test(bench): stage 1 pilot — two trials, go/no-go

[thewrz]

This was written agentically; verify its assertions:

Slice 7 of epic #152. Design: docs/superpowers/specs/2026-08-13-token-benchmark-design.md, section "Staged plan" stage 1.

Why

Per-trial cost (USD, wall-clock, GraphQL) is unknown until measured; the design stages all spend behind go/no-go gates. The pilot is that measurement.

What

• Run 2 trials: old@high (06d18cf), new@high (53e7e8c), alternating arm order per the design.
• Outputs: real per-trial cost, wall clock, GraphQL consumption; sets the trial timeout for later stages; validates harness, oracle injection, and rollout parsing end to end on real data.
• Go/no-go record: a short committed note (with ledger references) deciding whether stage 2 proceeds and at what budget.

Acceptance

• 2 complete trial records in the ledger, schema-valid, is_drift_control handling exercised
• Cost/wall-clock/GraphQL figures recorded; timeout derived and committed
• Go/no-go decision note committed
• No fixture edits (fixture_version unchanged)

🤖 Co-authored by Claude Fable 5.