Why a Rushed Guess Wrecks a Good Forecast

What each one is actually for

Attention Management

You look at what's actually in front of you — a report, a hard conversation you need to prepare for, a piece of code that needs real thought rather than patchwork — and you put a specific block on the calendar for it: say, 90 minutes on Tuesday morning. Then you defend that block the way you'd defend a meeting with someone important, because you are, in fact, the someone important in this meeting. Phone goes in another room. Chat apps get closed, not minimized. If you share space with other people, you tell them the block is happening so they don't wander in with something that feels urgent but can wait two hours. None of this is about tracking how your day went afterward — it's about deciding, ahead of time, that one specific stretch of the day is not up for negotiation.

Superforecasting

Superforecasting starts with a real question that has an actual, checkable answer at some point in the future — will the bridge repair finish by June, will the client sign by the end of the quarter, will the number on next month's report come in above a certain line. Instead of answering with a feeling — 'probably,' 'I'd be surprised if not' — you write down a number: 65%. Then, and this is the part that makes it training rather than guessing, you write down what would move that number up or down, and you keep every one of these numbers in a running log tied to the actual outcome once it's known. Calibration training is the discipline of going back through that log later and checking: of every time you said 65%, did the thing you were predicting happen close to 65% of the time? If your 90% calls are right only 60% of the time, that's not a feeling you can correct by trying harder next time — it's a specific, measurable miscalibration you now know to adjust for.

What you get out of pairing them

A defended block's payoff shows up the first time you actually finish something that's been fragmenting into fifteen half-attempts across a distracted week — most calendars, left alone, fill every open minute with things that are urgent without being important, and a block is what keeps that from happening to the one task that actually matters this month. Calibration training does something different: it's the only method here that tells you, with an actual number, when your gut has been wrong. Confidence and correctness are not the same thing, and almost nobody gets to see the gap between them without a logged track record to check against — once you have one, a habit of feeling very sure suddenly has to answer to a specific hit rate instead of just feeling right.

What Attention Management rests on

• Implementation intention closes intention action gap randomised trials
  Pre-committing to a specific situational cue paired with a specific response lets the behavior fire automatically when the cue occurs, without depending on in-the-moment willpower.
• Attention is trainable randomised trials
  Repeatedly practicing directing and redirecting attention strengthens the ability to control where it goes, making that control more available over time.
• Context switching degrades performance randomised trials
  Frequent switching between tasks imposes a cognitive cost that reduces the effective output of the time spent.

What Superforecasting rests on

• Decomposition improves estimate accuracy observational studies
  Breaking a complex uncertain question into smaller more tractable sub-questions produces more accurate estimates than judging the whole question at once.
• Explicit probability estimates improve calibration randomised trials
  Forcing a numeric probability rather than a vague verbal hedge makes a prediction falsifiable and trainable.
• Outcome feedback corrects judgment bias randomised trials
  Regularly scoring predictions against actual outcomes and reviewing the misses corrects systematic biases in judgment over time.

Where they actually meet

Line the two up and what they share is a refusal to decide anything important at the moment it's needed. The focus block is fixed in advance, tied to a specific time and place, so that when the moment arrives there's nothing left to negotiate — you don't have to decide, at 9am with a full inbox, whether the hard work matters more than answering an email; that was already decided on Sunday. The forecast works the same way: the number is committed to before the answer is known, tied to a specific, written form — 65%, not 'probably' — so that there's no room afterward to quietly round your memory of your own confidence up or down to match what happened. Both practices are, underneath, the same bet: that a decision made ahead of time, tied to a specific cue, survives contact with a normal day better than a decision made fresh in the moment does.

The principle underneath both

What actually connects these two is a habit of committing before the pressure hits, called an implementation intention in the research literature: a plan tied to a specific cue — 'when it's 9am Tuesday, I close chat,' 'when I'm asked to estimate something, I write a number' — rather than a general resolution like 'I'll focus more' or 'I'll be more careful.' Cue-tied plans like these get carried out where vague ones mostly do not. Layered on top of that is a second fact each practice depends on: the exact skill being protected, sustained attention on one side, calibrated judgment on the other, is not fixed. Both move, and move measurably, with repeated, deliberate practice.

What that makes the real goal

Turning a mental capacity you'd otherwise just hope shows up — enough focus for ninety minutes, enough honest doubt about your own confidence — into something built on purpose is what both practices are actually chasing. A cue-tied plan is what gets that building to happen instead of staying an intention; the fact that the capacity itself is trainable is what makes building it possible at all, whether the material is where your attention goes for an hour or how much weight a gut feeling deserves at a given number.

