Stop Grading Your Own Test Results

What each one is actually for

PDSA Cycle

A PDSA cycle starts by choosing one small, specific change you think will help — not the whole overhaul, one testable piece of it — and writing down, before you do anything, exactly what you predict will happen if you run it: a number, a specific outcome, something checkable. That's Plan. Do means actually running the change, once, on a small enough scale that a bad outcome doesn't cost much — a single shift, one client, one week, not the whole department switching over. Study is where most people quietly skip the hard part: you go back to the prediction you wrote down before you started and compare it, line by line, against what actually happened, not against your general impression of how it went. Act is the decision that follows directly from that comparison — adopt the change as-is, adapt it and run another small test, or abandon it — and the loop can run again with the next small change. The whole point of writing the prediction down before running the test is that memory reliably drifts toward whatever makes you look right in hindsight; a prediction on paper is the only thing that can catch that drift.

Commitment Contracts

A commitment contract is a specific agreement you make with yourself in advance, usually through a website or an app built for this, that names the exact goal, the exact deadline, and the exact cost of missing it — money you forfeit to a cause you'd hate to fund, control of a password, or simply a friend who finds out you didn't do it. Some versions add a referee: a person, sometimes a stranger paired through the service, whose job is to certify whether the goal was actually met, because leaving that judgment to the person with the money on the line defeats the entire design. The mechanism is that a future version of you, standing at the actual deadline deciding whether to follow through, isn't trustworthy in the same way the version of you signing the contract today is — today's version wants the goal in the abstract; the deadline version is tired, or busy, or simply doesn't feel like it anymore. Attaching a real, specific cost to failure, verified by someone other than the person who'd benefit from lying, is what keeps the deadline version of you honoring what the planning version signed up for.

What you get out of pairing them

A PDSA cycle is only as good as the honesty of its Study step, and that step has no built-in defense against the exact thing it's supposed to prevent — you grading your own test kindly because you already like the change, or because admitting it failed means starting over. A commitment contract is a tool built for precisely that problem: it takes the verdict on whether something actually worked out of the hands of the one person motivated to fudge it, and hands it to a real cost and, ideally, another person checking. Borrow that structure for your Study step specifically — not for the whole change, just for the moment you sit down to decide whether the result matched the prediction — and a PDSA cycle stops being something you can quietly talk yourself through and becomes something you have to actually answer to.

What PDSA Cycle rests on

• Feedback against prediction corrects belief plausible mechanism
  Comparing the actual result of a change against the predicted result before adopting it corrects mistaken beliefs faster than adopting or rejecting changes based on intuition alone.
• Small scale testing precedes scaling observational studies
  Running a change through a small planned test cycle before wider adoption reveals real-world effects while limiting the risk of being wrong.

What Commitment Contracts rests on

• Accountability to others increases follow through observational studies
  Having another person witness or enforce progress raises the likelihood that a commitment is kept.
• Loss aversion strengthens commitment randomised trials
  Framing a stake as a loss to avoid rather than a gain to earn increases adherence because losses loom larger than equivalent gains.
• Precommitment constrains future choice randomised trials
  Binding oneself in advance to a costly consequence removes the option to yield to in-the-moment temptation.

Where they actually meet

Both practices refuse to let the person running them be the sole judge of the outcome. A PDSA cycle does this by forcing a written prediction to exist before the result does, so there's a fixed target to check against rather than a memory that can drift. A commitment contract does the identical thing by forcing a real cost and often another person's sign-off to exist before the deadline arrives, so there's an external checkpoint the deadline-version of you can't quietly talk past. Neither one trusts the in-the-moment self to grade its own performance fairly — and neither one is wrong to distrust that, because the in-the-moment self has every incentive to be generous.

The principle underneath both

A verdict checked only by the person who wants a particular answer will tend to arrive at that answer, quietly and without anyone lying on purpose — and the fix in both cases is to fix the standard in advance, in writing, before the moment where motivated judgment would otherwise take over.

What that makes the real goal

Once you see that, the actual goal in both cases turns out to be the same thing dressed differently: get an honest verdict on whether something happened the way you said it would, at the one moment you're least inclined to give one. Anything that locks in the standard beforehand and puts the final call somewhere other than your own in-the-moment mood is doing that job, whether it calls itself quality improvement or an accountability app.

Other practices that serve the same goal

Pre-registration in scientific research does this at the level of an entire study: a researcher publicly commits to a hypothesis and an analysis plan before collecting data, specifically so the eventual verdict can't drift toward whichever result the data happens to support — a fix that meta-science research credits with a measurable drop in false-positive findings once journals started requiring it. A workout log paired with a wearable that records heart rate and pace automatically does something similar for physical training: the numbers exist independent of how the session felt, so a confident sense that you pushed hard has something to be checked against besides your own recollection. And an accountability partner who asks for a screenshot or a specific number, rather than a general how'd-it-go, is applying the referee half of a commitment contract without the financial stake — evidence for peer accountability alone, absent money, is weaker and more mixed than for a real financial or social cost attached.

