🧠 AI Literacy Lab HomeDeckPortal

Facilitator Guide

Irvington AI Training — Run of Show

For whoever is standing at the front of the room.

Before the Day

Two weeks out

and a completion-timestamp report from the credit-recovery platform

Three days out

Day-of setup

The one thing that will go wrong

The Wi-Fi. The portal and deck run entirely offline once the files are local — **that is why
they were built this way.** If the network dies, nothing stops. Tell the room that in minute
two; it buys you credibility with the skeptics immediately.

Timing Variants

LengthCut to
90 minWeather Report → Hallucination Hunt → Traffic Lights → Guardrails → Close
3 hoursAdd: Spot the Bot, Time-Back, Prompt Lab, Evidence Stack
Full day (6 hr)Everything below
4×45min after schoolSession 1: Clarity · 2: Capacity · 3: Community + Detection · 4: Build & share

THE RUN OF SHOW — Full Day

☕ 8:30 — Arrival (15 min)

Music playing. On the screen: the AI Weather Report question, nothing else.
Do not open with a welcome slide about the agenda. Open with a question they're already
answering in their heads.

🌤️ 8:45 — The Weather Report (15 min)

Goal: legitimize dissent in the first five minutes.

Everyone self-places: ☀️ Sunny / ⛅ Partly Cloudy / 🌧️ Stormy / ⛈️ Hurricane.
Hands up, or sticky notes on a wall chart.

Say this, close to verbatim:

*"Look around. This room does not agree about AI, and I'm not going to pretend it does.
Some of you use it every day. Some of you think it's actively harmful to kids, and some
of you have read enough to have specific, well-founded reasons for thinking so.*
*Here's the deal for today: I am not going to try to convince anybody of anything.
Nobody has to like AI. Nobody has to use it. That's not the ask.*
*The ask is that everybody leaves here competent about it — able to recognize it,
evaluate it, and teach students to do the same. That's a professional obligation whether
you love this stuff or you'd rather it had never been invented.*
*And if you're in the Stormy or Hurricane group — you have the most important job today.
I'll explain in a minute."*

Then immediately recruit the Red Team. Don't wait. The skeptics need to know their role
before they've had time to settle into resistance.

*"Stormy and Hurricane folks — you're the Red Team. Your job today is to break this stuff.
Find where it hallucinates, where it's biased, where it's bad for kids, where it leaks
data. Write every failure on that chart paper in the corner. At the end of the day, that
chart becomes a district document that goes to every teacher in Irvington with your names
on it.*
*You don't have to change your mind. I'd rather you didn't. Skeptics build better
guardrails than enthusiasts do."*

⚠️ Facilitator note: This is the most important 15 minutes of the day. If you lose the
skeptics here, the rest is theater. If you win them, they'll carry the room for you — and
they'll be the ones who make it stick in September, because their colleagues trust them.

🔍 9:00 — Hallucination Hunt (25 min)

Goal: let them beat the machine before you ask them to trust anything.

Portal → Hallucination Hunt. Timed. Table teams. Four AI-generated passages; find the
fabricated details.

Debrief questions:

The line to land:

*"It's not wrong the way a person is wrong. A person hedges. This thing fabricates a
citation and defends it with exactly the same confidence it uses for the truth. It does
not know which parts it made up, and it will never tell you."*

Red Team: first entries on the Failure Log. Expect them to be better at this than everyone else.

⏱️ 9:25 — The Time-Back Calculator (20 min)

Goal: give value before asking for anything. This is what converts the Hurricane group.

Portal → Time-Back Calculator. Each teacher estimates minutes/week on: parent emails,
rubric writing, differentiating materials, sub plans, practice problem generation,
newsletter/communication.

Then: pick one and do it live, right now, with a real task they actually have to do
this week.

Rules on screen the whole time:

The line to land:

*"That's Sunday afternoon. That's your kid's game. That's the difference between doing
this job for five more years and doing it for twenty."*

⚠️ This is the retention argument, and it's the honest one. Don't oversell it. Some
tasks won't work well. Let those fail publicly — it's more persuasive than a demo where
everything works.

☕ 9:45 — Break (15 min)

🚦 10:00 — The Traffic Light System (45 min)

Goal: the single most useful thing they'll take home.

Teach Red / Yellow / Green. Then Portal → Traffic Light Sorter: table teams sort 12
scenarios.

The sorter is rigged. Several scenarios are genuinely contested and tables will
disagree. That's the point.

Debrief:

*"Table 3 says Yellow, Table 5 says Red, and you're both right — because it depends on
what you're assessing. That's the whole lesson. There is no district-wide answer to
'is AI allowed.' There is only an answer per assignment, and only you can give it."*

Then: 15 minutes of actual work

Everyone opens a real unit they teach and labels every assessment R/Y/G.
They do not leave this block without at least one unit labeled.

The line to land:

*"The most common AI incident in a classroom isn't defiance. It's ambiguity. Four kids
used four different tools in four different ways, and all four thought they were fine,
because 'don't use AI' meant four different things. You can eliminate most of your AI
conflict this year with one sentence on each assignment."*

🤖 10:45 — Spot the Bot (30 min)

Goal: humility. This is the activity that prevents a false accusation.

Portal → Spot the Bot. Six writing samples. Human or AI? Individually first — **no
discussion**, so nobody can hide in a group answer.

Then reveal the scores.

Most people land near chance. Some confident people do worse than chance.

Let the silence sit for a second before you say anything.

