benchmark

The Token Value Benchmark (contract 1.16): your own runs and the league table per task class — cost per successful task, quality at the bar, tokens in, out, cached and reasoning, value per unit cost, relative token value, cache effect, latency and escalation cost — with the plain sentence for each class. Read-only: starting a run is a person’s action in LightHouse, because it spends money on models.

Role: viewer and above (any person, AAL staff, machine key).

Acts: no. It only reads; nothing an AI agent calls through MCP approves, pays, or changes a limit.

Inputs

Input	Type	Default
run_id	string or null	null
since	string or null	null
until	string or null	null
task_class	string or null	null
limit	integer	20

Call

The arguments of a tools/call request:

{
  "name": "benchmark",
  "arguments": {}
}