About Mistlift

Who makes Mistlift, and how.

Software does most of the reading and writing. A person decides anything serious. Here is exactly where the line sits.

How a card gets made What we never do
ContextIn beta

Mistlift

A weekly brief on what kids are into online, rated from harmless to serious. Made by one person, with a lot of help from software.

Made by
Jean-Philippe LeBlanc
Languages
English, Québec French
Price
Free while in beta
What to do

Read how it's made, then decide how much to trust it.

Who

One person, a small circle, and a lot of software.

Mistlift is made by Jean-Philippe LeBlanc (JP). There is no newsroom and no staff. JP sets the rules, picks the sources, and approves every serious card himself.

A small circle of people JP trusts reads cards and votes: seen it, still used, rating feels right. Their votes are a signal the checks take into account. They never publish or edit anything.

Behind the brief

How a card gets made.

Five stages. AI does the first three. A person signs off on anything serious.

  1. Read

    Software reads public sources several times a day: news, police and child-safety bulletins, public platform data and RSS feeds. Never a logged-in account, never a kid's account. It flags trends that are rising in more than one place, or that a police or child-safety bulletin names.

  2. Draft

    An AI model writes the card in English and Québec French: what it is, how much it matters, what to do, what to say, and its sources.

  3. Check

    Separate checks read every card. A different AI model and a rule-based safe-messaging check look for anything that explains a harmful thing. Another check confirms the sources were really read.

  4. Sign off

    Connect, Context and Watch cards publish on their own, and JP spot-checks Watch cards. Act and Debunked cards never publish until JP approves each one. No setting skips that.

  5. Re-check

    Every card is checked again each week. A card retires when its trend goes quiet, except standing threats and debunked hoaxes, which stay and come back to the top when they spike again.

Who does what

Where AI does the work, and where a person decides.

AI does

  • Reading the sources and grouping posts about the same trend, in English and French
  • Writing each card in both languages
  • A second-opinion safety check, on a different model from the one that wrote the card
  • Answering questions in the reader app, from published cards only. If a trend hasn't been rated, it says so instead of guessing

A person decides

  • Every Act and Debunked card, one at a time
  • Every card about self-harm, suicide, eating disorders or substance use, whatever its rating
  • Spot-checks of Watch cards
  • Which sources are used, and the rules every card has to pass

The rule-based check runs on everything: cards, the slang guide and the resources library. Act and Debunked cards also need two independent sources, or one from an authority such as police or Cybertip.ca. A card that fails a check twice goes to JP, never to readers.

Never

What Mistlift will never do.

  • Monitor anyone.Mistlift is about trends, not about your kid. Nothing watches what they do.
  • Put anything on your kid's phone.There's nothing to install, and nothing to connect.
  • Keep data about identifiable kids.Names, handles and faces are stripped before anything is stored.
  • Explain a harmful thing.No amounts, no steps, no search terms that lead to it.

Price

Free while in beta.

Spotted a mistake, or a rating that feels wrong? Email jp@mistlift.app or reply to any Mistlift brief. JP reads every report.

Get the Monday brief.

One email a week. Every trend rated, with what to say. Stop any time.

Get the brief