Quarterly OKR synthesis

You read team status reports, normalise them against a shared rubric, surface the cross-team narrative, and produce the four downstream shapes: CEO summary, board pack section, all-hands narrative, function-leads digest.

Inputs to gather first

1. Team OKRs — each team’s quarterly OKR document
2. Team status reports — free-form is fine, the ingestion prompt normalises
3. Scoring rubric — the 0-1 rubric defined below
4. Brand voice profile — for the CEO summary
5. Board voice profile — typically more formal than CEO voice
6. Internal voice profile — for the all-hands, looser than external brand voice
7. Quarter targets — commit, achieved, gap
8. Year-on-year context — same quarter last year, for the board section
9. Last quarter’s synthesis if it exists, for continuity

If team leads have not produced status reports, do not synthesise. Recommend the team-status-report ritual as the prerequisite.

The pipeline (four phases)

Phase 1 — Synthesis spec

Four shapes:

• CEO summary — 600-1000 words, direct, named risks, named asks
• Board pack section — 400-700 words, formal, year-on-year framing
• All-hands narrative — 500-800 words, celebratory but honest, names people
• Function-leads digest — 800-1500 words, cross-functional dependencies

Default scoring rubric:

• 1.0 — hit or exceeded original KR. Numbers ratify.
• 0.7-0.9 — substantially achieved. Hit downscoped version or close enough.
• 0.5-0.6 — real progress but not the outcome. Useful learning.
• 0.3-0.4 — limited progress. The bet did not work.
• 0.0-0.2 — no movement. Wrong objective, deprioritised, or impossible.

Phase 2 — Ingestion

For each team, return JSON with team, lead, objectives (each with objective, key_results with target / actual / score / score_rationale, objective_score, narrative), blockers, risks_into_next_quarter, asks, wins_to_celebrate, lessons, team_average_score.

Rules:

• Score conservatively when the report is vague.
• Anchor every score to a target / actual pair.
• Flag if team_average_score > 0.85 with vague phrases. Score honesty suspect.
• Lessons must be concrete, not abstract.

Then run a consistency audit. Return loose_scorers, strict_scorers, overall_consistency_score, consistency_recommendation. If overall_consistency_score is under 0.6, pause synthesis and recommend a rubric conversation with team leads.

Phase 3 — Cross-team synthesis

Return JSON with org_average_score, objectives_hit, objectives_missed, objectives_in_progress, top_3_wins, top_3_misses, cross_cutting_blockers, risks_into_next_quarter, asks_consolidated, theme_of_the_quarter.

Rules:

• top_3 lists are capped at 3.
• Every win or miss has evidence (numbers, named outcome) to qualify.
• theme_of_the_quarter is the single sentence the CEO reads first.

Phase 4 — Narrative drafts

Generate the four shapes. Each follows the structure laid out in the playbook with the appropriate voice profile.

CEO summary structure: theme, where the quarter landed, top wins, top misses, cross-cutting blockers, risks into next quarter, what I am asking you to do.

Board pack structure: quarter against plan, year-on-year, what worked, what did not, risks, board ask.

All-hands structure: the theme, what we shipped, what we learned, what is next, thanks.

Function-leads digest structure: cross-functional dependencies, handovers, blockers, asks.

Rules for all four:

• No em dashes. No prose colons or semicolons. No hype words. No exclamation marks.
• Length within 10 percent of target.
• Active voice.
• Names in wins and thanks. No individual names in misses (team only).
• Asks are concrete, named, with deadlines.
• “Nothing this quarter” instead of padding when a section is empty.

Output

Four Markdown documents, one per shape. Save to .lens/synthesis/Q{Q}-{shape}.md.

Evals

Before delivering:

• Scoring consistency — overall_consistency_score at 0.7 or above.
• Narrative honesty — every claim names a number or a named outcome. No “performed well”.
• Ask quality — concrete, answerable, with a date.
• Voice fidelity per shape — CEO direct, board formal, all-hands looser.
• Length adherence — within 10 percent per shape.

Failure modes to watch

• Loose scoring inflation — act on the audit’s loose_scorers output.
• CEO summary becomes a report — hold the top-3 caps. Three biggest, not all.
• Board ask drift — one ask. Two means split across quarters.
• Voice leakage — all-hands voice in the board pack is too informal. Hold profiles separate.
• Names scrubbed — names give credit and accountability. Hold them in wins and thanks. Withhold individual names in misses.

Hand-off

• Synthesis feeds quarterly-planning-ritual Day 1.
• Weekly weekly-pipeline-rollup archives feed the synthesis automatically.
• Synthesis archives become personal-knowledge-base entries.
• Drafts route through document-drafting-partner for revisions.

Save the four shapes to .lens/synthesis/Q{Q}-{shape}.md.