Playbook: OKRs
Objectives & Key Results — where the principal is trying to get to and how they'll know they got there. Your job is two-fold: track them, and be a candid critic when they're shared. Runs when the principal shares/updates OKRs, at onboarding, and inside weekly reviews.
The tracker — memory/okrs.md
A single file (not one per objective). Frontmatter type: okr. One section per objective with a KR table:
---
name: memory-okrs
description: The principal's objectives and key results, tracked to the timeframe.
type: okr
updated: <YYYY-MM-DD>
---
# OKRs — <timeframe, e.g. Q3 2026>
## O1: <qualitative, outcome-oriented objective> — owner: <me/person> · status: <on-track/at-risk/off>
| key result | baseline | target | current | confidence | owner |
|---|---|---|---|---|---|
| <measurable outcome> | <start> | <target> | <now> | <0–100%> | <me/person> |
- Objective: a sentence — where they're headed, why it matters. Not a metric, not a task.
- Key result: a measurable outcome with baseline → target over the timeframe; track
current+confidence. - Link KRs to the projects/commitments that drive them; a KR often spawns commitments (
memory/commitments.md) and decisions (memory/decisions.md). - Confidential objectives/KRs use the code name only (see
brain/schemas/confidentiality.md).
Critique when shared — pressure-test, don't just record
When the principal shares OKRs (or drafts), record them, then critique constructively and offer concrete rewrites. Match the principal's directness/warmth (brain/role/voice.md) — candid, not harsh. Check:
- Objectives — qualitative and outcome/aspiration, not a task or a metric; aligned to the mandate (
memory/profile/); few (≈3–5). Flag: an objective that's really an activity, or one disconnected from what they're accountable for. - Key results — each is a measurable outcome (lagging) with a baseline + target + timeframe; owned; 2–5 per objective; not a restatement of the objective; not a vanity metric. Flag the classics:
- Activity masquerading as a KR — "*hold weekly syncs*", "*launch the campaign*", "*ship X*" (that's a task/output → ask "what outcome will that produce?" and propose the outcome KR).
- No measure / no baseline — "improve reliability" → propose a number and a starting point.
- Unowned, too many, duplicative, or misaligned with the objective.
- Sandbagged (trivially hittable) or impossible — push for ambitious-but-achievable.
- End with the 1–2 highest-leverage fixes, phrased as a suggested rewrite, and ask before overwriting their wording.
Tracking & surfacing
- Update
current/confidenceas you learn progress; move objectives toon-track/at-risk/off. - In weekly-retro / weekly-preview / daily-brief, surface at-risk KRs (low confidence, flat
current) and what would move them — tie to commitments and decisions. - At the timeframe's end, help score them and carry forward learnings.