trust · whole group · offline + online
The Safety Check.folder
Psychological safety isn't a mood — it's four concrete questions, and one always lags. Rate them privately, reveal the bars, and fix the lowest one.
// statement + behavior cards · 20 min · print & run · tool pack · no projector, no account
A trust tool that makes psychological safety visible, built as a complete workshop folder: a timed run sheet, printable statement and prompt cards, and a mark → reveal → find-the-lowest → raise-it mechanic that turns 'do we feel safe here?' into four bars with one clear target.
The props are poker-spec print cards (63.5 × 88.9 mm, bleed and cut guides included) — four statement cards (ask a dumb question · admit a mistake · disagree · take a risk), a lowest-one card, a behavior card and branded backs. Nothing needs a projector, and nothing needs an account to download.
the run sheet
// this is the guide file, word for word — read it here, run from it here. the .md at the bottom of the page is the same text, offline.
Folder: `the-safety-check` · Category: trust · 4–12 people · 20 min · offline + online · print props · tool pack License: CC BY 4.0
A 20-minute, 4–12 person trust tool. Psychological safety isn't a mood — it shows up in four concrete questions, and teams almost always have one that scores far below the rest. Everyone privately rates those four, the team reveals together, and the lowest-scoring one becomes the target: name a single behavior that would raise it, and commit to try it before you check again.
How this relates to the other trust folders: The Safety Check is the trust folder that *measures* — four questions, dot-rated, one visible lowest bar. Step In If… shows shared experience without numbers; The Reliability Ledger writes promises down; The Trust Timeline maps past ruptures. Run this when the question is "where exactly are we holding back?" — then follow with whichever repair tool the answer points to.
Objectives
By the end of the activity, participants will have:
- Privately rated the team's real 1–5 score on all four safety statements — ask a dumb question, admit a mistake, disagree with the decision-maker, take a risk without it being used against me.
- Revealed the four bars side by side and agreed out loud which single one is lowest — without averaging it away.
- Committed to one concrete behavior anyone could start this week to raise that bar, with a re-check date attached.
When to use this activity
- A team where things feel fine in meetings and quietly don't between them — unreported problems, unasked questions.
- After something went wrong and nobody said so at the time; a lagging "I can't admit a mistake" bar is usually why.
- On a rhythm (quarterly pulse) to watch the four bars move — or to prove they won't move by themselves.
Setup
Clock: twenty minutes end to end. The rating is fast; the argument about the lowest bar is the activity — don't let it get squeezed out by the reveal.
Space: a wall or shared board with four columns, one per statement, and a spot for The lowest one. Online, the columns live on a shared board and dots become anonymous votes.
Materials per person:
| Item | Source |
|---|---|
| 4 × private 1–5 dots (sticky notes or markers, no names) | your own |
| The four statement cards — Ask a dumb question · Admit a mistake · Disagree with the decision-maker · Take a risk without it being used against me | `the-safety-check-prop-cards.pdf` |
| 1 × "The lowest one" card + 1 × "One behavior to raise it" card + blank | `the-safety-check-prop-cards.pdf` |
| Optional: the poster with the two rules in view | `the-safety-check-activity-poster.pdf` (A3) on the wall |
Grouping: 4–12 people who work together and share decisions — a team, a pod, a leadership group. Below 4 the anonymity dies; above 12, keep the reveal to the shape, not the count.
Roles: everyone rates privately and un-names their dots; you run the reveal without commentary, refuse the average, and force the close — one behavior, who does it, which day, and the re-check date. No projector, no account, no online form required.
How to run it
| Time | Phase | Facilitator does | Participants do |
|---|---|---|---|
| 0–4 | Mark | Read the four statements aloud, exactly as worded. "1 = definitely not safe here, 5 = completely safe. No names on the dots." | Privately place a 1–5 dot on each of the four statements |
| 4–8 | Reveal | Count the dots per statement together, no commentary. Let the shape of the four bars land before anyone speaks. | Watch one bar clearly lag while the others hold up |
| 8–12 | Find the lowest | Circle The lowest one. Refuse "but our average is fine" — the lowest bar is where people actually hold back | Agree out loud which single statement is the problem, and what it cost recently |
| 12–20 | Raise it | Ask for one concrete behavior anyone could start this week that would raise that bar. Push from "everyone should" to "I will, from Tuesday". Commit, and set the re-check date | Name the behavior, who starts it, and what the first visible proof would look like |
Then run the debrief — see `the-safety-check-debrief.md`.
The card
The four statement cards are the instrument — fixed wording, because the whole readout depends on comparing the same four bars every time you run it. The labels are deliberately in first person ("I can admit a mistake"), so each dot is about the room, not about someone's leadership style. The lowest one and One behavior to raise it keep the check from ending in a score: the tool's product is a dated behavior, not a chart.
Failure modes
- Everyone rates 5s. The check itself feels unsafe — a classic sign. Rescue: switch to anonymous written dots, or ask "what would have to be true for a 3 to be safe to say?"
- The team wants to fix all four. Rescue: you'll fix none. Force the single lowest bar; one raised bar lifts the others faster than four half-efforts.
- A leader explains why the low score is wrong. Rescue: that explanation is the data. Thank them, park it, and keep the anonymous result — the check measures the room, not the org chart.
Variations
- New team, week one: run it to set expectations out loud — "here's what we want to be safe for, and we're not there yet."
- After something went wrong: a spike in "I can't admit a mistake" is usually why things stopped being reported. Check that bar first.
- Quarterly pulse: re-run on a rhythm; the four bars over time are the truest health readout a team has.
- Two teams that hand off: run it once per team; the bars that differ are where the inter-team trust leaks.
