mushanghai-retrospective

name	mushanghai-retrospective
description	muShanghai post-experiment retrospective skill for Experiment Leaders after Demo Day: BioBloom retros start after May 22; Longevity Month retros start after June 5. Use for end-of-experiment summary, reproduction notes, what worked, what did not work, and next seed hypothesis.
when_to_use	Trigger phrases: 'retrospective', 'retro for my experiment', 'post-experiment reflection', 'what worked', 'what didn't work', 'reproduction notes', 'next experiment seed', 'end-of-experiment summary'. Use after Demo Day. Daily logging belongs to mushanghai-participant-log; mid-experiment status belongs to mushanghai-experiment-leader Mode E.
allowed-tools	Read Write

muShanghai Retrospective

Valid for: May 2026 cycle. BioBloom retros start after May 22; Longevity Month retros start after June 5. Refresh dates before later reuse.

Prerequisite: Before doing anything else, read ../mushanghai-canon/references/safety.md and ../mushanghai-canon/references/four-musts.md. These hold the safety banner, Four Musts labels, and HARK failure mode.

Always: show the canon safety banner first

Before any other content or question, surface this banner:

Before we start, do NOT paste:

• Identifiable health data (HRV with timestamps + location, blood labs, body weight, sleep records)
• Real names of other participants
• Other people's data without their explicit consent
• Details of medical conditions, pregnancy, or current medication

muShanghai skills help you plan, log, and reflect. They do not provide medical advice. All experiments are for education, exploration, and self-observation only - not diagnosis or treatment. High-risk interventions (anything beyond lifestyle, breath, movement, or off-the-shelf wearables) require Track Leader approval AND escalation to the Month organizer.

If the user pasted anything from the never-paste list, stop and ask them to redact before continuing.

Single-mode walkthrough

This skill has one mode: build a post-experiment retrospective at experiments/<track>/<seed-name>-retro.md using assets/retro-template.md.

Ask one clarifying question up front:

Are you running this solo as the Experiment Leader, or with your participants?

• Solo retro: address the user as the EL; focus on protocol design, observed deltas, and next seed.
• Group retro: invite participant-facing prompts too: what changed for you, what did not change, would you do this again, what did logs reveal that memory alone would have missed, and what should stay attached to the dataset.

Then collect the track, seed name, cohort, dates, EL, retro author, named signal(s), baseline state, final state, logs/artifacts, missed days, anomalies, and consent limits.

Five-step retro

Walk the user through these steps in order. Keep uncertainty visible; no polishing away no-change results or adherence problems.

1. What changed / what didn't - observed deltas across baseline -> final on the named signal(s). Include one row per Four Must: Problem, Monitoring, Intervention or Interaction, Deliverables.
2. Possible explanations - at least two competing explanations. Add a HARK warning if the hypothesis was rewritten mid-protocol: do not present a post-results hypothesis as if it was the original prediction.
3. Reproduction notes - what a future EL needs to repeat this experiment with no further questions: instruments, data fields, daily schedule, intervention/interaction dose or cadence, templates, reminders, consent constraints, and gotchas.
4. What I would do differently - separate adherence problems, instrument problems, protocol problems, and motivation problems.
5. What I want to do next - propose a tighter or different next experiment seed hypothesis. Hand off to mushanghai-experiment-leader Mode A for proposal bootstrap.

Output protocol

1. Fill assets/retro-template.md with the user's answers.
2. Propose the output path: experiments/<track>/<seed-name>-retro.md.
3. Show the user the proposed path and full markdown content for confirmation before writing.
4. After confirmation, write the file exactly at that path.
5. If the user does not confirm, revise in chat only; do not write.

References

• ../mushanghai-canon/references/safety.md
• ../mushanghai-canon/references/four-musts.md
• mushanghai-experiment-leader Mode A for the next seed handoff
• mushanghai-participant-log final reflection language for group retros

Assets

• assets/retro-template.md

Common pitfalls

• Do not turn a retrospective into a new proposal until Step 5.
• Do not hide null results, missed days, or device failures.
• Do not let a single explanation dominate when adherence, instrument, protocol, and motivation explanations still compete.
• Do not write identifiable health data or participant names into the retro.