Other practices that serve the same goal

Other practices aim at the same target through different mechanisms. Deliberate practice, the concept researchers like Anders Ericsson built around expert performance, runs on the same logic — narrow, effortful, feedback-scored repetition beats broad, uncorrected experience — and its evidence base sits mostly in domains like music and chess rather than everyday judgment, which is a real limit worth naming. Prediction markets, where people bet real stakes on the same kind of yes/no questions a forecaster scores privately, chase the identical objective of forcing an honest, checkable number out of a belief, and carry their own separate track record of accuracy in domains like elections and product launches. A structured weekly review — going back over the week's actual output against what was planned for it — reaches toward the attention side of the objective without the numeric-forecasting piece, and it's a far lower-effort way to start if calibration logging feels like too much at once.

A tradition that already does this

Actuarial and underwriting work has, for a long time, combined both halves of this pairing without naming either one: a quiet, protected block of calculation time to work through a table of risk, paired with the after-the-fact discipline of checking predicted loss rates against the claims that actually came in and adjusting the next table accordingly. Neither habit was branded as attention management or calibration training in that world — it was just how the job got done — but the underlying moves, protect the thinking time, score the number against reality, are the same two moves this page is describing.

Giving a Forecast the Undistracted Time It Needs

A forecast made between two meetings, typed out in ninety seconds because someone wants a number, is usually a gut reaction wearing a percentage sign — it hasn't been decomposed into the sub-questions that would actually move it, because there wasn't time to do that. Put the forecast inside a protected block instead: fifteen or twenty minutes, phone away, one hard question in front of you, and actually work through what would need to be true for the answer to be yes, what the base rate for something like this tends to be, what specific piece of information you're missing that would move your number the most. The number that comes out the other side of a protected block is not automatically more accurate, but it is a number you can actually defend later, because you know which sub-question it rests on — and that record is what makes the next round of calibration checking mean something, instead of just re-litigating a gut feeling you can't reconstruct.

Using a Calibration Log to Test Whether the Block Is Working

A protected block is easy to defend by feel — it felt productive, it felt like real work got done — and feel is exactly the kind of measure that calibration training exists to distrust. Borrow the logging habit and point it at your own focus practice: before a block starts, write down a specific prediction about what you'll actually finish in it, as a number if you can manage one — 'I'll have a complete first draft of section two, 70% confidence' — and check it against what actually happened when the block ends. Do this for a few weeks and a pattern tends to show up fast: some blocks reliably deliver close to what you predicted, and some kinds of work you consistently overestimate finishing in ninety minutes no matter how undistracted the room is. That's information the felt sense of a good, quiet morning cannot give you on its own, and it tells you whether the block itself needs to be longer, earlier, or aimed at a different kind of task altogether.

Tuesday, 9:40am, One Number on a Sticky Note

You've blocked 9 to 10:30 on your calendar for the client proposal, phone in the kitchen, door closed. Fifteen minutes in, before you start writing, you make yourself answer a question you've been avoiding: will this actually close by month end? Not a feeling — a number. You sit with it for a minute, break it into what you actually know (the client's budget cycle, how many of these calls have closed before at this stage, what's different this time) and land on 55%, which surprises you a little because you'd have said 'pretty likely' out loud if someone asked. You write '55% — client close, month end' on a sticky note stuck to the monitor, then get back to the actual proposal for the rest of the block. Six weeks later, going through old sticky notes while cleaning your desk, you find it: the deal closed in week five. One data point shows nothing on its own, but it's now sitting in a small, growing pile of numbers you can eventually check yourself against — a pile that only exists because a protected ninety minutes gave you the fifteen quiet minutes it took to write an honest number down instead of a comfortable feeling.

When this is worth reaching for

Your job regularly requires a real prediction — will this ship, will this close, will this hold up — made under time pressure that tempts a fast, unexamined gut call instead. This also holds once calibration training has already been tried on its own and the numbers come out mushy: often that's less a forecasting problem than an attention problem, because a rushed, interrupted five minutes rarely produces a number worth logging in the first place.

Where this does not apply

If your work rarely involves a genuine prediction with a checkable future answer — most creative or relationship-based work doesn't reduce to a percentage in any useful way — bolting a calibration log onto your calendar adds bookkeeping without a real target for it to sharpen. And someone just starting out with protected time blocks has enough to manage in defending the block itself; adding a second discipline, writing and later scoring a number, on top of a habit that hasn't stuck yet is a common way to abandon both. Get the block holding on its own for a few weeks before layering a forecast log inside it.

What the evidence does and does not support

The trial support underneath these two practices sits in different places. Implementation intentions — the 'when X happens, I do Y' structure behind both the calendar block and the logging habit — have a genuinely large body of randomized-trial support across dozens of studies and behaviors, which is part of why both practices lean on the same cue-based structure. The forecasting side has its own separate support: Philip Tetlock's Good Judgment Project found that ordinary people, given feedback on their calibration over time, measurably out-forecasted trained intelligence analysts in some tournaments — a real, checked result about calibration training itself. What nobody has measured is this specific pairing: whether doing your forecasting inside a protected attention block produces better-calibrated numbers than doing the identical logging and scoring without one. That's a reasonable bet based on what each half is good at, not a finding anyone has run.