Facilitation notes
- Dots carry no names. The moment someone's score is traced to them, the ratings turn safe and useless — protect the anonymity harder than you protect any other rule here.
- The lowest score is the target, not the average. A team averaging 4 with one statement at 2 has a real problem in that one place.
- "One behavior" means observable: if a stranger couldn't spot it happening, it isn't one yet.
Put it on the wall
Print `the-safety-check-activity-poster.pdf` (A3) to keep the four statements and the two rules — name the lowest, not the average; anonymous dots — in view during the reveal.
Props in this folder
- `the-safety-check-prop-cards.pdf` — poker-spec print sheet (63.5 × 88.9 mm, bleed and cut guides): four statement cards, the lowest one, one behavior to raise it, blank card, branded backs
- `the-safety-check-guide.md` — this run sheet
- `the-safety-check-debrief.md` — the debrief questions and the failure modes of the debrief
- `the-safety-check-run-sheet.pdf` — the print edition: this run sheet + the debrief, bound as one A4 document with bookmarks
- `the-safety-check-activity-poster.pdf` — the whole tool on one A3 wall sheet: the four bars, the flow, the watch-outs
- `the-safety-check-start-here.md` — one page, thirty seconds, then print and go
- `the-safety-check-license.md` — what you may do with all of this
- `SKILL.md` — machine-readable manifest for AI agents
- `*-Letter.pdf` — US Letter editions of the print documents
Agent version
Every file in this folder has a stable public URL — an AI agent can fetch the cards, the debrief and this run sheet with three curl commands (see `SKILL.md`). The paper version needs nothing but four statements on a wall and anonymous dots.
the debrief sheet
// the debrief questions, in full — the .md version prints on one page for the room.
Four bars are a diagnosis, not a verdict — the debrief is where the lowest one turns into one dated behavior. Use with `the-safety-check-guide.md`. The raise-it phase at minute 12 is the compressed version; this sheet is the full one.
How to run this debrief
- Keep the four bars visible on the wall. This debrief argues with a picture the team just drew; folded away, it becomes opinion again.
- Stay on the lowest bar. Curiosity about the other three is how teams dilute the one finding they have.
- Close on behavior, owner, start day, re-check date. A debrief that ends at "we should be more open" measured nothing.
Read the shape
- Look at the four bars side by side. Which one is the outlier — clearly lower than the rest?
- Did the shape surprise anyone? A leader who thought "we're totally open" and sees "disagree with the decision-maker" at a 2 has learned something.
- Are any two bars linked? Low "admit a mistake" and low "take a risk" usually travel together — one silence causes both.
The six debrief questions
1. The outlier.
"Which bar is clearly lower than the rest — and who was unsurprised by it? Unsurprised people know where it comes from."
→ Targets: reading the shape before feelings; the unsurprised are your first witnesses, not your enemies.
2. The last test.
"For the lowest statement: when was the last time someone tested it here and it went badly?"
→ Targets: the incident behind the score — usually one, recent enough, and known by more people than will say so.
3. Rule or signal.
"Is the low score about a rule (punishment is real) or a signal (nobody has seen it go well)? What would a person have to believe to keep their question, mistake or disagreement to themselves?"
→ Targets: the diagnosis the fix depends on — a rule needs changing, a signal needs one visible good outcome. Name the belief, not the person.
4. The signal-maker.
"Who is most able to change that signal — and are they in the room?"
→ Targets: whether the behavior can come from today's seats; if not, the ask becomes a conversation with someone who isn't here.
5. The first proof.
"What's the first visible proof that it's safe now — the small thing that, if it happened, would move the next dot up? And what happens to the person who tests the new behavior and it goes badly?"
→ Targets: making safety observable and the retry protected; otherwise you accidentally confirm the old fear.
6. The owner and the date.
"The behavior you picked: who will do it, starting which day — and when is the re-check? What will you do if the bar hasn't moved?"
→ Targets: "'everyone should' means no one will"; the re-check plan is what makes this a measurement, not a mood.
Failure modes of the debrief
| Failure mode | Why it happens | The rescue |
|---|---|---|
| Everyone rated 5s | The check itself felt unsafe — a classic sign | Redo the marking anonymously in writing, or ask "what would have to be true for a 3 to be safe to say?" |
| The team insists on fixing all four | The average feels safer than the outlier | Force the single lowest bar; one raised bar lifts the others faster than four half-efforts. |
| A leader argues the score away | The check measured the room, and the org chart objects | That explanation is the data. Thank them, park it, keep the anonymous result. |
Who is The Safety Check for?
A team that 'gets along' but quietly holds back questions, mistakes, or disagreement — and any leader who wants a real read on where people don't feel safe. It needs 4–12 people and 20 minutes; it runs on a shared board online.
What does it produce?
Four scored statements with their real shape revealed, the one missing kind of safety named out loud, and a single concrete behavior the team committed to start — with a date to re-check whether it moved. No pre/post measure; the artifact is the four bars and the behavior.
Can my AI agent use this folder?
Yes. Every folder ships a SKILL.md skill file: frontmatter for matching plus runnable prep commands — fetch the cards, pull the debrief. The three-step workflow (discover, install, run) lives on the skills page, every folder ships a one-command zip bundle, and the machine index is skills.json.
Is it really free?
Yes — CC BY 4.0, no account, no email wall inside a folder. Download it, run it, copy it, rebrand it for your own workshop. Attribution keeps everything free.
make it yours
This check needs no online form and no account — four statements on a wall and nameless dots. The card prompts are the one file to change: rename the four statements and the lowest/behavior cards to how your team already talks about speaking up, and reprint at poker spec; the 'lowest score is the target' rule stays.
Files in this folder
All files are free under CC BY 4.0. Attribution: “The Safety Check · WorkshopSkill · CC BY 4.0.”