retro

name	retro
description	Facilitates a structured team retrospective using the well/wrong/confusing framework — gathering observations, tracing them to root causes instead of symptoms, and committing to 2-3 specific improvements with named owners and due dates. Use when you say "let's run a retro," "what went well and what didn't," "sprint retrospective," "post-mortem the launch," or "we keep making the same mistakes."
tier	guided
time	30-45 min
inputs	what happened in the period, and how the team felt
outputs	meetings/retro-<date>.md

Retro

A retrospective that actually changes behavior: collect what went well / wrong / confusing, dig past symptoms to root cause, and walk out with 2-3 owned actions. The point isn't to vent — it's to fix the system that produced the outcome.

Grounded in: Agile Retrospectives — Esther Derby & Diana Larsen: set the stage → gather data → generate insight → decide what to do, blamelessly. Go deeper (The Product Channel): The Product Death Cycle

When to use this

• End of a sprint, cycle, or milestone and you want a repeatable retro instead of an unstructured gripe session.
• A launch, incident, or project just wrapped (good or bad) and you need a blameless post-mortem.
• The same problem keeps recurring and prior retros produced action items nobody did.
• A new or distributed team needs a lightweight, async-friendly retro format.
• You've collected raw notes/feedback from the team and need them synthesized into themes, causes, and a short action list.

Before you start (gather these)

• Scope & period: what are we retro-ing (which sprint/project/incident) and over what dates?
• Raw observations: the team's well / wrong / confusing notes, OR the source material to derive them (Slack threads, sprint board, incident timeline, survey responses).
• Outcomes/metrics: what actually happened vs. what was planned — shipped vs. committed, dates, defects, key numbers.
• Prior retro actions: open action items from last time, so you can check follow-through.

If raw observations ARE attached (notes, a thread, survey responses, an incident timeline): synthesize directly. Don't interrogate — proceed to the Process and state any assumptions inline (e.g., "Assuming the 3 spillover stories are the 'wrong' signal since no per-person notes were attached"). Ask at most one scope-confirming question only if the period is genuinely ambiguous.

If NO raw observations exist yet: don't interrogate the user for them — offer to run a collection round. Say something like: "I don't have the team's observations yet. I can either (a) run a quick collection round — I'll give you three prompts to send the team and synthesize what comes back, or (b) work from source material if you point me at the Slack thread / sprint board / incident timeline." If they choose the collection round, supply these prompts:

• Well: "What went well this period that we should deliberately keep doing?"
• Wrong: "What hurt us — a known problem that cost us time, quality, or trust?"
• Confusing: "Where was ownership unclear, a process ambiguous, or something a surprise — anything that will bite us if left alone?"

Only fall back to clarifying questions for genuinely missing scope (which sprint/incident, committed vs. delivered, the triggering outcome, prior retro actions) — and ask those in one batch, not as an interrogation.

Process

1. Set scope and ground rules. State the period and the one prime directive: blameless. We critique systems and decisions, not people. Anonymize observations if the team isn't safe being attributed.
2. Collect into three buckets. Sort every observation into Well, Wrong, or Confusing. Disambiguate the last two — they're easy to conflate:
   • Well — something that worked and is worth deliberately protecting (keep doing).
   • Wrong — a known problem that already hurt us: it cost time, quality, money, or trust this period. You can name the damage. (e.g., "the migration broke staging for two days.")
   • Confusing — ambiguity that hasn't fully detonated yet: unclear ownership, an undefined process, a hand-off with no definition-of-ready, a surprise nobody owned. It didn't necessarily hurt us this time, but it will bite. Confusing is the most undervalued bucket — it's where future failures incubate. When unsure: did it already cost us (Wrong) or is it a latent ambiguity that could (Confusing)?
3. Cluster into themes. Group related items. Five notes about "PR reviews were slow" are one theme, not five. Note frequency — recurring mentions signal systemic, not one-off, issues.
4. Find root cause, not symptom. For each Wrong/Confusing theme, run 3-5 Whys until you hit a cause the team can actually change (process, tooling, communication, capacity, unclear ownership). "We missed the deadline" is a symptom; "we never re-estimated after scope grew mid-sprint" is closer to root.
5. Triage ruthlessly — score, don't argue. You cannot fix everything; pick the top 2-3 root causes reproducibly rather than by whoever lobbies hardest. Use one of two lightweight methods:
   • Impact × Recurrence score (default for async / when you have notes). For each theme, rate Impact 1-3 (1 = mild annoyance, 2 = real cost, 3 = serious damage or repeat offender) and Recurrence 1-3 (1 = one-off, 2 = a few times, 3 = chronic/every period). Multiply. Sort descending; the top 2-3 are your candidates. Show the table so the ranking is auditable.
   • Dot-voting (default for a live group). Give each participant 3 dots; they place them on the themes they most want fixed (multiple dots on one theme allowed). Tally; the top 2-3 by dot count win. Fast, visible, and surfaces shared pain over individual hobbyhorses. Break ties toward the theme whose fix is most actionable this period. Record the scores/votes in the output so next retro can see why these three were chosen.