The debrief that matters:

*"Now look at what you used to decide. 'Too polished.' 'Weird transitions.' 'No
personality.' 'Vocabulary above their level.'*
*Every one of those also describes a multilingual student writing carefully in their
third language. Every one of them describes a kid who just learned to use a thesaurus.
Every one describes a student who was taught a rigid five-paragraph formula and is
following it exactly the way we told them to.*
*In a district as linguistically diverse as Irvington, if we accuse based on those cues,
our errors are not going to be evenly distributed. They're going to concentrate. And
that's not a technical problem — that's a civil rights problem with a name."*

⚠️ Do not soften this. It's the ethical center of the day. Then pivot immediately to
what does work, so nobody leaves the block feeling disarmed.

🍕 11:15 — Lunch (45 min)

Leave the portal open. People will play with it. That's the design working.

🕵️ 12:00 — The Detective's Desk (40 min)

Goal: replace detection with something that actually works.

Portal → Detective's Desk. A suspected case. Evidence cards, weighted. Tables decide.

Teach the Evidence Stack: oral explanation and in-class baseline at the top,
detector score at the bottom, below the line marked "not evidence."

Then practice the Conversation Protocol — role-play in pairs. One teacher, one student.
Run it twice, swap roles.

The line to land:

*"Never ask 'did you use AI.' That's a yes/no trap — it invites a lie, and if they say no
you've got nowhere to go. Ask 'walk me through how you made this.' You cannot fail that
question if you did the work, and you cannot pass it if you didn't. And you never accused
anybody of anything."*

The second line to land:

*"A zero teaches 'don't get caught.' A redo teaches the content. Pick the one that does
your actual job."*

🏃 12:40 — Speedrun Autopsy (35 min)

Goal: address the district's actual pain, honestly.

Put up the district's real completion data if you have it. If not, describe the pattern.

The reframe — say it plainly:

*"Six months of coursework in six hours. Here's the uncomfortable part: AI didn't create
that. It exposed it. If a course can be finished by a machine in an afternoon, that
course was measuring compliance, not learning — and the machine got very good at
compliance, very fast, for free.*
*That's uncomfortable. It's also good news, because it means the fix is something we
control."*

Then the who-fixes-what table from the framework. Teacher / building / district.

The line that buys you the room:

*"Rows two and three aren't your job. I'm showing them to you so you know they're being
handled — and so you can stop feeling personally responsible for a hole you didn't dig
and can't fill by yourself."*

⚠️ Have an administrator in the room for this block if at all possible. When teachers raise
pacing gates and credit-recovery policy — and they will — someone needs to be able to answer.

☕ 1:15 — Break (15 min)

✍️ 1:30 — Prompt Lab + Become the Reviewer (45 min)

Part 1 (20 min): Portal → Prompt Lab. Rebuild a lazy prompt using
Role · Context · Task · Constraints · Format. Live rubric scoring.

Part 2 (25 min): The reviewer flip. Have AI produce a piece of student work, then
teachers grade it against their own rubric — in track changes, in red pen, out loud.

The line to land:

*"This is your best Green-light assignment, and you just built it. Have the AI write the
essay, then have students tear it apart. A kid who can red-pen an AI essay understands
the genre better than a kid who wrote a mediocre one from scratch."*

🛡️ 2:15 — Guardrails & The Failure Log (30 min)

Hand out the laminated guardrail cards. Walk the non-negotiables.

Then: give the floor to the Red Team. They present the Failure Log.

⚠️ Do not rush this and do not editorialize afterward. This is the moment the skeptics
become authors of the district's AI policy instead of obstacles to it. Let their findings
stand on their own. If a finding is uncomfortable, that's a sign it's valuable.

Photograph the chart. It becomes Irvington AI Failure Log v1, distributed district-wide
with the contributors named.

🎯 2:45 — Commitments & Close (30 min)

Portal → Build Your Syllabus Statement. Everyone generates and saves a real AI-use
statement for a real course.

Each person writes one commitment on a sticky note and puts it on the door on the way out:

Certificates print from the portal.

The closing line:

*"We're not the last generation of teachers before AI. We're the first generation of
teachers who get to decide what it's for."*

Facilitation Notes

If the room turns hostile

Do not defend AI. Agree with the critique, then redirect to the professional obligation.

*"You might be right. It might be bad for kids. And they're using it anyway, right now,
today, whether we're ready or not. So the question isn't whether we approve. It's whether
we're the adults who help them think about it, or the adults who didn't."*

Then put them on the Red Team. Their critique is data, not disruption.

If someone says "this is just cheating with extra steps"

*"Sometimes, yes. That's real and I'm not going to pretend otherwise. And a calculator is
cheating if the skill is arithmetic, and it's a tool if the skill is modeling. Same
object, different assignment. Which is exactly why the traffic light goes on the
assignment and not on the district."*

If someone asks "will this replace us?"

Answer it once, briefly, honestly, then move on. Don't let it eat the day.

*"I don't know. Nobody does. What I'm confident about is that a room full of teenagers
who need an adult to notice them isn't a problem a language model solves. The parts of
this job that are paperwork are genuinely at risk. The parts that are the job aren't."*

If someone demands a detector recommendation

Do not give one. Explain the evidence stack. Offer the process tools instead.
If they push: *"I'll get you something that works. A detector isn't it, and I'd rather tell
you that now than after you've accused a kid who didn't do it."*

If the enthusiasts get bored

Give them jobs: table facilitator, peer coach during the work blocks, or the advanced
challenge in the Prompt Lab. Bored enthusiasts become disruptive; deployed enthusiasts
become your co-facilitators.

Pacing discipline

Nothing on screen for more than 6 minutes without the room doing something.
If you're explaining for 10 minutes, you've lost the back three tables. Cut and move to the
activity — the activity is the content.

Things that will come up and you should have an answer ready

have devices or accounts, and most services are 13+.

evaluate it. That's the whole requirement."*

After the Day

That subgroup is the real measure of whether this worked.