Facilitator runbook

Everything needed to run the session, published as a page rather than buried in speaker notes. A co-facilitator, a translator, or a reviewer cannot read speaker notes while the session is running.

Profile 1, presenter-led, 90 minutes. Section allocation 15, 15, 25, 15, 15, and 5 minutes.

Run of show

Minutes Slide What happens Hard stop
0 to 3 S01 Title, duration, section split, independence notice read aloud  
3 to 7 S02 What Copilot actually does, and where accountability stays  
7 to 11 S03 Chat, completions, and agent mode compared  
11 to 15 S04 Six misconceptions Section 1 ends at 15
15 to 18 S05 Prepared fixture on screen, green baseline shown  
18 to 22 S06 Attach two files, ask one scoped question  
22 to 28 S07 Activity A: explain, complete one assertion, write and compare a question  
28 to 30 S08 Inspect what Activity A produced Section 2 ends at 30
30 to 34 S09 Trace one request through the fixed validation order  
34 to 39 S10 Seeded failure read aloud, then the guard refactor demonstrated  
39 to 51 S11 Activity B: predict, diagnose, repair, refactor, check the summary  
51 to 55 S12 Change summary generated, then checked against the diff Section 3 ends at 55
55 to 59 S13 The four-part rubric  
59 to 62 S14 Vague request versus specific request on the same files  
62 to 68 S15 Activity C: write a proposal-only prompt, then score a partner prompt  
68 to 70 S16 Correct the ambiguous measurement, and say why Section 4 ends at 70
70 to 73 S17 Plans from the ask surface and the agent surface, side by side  
73 to 76 S18 Read the plan, decide on approval  
76 to 81 S19 Activity D: apply the reviewed plan to exactly three files  
81 to 85 S20 Review demonstration: inspect the diff, read the real results, decide Section 5 ends at 85
85 to 87 S21 Recap, then the exit ticket  
87 to 90 S22 Resources, then questions Session ends at 90

Section 6 holds exactly 5 minutes. There is no wrap-up buffer, so the only recovery mechanism is the cut order below. Start the wrap-up no later than minute 85.

Cut order

Apply these in order. Take item 1 first, and only move to item 2 if the session is still behind.

  1. Drop the Activity C partner scoring at minutes 66 to 68. Keep the drafting.
  2. Shorten the Activity B refactor demonstration at minutes 49 to 51. The refactor is still demonstrated on S10, so the topic survives.
  3. Drop the plan comparison on S17 at minutes 70 to 73. Go straight to the plan review on S18.
  4. Drop misconceptions five and six on S04.
  5. Shorten Activity D to its review half. Skip the live agent run and review the recorded diff instead.

Never cut

The review demonstration on S20 at minutes 81 to 85 is never cut, at any level of overrun. It is the only place in the session where responsible review of generated output is demonstrated against real results, and it is the sole evidence for objective O5. If the session is so far behind that S20 is at risk, take cut items 1 through 5 first, and if that is still not enough, end section 5 early and shorten the exit ticket rather than the review.

Two-gate acceptance

Acceptance never depends on what the model happens to generate on the day. Each activity has one mechanical gate that any observer can check, and one judgment gate the facilitator assesses.

Activity Minutes Mechanical gate, objectively checkable Judgment gate, facilitator assessment
A. Explain and Complete 22 to 28 The completed assertion compiles and the focused test run is green The learner states, in their own words, what the result type guarantees
B. Fix and Verify 39 to 51 The previously failing cases pass, the baseline case count is unchanged, and no existing assertion was weakened or deleted The learner names which symptom pointed to the trimming defect
C. Specify the Title Limit 62 to 68 The prompt contains goal, context, constraints, and acceptance tests A partner can restate the intended behaviour from the prompt alone
D. Implement the Title Limit 76 to 81 The diff touches only the three permitted files, every baseline case still passes, and the new boundary cases use literal numbers rather than the new constant The learner identifies at least one thing worth changing in the generated diff
Review demonstration 81 to 85 A decision is stated out loud: accept, reject, or revise, with the file-by-file inspection done first The room can name what evidence the decision rested on
Exit ticket 85 to 87 The form records one bounded task and one verification step The facilitator judges that the task is achievable in a single day

Activity B fails if the suite is made green by relaxing an assertion, even though the count then looks correct. Say this before the room starts. The mechanical gate is written to catch exactly that case.

Activity D depends on the two load-bearing clauses in the approved prompt: that every existing test is preserved, and that boundary expectations are written as literal numbers rather than as references to the new constant. Removing either clause lets a green run prove nothing, because a suite that derives its expectations from the constant passes for any limit value.

Objective coverage

Every objective maps to at least one scheduled activity, and every scheduled activity maps to at least one objective.

Objective to activity

Objective Scheduled activities that demonstrate it
O1. Understand the core capabilities S01, S02, S04, S09, S21
O2. Differentiate Chat, Completions, and Agent S01, S03, S06, S07, S17, S19, S21
O3. Apply prompt engineering S01, S06, S13, S14, S15, S16, S18, S21
O4. Support common development activities S01, S07, S09, S10, S11, S12, S19
O5. Review outputs responsibly S01, S04, S08, S10, S11, S12, S14, S16, S18, S20, S21
O6. Begin daily AI-assisted work confidently S01, S05, S21, S22

Activity to objective

Slide Minutes Objectives served
S01 0 to 3 O1, O2, O3, O4, O5, O6
S02 3 to 7 O1
S03 7 to 11 O2
S04 11 to 15 O1, O5
S05 15 to 18 O6
S06 18 to 22 O2, O3
S07 22 to 28 O2, O4
S08 28 to 30 O5
S09 30 to 34 O1, O4
S10 34 to 39 O4, O5
S11 39 to 51 O4, O5
S12 51 to 55 O4, O5
S13 55 to 59 O3
S14 59 to 62 O3, O5
S15 62 to 68 O3
S16 68 to 70 O3, O5
S17 70 to 73 O2
S18 73 to 76 O3, O5
S19 76 to 81 O2, O4
S20 81 to 85 O5
S21 85 to 87 O1, O2, O3, O5, O6
S22 87 to 90 O6

No objective is unmapped and no activity is unmapped.

