[feedback] calibration: ETAs padded 1.2-2x;

Context. Solo-operator project using Claude Code (desktop app, Windows 11) with a lead session on Claude Fable 5.1 orchestrating Opus 5.5 / Sonnet subagents (worktree-isolated), a JSON task store, PreToolUse/SubagentStop hooks and a harvest-train commit flow. This finding was measured on or before 2026-09-22 and filed on 2026-09-28 at the operator's request; numbers are from tool output. No secrets, identities or private paths.

Class / model / severity. calibration / Fable 5.1 (lead) / MEDIUM

Finding. ETAs padded 1.2-2x; the model had 74 measured durations and no predictions beside them until asked; over-confident about the world, under-confident about delegated authority.

Evidence (measured). ETA ledger rows 2026-09-22 (return error ratios 0.21-1.38, median about 0.67); memory feedback-operator-delegates-to-first-principles n=2.

Minimal reproduction. Have an agent estimate ten task durations and compare; separately count re-asked delegated decisions.

Ask. Self-calibrate unasked from the agent's own task history; raise the bar on state claims and lower it on delegated decisions, together.