Stakeholder synthesize

Stakeholder Synthesize

Find the signal in accumulated reflections. Operates over one stakeholder, a category, or all stakeholders, depending on the user's framing.

Inputs to clarify

Before synthesizing, settle:

1. Scope. One stakeholder, one category (e.g. all of managing-down), or everything?
2. Time window. Default: last 90 days. The user can override ("this quarter", "since the reorg in March", "all time").
3. Lens. What's the user looking for?
   • Trend over time ("has Jill's morale improved?")
   • Cross-stakeholder pattern ("am I getting consistent feedback across managing-up?")
   • Pre-meeting prep ("what do I know about John before our 1:1?")
   • Performance review prep ("what's the case for Draymond's promotion?")
   • Blind spots ("what haven't I been thinking about?")

If the user gives an open-ended ask ("what's going on with my team?"), pick a default lens and state it: "I'll do a cross-stakeholder pattern read across managing-down for the last 90 days — let me know if you want a different cut."

Loading

1. Read ~/bettersense-work-reflections/stakeholders.json for the registry.
2. Identify which files to load based on scope.
3. Read each file in full. Within each file, read all ### YYYY-MM-DD entries falling inside the time window.

Synthesis discipline

This is the load-bearing part. Bad synthesis on stakeholder data is worse than no synthesis — it manufactures false confidence about real people.

Every claim must cite at least one specific dated entry. Format: "Morale has been declining since mid-March (entries 2026-03-18, 2026-04-02, 2026-04-15 all reference disengagement signals)." If you cannot cite, do not claim.

Distinguish your three modes of evidence:

• ask entries: what the stakeholder said. Treat as their stated position, which may or may not be the truth.
• sense entries: what the user observed. Treat as the user's read, which is filtered by the user's biases but anchored in behavior.
• askandsense entries: where stated and observed agree, conviction is highest; where they disagree, surface the gap explicitly.

Surface contradictions, don't paper over them. If the user wrote on March 5 that John "seems all-in" and on April 12 that John "is checked out," call out the gap rather than averaging it.

Distinguish trend from snapshot. Three entries in the same week saying the same thing is a snapshot. Three entries across two months saying the same thing is a trend. Don't conflate them.

Watch for the user's blind spots. Patterns of non-reflection are themselves signal:

• Stakeholders with no entries in 60+ days despite due items: avoidance? Genuine no-news?
• Questions consistently skipped across stakeholders: unwillingness to look at certain topics (compensation, conflict, morale)?
• Entries that always say "fine, no issues": real, or surface-skating?

Surface these as "patterns of attention," not as accusations.

Output structure

Match the lens. For most lenses, this shape:

# Synthesis: [scope] — [time window]

## Top-line read
One paragraph. The most important thing the user should walk away with.

## Patterns
- **[Pattern]** — one-sentence claim. (Sources: 2026-03-18, 2026-04-02 in [stakeholder])
- ...

## Contradictions / open questions
- **[Tension]** — what was said vs. what was observed. (Sources: ...)
- ...

## Blind spots
- Stakeholders or topics the user hasn't engaged with recently, with what would be worth doing.

## Suggested next conversations
- [Stakeholder]: what's worth raising, with concrete framing.

For pre-meeting prep, swap "Suggested next conversations" for a meeting agenda. For performance review prep, swap for a structured promo-packet-style write-up (strengths with evidence, growth areas with evidence, recent impact).

Composing with other skills

After producing the synthesis, surface the natural next steps:

• If a "next conversation" would benefit from structured feedback drafting → offer to invoke feedback-frameworks.
• If a meeting was named → offer to invoke one-on-one-prep.
• If the synthesis surfaced a decision the user is implicitly making (e.g. "I'm going to stop investing in this peer relationship") → offer to invoke decision-log so it's named and revisitable.
• If the synthesis is part of pre-launch / pre-review prep → mention that the-translator can help reframe technical findings for execs (when the audience matters).

Operating principles

• Synthesis ≠ summary. A summary describes what the entries say. Synthesis derives a claim the user wouldn't have seen by reading them sequentially. If the output reads like a recap, you haven't finished.
• Cite or strike. Any claim without an entry-date citation gets removed. No exceptions.
• Honor the user's voice. Quote the user's own phrasing where it's vivid; don't sand off the candor. The reflections are private — clean copy is not the goal, accurate insight is.
• Quantify scope honestly. "Across 4 entries over 6 weeks" is honest; "consistently" is loaded.
• Don't moralize. If the user has been avoiding a stakeholder, name the pattern, not the person. The skill is a mirror, not a coach.
• Resist the temptation to therapize. Stakeholder reflection is for getting more useful at work, not for psychological insight about the stakeholder.

Anti-patterns to flag

• Bulleted summary that reads "John said X. John said Y. John said Z." That's a list, not synthesis. Do better.
• Smoothing over a contradiction. If three entries disagree, surface the disagreement.
• Inventing patterns from sparse data. Two entries is not a trend. Say "one signal" if that's what it is.
• Synthesizing across stakeholders without naming the segment. "My managing-up relationships are healthy" is a claim about a population; specify which stakeholders, over what window, with what evidence.
• Producing a synthesis the user can't act on. End with concrete next moves, even if the next move is "I should talk to her about X."