Bet-Writing Primer
One page of patterns and anti-patterns for writing a hypothesis that can actually be tested in six weeks.
A good SHIP bet is small enough to run in six weeks, specific enough to know whether it worked, and broad enough that the team has real choices about how to test it. This is one page of patterns to get you to that.
§ I · The shapeWhat a hypothesis looks like
If [we do X], we expect [Y], because [the underlying reason].
Success means [two or
three concrete, measurable criteria].
The "because" is optional but useful — it forces you to name what you actually believe is happening. The criteria are not optional. A hypothesis without falsifiable criteria is just a wish.
§ II · Weak vs. strongThe same problem, two ways to write it
"We want to use AI to improve student writing in 8th grade ELA."
"If 8th grade ELA teachers use an AI feedback tool on the first essay draft, we expect students to revise more substantively on second drafts. Success means revision counts double, teacher review time drops 20%, and reading-specialist quality ratings stay flat or rise."
"Use AI to help with IEPs."
"If special ed case managers use an AI assistant to draft present-levels paragraphs from MTSS data, we expect drafting time to drop by 50% per IEP. Success means measurable time recovery, parent-reported clarity stays high, and case-manager review remains the final say."
"Modernize how we communicate with families."
"If Communications uses an AI drafting tool for new-family welcome letters in the receiving family's home language, we expect drafting time per letter to drop by 40%. Success means measurable time recovery, no decrease in family satisfaction, and translation accuracy at or above current standard."
"Help teachers use AI better."
"If we give Tier 2 reading teachers a 30-minute Logbook session each Friday on prompt patterns specific to their work, we expect prompt quality (judged by reading specialists) to improve and teacher self-reported confidence with AI to rise. Success means measurable improvement on both."
§ III · Patterns that workWhat good bets share
Pattern 01 · Name the user, the action, and the change
A good hypothesis says who is doing what and what gets different. "Special ed case managers use an AI assistant to draft present-levels paragraphs" — who, what, change.
Pattern 02 · One measurable, one observable
Include at least one quantitative criterion (time, count, rate) and one qualitative one (judgment, satisfaction, feel). Pure number-crunching misses things; pure feel can't be argued with.
Pattern 03 · The "no degradation" criterion
Most bets aim to improve one thing without breaking another. Name the thing you don't want to break. "Drafting time drops by 40%, and family satisfaction does not decrease." The second clause is what keeps you honest.
Pattern 04 · State the underlying belief
The "because" clause is the bet's mental model. "Because AI assistants are good at structural feedback but bad at voice — if we scope feedback to structure only, teachers will trust the output." Naming the belief lets you check it when evidence comes in.
Pattern 05 · Make it small enough to fail
If your bet can't fail in six weeks, it's not really a bet. Bets that are too big either (a) succeed by definition because you redefine success, or (b) become Continuation Briefs forever. Both are tells.
§ IV · Anti-patterns to avoidHow bets go wrong
Anti-pattern 01 · The vision statement
"We want to transform student writing instruction through AI."
Vision, not hypothesis. Strip it down to a specific user, a specific action, and a specific change you can measure in six weeks.
Anti-pattern 02 · The deliverable disguised as a hypothesis
"We will build an AI tool for IEP drafting."
That's the work, not the bet. The bet is what you expect to be true after the tool is built. "If case managers use the tool, drafting time drops by half."
Anti-pattern 03 · Criteria with no failure condition
"Success means teachers find the tool helpful."
How do you know if teachers don't find it helpful? Anchor to something specific — a survey question, a count, a judgment from a third party.
Anti-pattern 04 · The hedged hypothesis
"AI might help speed up some communications work in some cases."
Pre-fail. Don't write a hypothesis you'd be willing to defend either way. Commit, then test.
Anti-pattern 05 · The kitchen-sink bet
"AI for IEPs, lesson planning, family communication, scheduling, and assessment."
One bet per cycle. If you have five candidates, you have five Bet Briefs to write — over five cycles, not one.
§ V · The Logbook can sharpen itIf you're stuck
The Logbook coaches teams through bet-writing. Open a thread in your Google Chat Space and describe the problem in plain language. The Logbook will push back on weak hypotheses, suggest measurable criteria, and offer the formula above filled in for your situation. It is faster than writing alone.