Time Reality Check

The user gave themselves an estimate. The user is almost certainly wrong. Time blindness is the default for ADHD and many autistic brains — initial estimates skew low because the brain models the best-case path and silently drops setup, interruptions, decision points, and the cost of re-entry from context switches.

Your job is to produce a calibrated estimate that accounts for the gap between the user's mental model and reality.

Lens: statistics-expert + devils-advocate

For non-trivial estimates (anything the user is using to commit to others, anything over an hour, anything that will affect their schedule), invoke clarity-council via Skill in multi mode with personas [statistics-expert, devils-advocate]:

• user_problem: "The user estimated [X] for [task]. Produce a calibrated range that accounts for known time-blindness factors: setup cost, interruption probability, decision points, re-entry cost from context switches."
• context: task description, user's gut estimate, deadline if any, similar tasks they've done before (if recalled).
• desired_outcome: "A three-point estimate (best / likely / worst), an explicit list of factors the user's gut estimate probably missed, and a recommended buffer."
• constraints: [do not flatter the gut estimate, do not catastrophize, base multipliers on stated patterns rather than generic ADHD lore, separate task time from elapsed time]
• depth: brief.

For trivial estimates (a single small task with no setup), skip the council and produce a one-line calibrated read.

When to activate

When the user says:

• "I'll do it in [X minutes]"
• "How long will this take?"
• "Can I get this done before [time]?"
• "Am I being realistic?"

The questions

Ask these in one AskUserQuestion call:

1. What's the task? (Concrete enough to estimate against.)
2. What's your gut estimate? (Capture this before anything else — the gut number is the data point.)
3. What's the last similar task you did and how long did it actually take? (Anchor against reality, not memory.)

Output format

Your gut: [X]

Restate the user's gut estimate, plainly. No judgment.

Calibrated range

	Time	What it assumes
Best case	[X]	Zero setup, zero interruptions, no decision paralysis
Likely	[X]	Normal setup, 1-2 small interruptions, expected decision points
Worst case	[X]	Multiple context switches, hidden subtasks surface, re-entry cost

What your gut probably missed

A bulleted list. Be specific to the task, not generic. Examples:

• Setup: opening the right files, finding where you left off
• Decision points: every place the task forks and requires a choice
• Re-entry cost from any interruption longer than 5 minutes
• The non-coding parts (PR description, screenshot, message to teammate)

Recommended commitment

If the user is committing to another person, use the worst case minus a small buffer. If only committing to themselves, use the likely.

Commit to: [X by Y] Personal target: [X]

One-line summary

If the user only reads one line, this is it: a flat declarative sentence with the recommended time.

What NOT to do

• Do not multiply the gut estimate by a fixed factor (2x, 3x) — that's lazy and wrong as often as it's right
• Do not catastrophize — "it'll take all day" is not useful
• Do not assume the user can't focus; some tasks genuinely take 20 minutes
• Do not skip the gut estimate question — you need their number to calibrate against
• Do not lecture about time blindness; the user knows

Host portability: tool names in this skill follow Claude Code conventions; on other hosts (Codex, opencode) map them by intent — see PORTABILITY.md.

<!-- self-evolve:start -->

Self-Evolve Loop

Journal ~/.ink-and-agency/learnings/time-reality-check.md (or workspace-local .ink-and-agency/ where the sandbox confines writes). Read it first; append what the run taught — SELF-EVOLVE.md.

<!-- self-evolve:end -->