Methodology
How we collect signals, how every score on the site is calculated, and what the numbers can and can't tell you.
What we track
Public sentiment and political will around technology in K-12 schools: AI of every kind, screens and school devices, personal phones, social media, student data privacy, surveillance tech, and named edtech products. We record pushback and support alike.
Each signal is one distinct data point: one board agenda item, one round of public comment at one meeting, one policy, one contract action, one state action, one news story, op-ed, petition or poll. If a story covers a board vote, the vote and the story are two signals grouped as one event.
Every signal links to a source someone opened, with a confirmed date. We paraphrase sources and quote at most a few words. We never name private individuals; we count them. We tag interested parties: vendors, advocacy groups, law firms recruiting districts into lawsuits, and partisan or sponsored polls. Each signal is coded for topic, who's speaking, which grades, what kind of action it is, how strong it is (1 for a single post up to 5 for a binding vote), and whether it pulls back on the technology or leans in.
The temperature
States and districts, 0 to 100, last 90 days1
Score each signal
How hard it pushes back (1.0 strong, 0.7 clear, 0.35 mixed, 0 neutral or supportive) times how much it weighs: a post 1, a news story 2, organized action 4, a formal proposal 6, a binding vote or contract change 8.
2
Count each story once
Signals about the same event are grouped. The strongest counts in full, and extra coverage adds 25% each, capped at double. Only a district's 5 strongest events count (a state's 6), so more research can't inflate a score.
3
Adjust for context
Recent signals count more (60-day half-life). Broad coalitions and big petitions add up to 50%. Larger districts and states are damped for their bigger megaphone.
4
Convert to 0 to 100
A saturating curve: one binding vote against school tech lands a district around 55. A state score pools every signal in the state, statewide and local.
5
Show what we don't know
Checked districts with no pushback score 0. Districts we haven't checked yet are left blank and get no score. Every score carries a confidence level.
Bands: 0 Quiet, 1 to 19 Cool, 20 to 39 Mild, 40 to 59 Warm, 60 to 79 Hot, 80 and up Very hot. When you filter the map to one category or product, the same method runs on just those signals, so a filtered mark reads like the full map.
Ventures
Product and category temperatures1
Pooled nationally
A product's mark pools every signal that names it, anywhere, and counts its 6 strongest events. A category counts its 20 strongest. Both use a wider scale than districts, since they draw on the whole country.
2
Avg. district
The average temperature of the school districts where a product came up. It tells you whether a product tends to show up in hot places, whatever is driving the heat.
3
Named isn't targeted
A signal that names several products counts toward each, and some mentions are incidental (a policy listing approved tools, say). Open a product to see the actual signals.
4
Categories
Eight categories: elementary and secondary screen time, direct student AI use, ambient AI, direct staff AI use, backend and operational tech, personal phones, and consumer apps. Screen time splits by the grades a signal names. A signal can count in more than one category.
5
Leaning in counts too
Contracts approved, pilots launched and supportive comment show up as leaning in. Phone restriction products have no mark, since pushback on phones is demand for them. Consumer apps are listed because they come up constantly, though schools don't buy them.
Policy
Status and stance1
What counts
Policies adopted or revised, policies proposed or on a first reading, state guidance and model policies, and bills and rules. News about the same policy is grouped with it as a source.
2
Status
The furthest step we've seen: adopted (including laws passed), guidance issued, or under consideration. A proposal may have moved on since we last checked.
3
Stance
0 to 100. The average of two codes: the policy's stance (from full support through limits to blanket opposition) and its tone toward the technology. 60 and up pulls back, 40 and under leans in.
4
Categories
The same categories as Ventures, plus phones, social media, and AI and screen rules that don't name specific uses or grades.
5
Coarse for now
Most phone and screen limits land at 75. A later version will code how restrictive each policy is, so an instructional-time phone rule and a bell-to-bell ban score differently.
Get involved
Petitions are the ones we found in the last 90 days, in every direction. Signature counts are what the petition page or news story showed when we checked, so most are higher now. Petitions started by advocacy groups carry the group's name.
National campaigns are organizations with chapters, pledges or toolkits. We opened each site, paraphrased what it asks for, and noted who runs or funds it, including vendor funding and other interested parties. We don't endorse any of them.
Board meetings come from district board portals where the portal allows automated reading. Many portals block it, so for most districts we link to the portal instead. Dated events come from signals we've already collected, each with its source.
Reader tips
Anyone can send us a tip on the Contribute page. Every night we check new tips against public sources: the agenda or minutes, the policy, the bill, the petition, the news story. A tip goes in only when we can confirm it with a source we opened and a date, and then it's coded the same way as everything else. Tips with a political angle need two independent sources.
Tips we can't confirm are set aside. We don't ask who sent a tip, and we never name private individuals from one.
Known limits
Scores reflect research effort as well as pushback. Colorado was our pilot and was searched far more deeply, so its scores partly reflect that. Low scores in thinly covered states mostly mean we haven't found much yet.
We searched more deeply in recent weeks than in earlier months, so monthly counts show our coverage as much as the pace of events.
Every record was classified with AI and hasn't had a human review yet. Some dates are publication dates where the meeting date wasn't known. Social media is mostly behind login walls, so board public comment stands in for much of the grassroots picture.