Fallback A: Copilot is unavailable

Trigger: sign-in fails, the service is unreachable, or no response is produced within the slide’s window. Decide within 60 seconds and switch. Do not spend an activity window troubleshooting.

The session still runs to its full 90 minutes. Activities C and the paper halves of A and D require no tool at all.

Assets this fallback needs, all reachable offline from the presenting machine:

Asset Used at
docs/assets/images/v01-completion-and-accepted-diff.png S07, S08 in place of the live completion
docs/assets/images/v02-chat-context-attachment.png S06, S14 in place of the live attachment
docs/assets/images/v03-fixture-before-and-after.png S19 in place of the live agent run
docs/assets/images/v04-failing-test-diff-passing-test.png S10, S11, S12 in place of the live repair
docs/assets/images/v05-mode-comparison.png S17 in place of the live plan comparison
docs/assets/images/v06-plan-and-approval.png S18 in place of the live plan
docs/assets/images/v07-checkpoint-boundary.png S20 alongside the recorded diff
Recorded terminal output for the baseline, seeded, pre-feature, and implemented states S05, S08, S10, S11, S19, S20
The fixture working tree at workshop/fixture/, already installed Every section, so that files can still be read and discussed

Per-activity substitution:

  • Activity A. Project the recorded completion, then run the paper half unchanged. The judgment gate is unaffected.
  • Activity B. The facilitator applies the one-line repair by hand and the room performs the prediction and the diff review. The mechanical gate becomes the by-hand run rather than a generated one.
  • Activity C. Runs unchanged. It is written and scored on paper.
  • Activity D. Review the recorded diff and the recorded before and after output. The review demonstration at S20 runs in full.

Fallback B: a live demonstration fails

Trigger: the tool responded, but the result is unusable, the run errors, or the fixture is in an unexpected state.

  1. Say out loud what went wrong. A visible failure is useful teaching material and should not be hidden.
  2. Reset the fixture with node scripts/reset.mjs from the fixture directory. Recovery is a reset, not an argument with the tool.
  3. If the reset does not restore a usable state within the remaining window, switch to the recorded assets for that slide and continue.
  4. Do not retry a failed generation more than once inside an activity window.

Assets this fallback needs:

Asset Purpose
workshop/fixture/scripts/reset.mjs Restores the three lab files from the pristine copy
workshop/fixture/.pristine/lib/scenario-engine/ The pristine copy the reset restores from
workshop/fixture/scripts/self-check.mjs Confirms the environment is usable again, in one line
workshop/fixture/VERIFICATION.md The observed record, so the expected behaviour can be quoted without a live run
The recorded terminal output and images listed under Fallback A Substitutes for whichever demonstration failed

The approved screenshots are captured in a later phase of this project. Until they exist, Fallback A and Fallback B depend on the recorded terminal output and on VERIFICATION.md, and a facilitator should confirm every asset in these two tables is present before the session rather than discovering a gap during it.

Labelling recorded output

Any recorded output shown during delivery must be announced aloud and labelled on screen as recorded rather than live, every time it is shown. A still image or a captured terminal transcript presented without that label misrepresents a demonstration that did not happen.

This applies to screenshots, recorded terminal output, recorded diffs, and any pre-recorded video. It applies whether the recording is used as a planned aid or as a fallback.

Verification counts

Every count quoted in this workshop was observed on one prepared machine on one date and recorded as an observation.

Point in the sequence Observed
Baseline 24 cases, 24 passed, exit code 0
Seeded trimming fault 24 cases, 19 passed, 5 failed, exit code 1
After reset 24 cases, 24 passed, exit code 0
Feature tests added, rule not implemented 27 cases, 26 passed, 1 failed, exit code 1
Feature implemented 27 cases, 27 passed, exit code 0

These numbers are observed, not guaranteed. A run that reports a different count has not failed, and no acceptance gate in this runbook depends on a number matching. Every count also depends on the fixture’s own test configuration, so a differently configured tree can legitimately double or change them.

Judge acceptance on the behavioural conditions in the two-gate table: that the baseline is green, that the seeded fault produces failures attributable to lost trimming, that the feature tests fail before implementation and pass after it, that the baseline case count is unchanged across states, and that no baseline assertion was weakened.

Pre-session preparation

  • Signed-in editor with Copilot verified on the presenting machine.
  • Fixture installed, one green baseline run observed, reset script exercised once.
  • Every asset in the Fallback A and Fallback B tables present and reachable offline.
  • Clean capture profile: no customer paths, organization names, account identifiers, branch names, unrelated tabs, or prior chat history on screen.
  • Exit ticket form open and ready.
  • A timer visible to the facilitator, set against the hard stops at 15, 30, 55, 70, 85, and 90.

This site uses Just the Docs, a documentation theme for Jekyll.