Stats

Everything, aggregated.

Totals

Run-level counters.

Functions

by_repo

Group states by RunState::repo and aggregate each group with collect_refs — the same counting logic as collect, just scoped to one repository at a time.

collect

Aggregate states.

collect_refs

Aggregate states, over any iterator of references rather than a slice — what by_repo uses to run the same counting logic over each repository’s own group without collecting it into an owned Vec<RunState> first. collect is a thin wrapper around this for the common whole-slice case.

filter_repo

Narrow states to those recorded against repo exactly.

load_all

Load every run on disk, skipping any that cannot be read.

node_durations

Derive a per-node duration breakdown from every loaded run’s events.