A tradition that already does this

Weekly weigh-in programs like Weight Watchers have combined both moves for decades without naming either one this way: a real number on a scale, checked against a stated goal, is the Study step; reporting that number out loud in front of a room of other members doing the same thing is the referee and the social cost, together, in one meeting. Neither the quality-improvement pioneer W. Edwards Deming, who popularized the plan-do-study-act cycle in manufacturing, nor the commitment-contract services built decades later set out to reproduce that structure — a weekly group weigh-in got there first, for an entirely different problem.

Using the PDSA Cycle to Sharpen a Commitment Contract

Before writing a commitment contract, run its goal through a genuine Plan step first: write down not just the goal and the deadline, but what you actually expect to happen and why you think this particular version of the goal is achievable, in the same specific, checkable language a PDSA prediction requires. Most commitment contracts fail not because the stake wasn't big enough but because the goal was set the way a New Year's resolution gets set — vague, ambitious, untested — rather than the way a PDSA cycle insists on: small enough that you can name, in advance, exactly what evidence would count as it working. Set the first version of the contract around a genuinely small, testable slice of the real goal — not exercising four times a week for a year, but walking on Tuesday and Thursday this week, verified by a photo of the time-stamped route — treat missing it as data rather than as a moral failure, and let the Study step tell you whether the size of the commitment, not just your willpower, was the actual problem before you scale the stake up.

Using a Commitment Contract to Sharpen the PDSA Cycle

The weak point in almost every PDSA cycle is the Study step, and the fix is to attach a small, real cost to skipping or fudging it, the same way a commitment contract attaches a cost to missing a deadline. Before running the test, write the prediction down and hand a copy — not a summary, the literal prediction — to a colleague or a manager who has agreed to ask you, on a specific date, whether the actual result matched it, with a small real consequence attached to dodging the question: you buy the coffee, you present the result at the next team meeting regardless of whether it's flattering, you owe a specific favor. This is not about making the test itself higher-stakes; the test can and should stay small. It's about making the moment of grading it honestly cost something to avoid, exactly the way a referee on a commitment contract makes lying about the outcome cost something. A Study step that nobody but you will ever check is a Study step you can quietly skip on a bad week, and most people do.

You ran the new intake script for one week, and the numbers came back worse, not better

You'd been sure it would help — you Planned for a fifteen percent drop in callback requests, wrote the number down before you started, and ran the new script for one week, the small test PDSA calls for. The actual result came back at a four percent increase in callbacks, not the drop you predicted, and your first instinct, alone with the spreadsheet at your desk, is to reach for a story: the week had two holidays in it, the sample was too small to mean anything, next week will look different. That instinct is exactly the moment a commitment contract's referee exists to interrupt. Because you'd already told your manager, before running the test, that you'd report the actual number at Friday's stand-up no matter what it was, you don't get to quietly extend the test another week to see if the story about the holidays holds up. You report the four percent increase as written, out loud, in the room, and only after that verdict is on the record do you get to propose Study's honest next move — Act: abandon this version of the script, not adapt it, because the prediction and the result didn't just miss, they went the wrong direction entirely.

When this is worth reaching for

The person running the test and the person who badly wants a particular answer are the same person — your own process change, your own habit experiment — and that's precisely the condition under which a self-graded Study step goes soft without anyone ever deciding to let it.

Where this does not apply

For PDSA cycles run inside a team with an actual manager already reviewing the results, or for low-stakes tests where a wrong verdict costs almost nothing either way, adding a formal commitment contract on top is overhead without a real problem to solve — the external check that matters is already there, or the size of the mistake doesn't justify building one. And a commitment contract's financial stake works against you if the goal itself was mis-set by an overly ambitious Plan step; punishing yourself for missing a badly designed test teaches the wrong lesson entirely. Fix the size of the test before you attach a cost to grading it.

What the evidence does and does not support

Loss-based commitment contracts have randomized trial support specifically for goals like weight loss, exercise adherence, and smoking cessation, where a real financial stake measurably raised follow-through compared with no stake at all. The plan-do-study-act cycle itself is a widely used quality-improvement structure with strong observational and case-study support across manufacturing and healthcare settings, though it's rarely been tested as a randomized intervention the way a single behavior-change technique is. Nobody has run a trial that puts a financial commitment contract specifically around the Study step of a PDSA cycle — the argument on this page rests on how the two structures are built, not on a measurement of the combination, and the underlying mechanisms it borrows from, precommitment and loss aversion, are each independently well-evidenced rather than the pairing itself.