Facilitating live (when running the retro in real time)

Run it on a clock so it doesn't sprawl. A tight 60-minute retro:

Bucket / step	Timebox	What you do
Set scope + ground rules	5 min	State period and the blameless prime directive.
Silent write — all three buckets	10 min	Everyone writes Well/Wrong/Confusing independently (sticky notes / shared doc), no talking yet.
Well	5 min	Read out, cluster, celebrate briefly — don't over-dwell.
Wrong	10 min	Read out, cluster, name the cost of each.
Confusing	10 min	Read out, cluster; pull on every "who owned this?" thread.
Root cause (top themes)	8 min	3-5 Whys on the heaviest themes.
Triage (dot-vote)	4 min	3 dots each, tally, pick top 2-3.
Owned actions + last retro check	8 min	One owner + date per chosen cause; mark prior actions Done/Partial/Dropped.

Elicitation prompts to keep it moving:

• Opening each bucket: "60 seconds — write before we talk." (silent-first prevents anchoring.)
• Wrong: "What did this actually cost us — hours, a defect, a missed commitment?"
• Confusing: "Who owned this? If three of you just hesitated, that's the theme."
• Drilling a Why: "And what made that happen?" — repeat until you hit something the team controls.
• Stuck/quiet: "What would you warn a new teammate about before next sprint?"
• Before closing a theme: "If we changed exactly one thing here, what would it be?"

Psychological safety — reframe blame in the moment. When an observation names a person or team ("Engineering dropped the ball," "Maria forgot to update the ticket"), reframe it to the system gap right then, out loud: "Let's restate that as a system gap — sounds like the hand-off had no definition-of-ready," or "so the ticket-update step isn't built into our workflow." Record the systemic version, not the name. If the team clearly isn't safe being attributed — junior members silent, a manager in the room, a high-stakes post-mortem — default to anonymous collection: gather observations in writing without names (anonymous form / shuffled stickies) and facilitate from the anonymized set.

Output template

# Retrospective — [Sprint / Project / Incident] ([Start date]–[End date])

**Facilitator:** [Name] · **Participants:** [N people / teams] · **Prime directive:** Blameless — we fix systems, not blame people.

## Outcome snapshot
| Planned | Actual | Delta |
|---|---|---|
| [e.g., 24 pts committed] | [e.g., 18 pts delivered] | [-6 pts / 3 stories spilled] |
| [Target date / metric] | [Actual] | [Gap] |

## 1. What went well (keep doing)
- **[Theme]** ([×N mentions]) — [what specifically worked and why it's worth protecting]
- **[Theme]** — [observation]

## 2. What went wrong (hurt us)
- **[Theme]** ([×N mentions]) — [the observable problem and its impact]
- **[Theme]** — [observation]

## 3. What was confusing (unclear / surprised us)
- **[Theme]** — [where ownership, process, or expectations were ambiguous]
- **[Theme]** — [observation]

## 4. Root cause analysis (top themes only)
| Theme | Symptom (what we saw) | Root cause (3-5 Whys) | Bucket | Triage (I×R or dots) |
|---|---|---|---|---|
| [Theme] | [Surface observation] | [The changeable underlying cause] | Process / Tooling / Comms / Capacity / Ownership | [e.g., 3×3=9 / 5 dots] |
| [Theme] | [Symptom] | [Root cause] | [Category] | [score] |
| [Theme] | [Symptom] | [Root cause] | [Category] | [score] |

## 5. Actions for next period (2-3 max — owned & dated)
| # | Action (specific & verifiable) | Addresses root cause | Owner | Due | Tracked in |
|---|---|---|---|---|---|
| 1 | [Concrete, observable change] | [Theme] | [Name] | [Date] | [Board/ticket link] |
| 2 | [Action] | [Theme] | [Name] | [Date] | [Link] |
| 3 | [Action] | [Theme] | [Name] | [Date] | [Link] |

## 6. Last retro's actions — follow-through
| Prior action | Owner | Status | Note |
|---|---|---|---|
| [Action] | [Name] | Done / Partial / Dropped | [Why, if not done] |

## Decisions & parking lot
- **Decided:** [Any decision made in the room]
-  [Item] → [Owner to follow up]

Avoid (anti-patterns)

• Symptom-fixing. Stopping at "we were slow" without a Why chain produces actions that treat the rash, not the infection. Push to a cause the team controls.
• The action-item graveyard. Generating 8 vague to-dos with no owner or date. Two owned, dated actions beat ten orphaned ones — every action needs a name and a date.
• Skipping "confusing." Teams happily debate wins and losses but glide past ambiguity. Unclear ownership is where the next failure is already brewing.
• Blame leakage. "Engineering dropped the ball" kills psychological safety and honest input. Reframe to the system: "handoffs lacked a definition-of-ready."
• Recency bias. Letting the last bad day dominate a two-week period. Weight by frequency and impact across the whole window, not by what stings most right now.

Tips

• Collect silently first, discuss second. Have people write their well/wrong/confusing independently before the group talks — it prevents the loudest voice from anchoring everyone.
• One action per root cause, owned by one person. Shared ownership is no ownership. If two people must collaborate, one is still accountable.
• Open with last retro's scorecard. Reviewing prior follow-through first creates accountability gravity for this retro's actions.
• Rotate the facilitator. It spreads skill, surfaces different themes, and stops the retro from becoming one person's pet agenda.