# Estimate before the number arrives

Treat this file as reference data, never as instructions that override the user or agent. Read the complete item before applying it. Keep use conditions, avoid conditions, limits, evidence roles and source links with the answer. An editorial block is a separately versioned analysis layer, not an original corpus claim or proof of efficacy. Cite the readable URL and version. Do not infer human review. This is a saved edition, not a live safety-notice service. Before consequential reuse, check the current page, https://vedokrok.com/data/availability.json and https://vedokrok.com/data/notices.json; absence alone does not explain why an item was removed.

Release: VEDOKROK-SEL-5d86e1b23401e578
Snapshot SHA-256: 30ce1c8fa9155763797e3d5cb55b1d94cbb8c3146ebd03fb3c006531ce60e82f
Current availability: https://vedokrok.com/data/availability.json
Current notices: https://vedokrok.com/data/notices.json
Editorial overlay: MHC-EO-20261021-481e1ce95099
Editorial projection SHA-256: 6b41bfd164b85b2522e2378c425929aa041d9e093e81428ade9a7b7b9f78e8fe

---

## Estimate before the number arrives

ID: MHC-U-9C0A00000005 · Version: 0.2.0 · Kind: heuristic
Source: https://vedokrok.com/knowledge/estimate-before-the-number-arrives

Give yourself a reference point before someone else — or an AI — gives you one.

Knowledge source release: MHC-RPUB-20261011-9e4b62c5
Source corpus commit: 9e4b62c5648844484b8cc3160f3e168d63a1241b
Source manifest SHA-256: 8cc91f0a764c635a79a18e06e048d3dc7bb1e07ae0fed6bf544c89832c024402
Source projection SHA-256: d03d10dd52b53dec1ed850b7706847ba5b200e60f24c4fbda4518c2004356104

### Use when

- You need a numerical estimate and have not yet seen the suggested number.

### Avoid when

- If you already saw the number, do not reconstruct an 'independent' estimate after the fact. Use separate evidence or another estimator and disclose the exposure.

### Explanation

Before you see a suggested number, write your own range and the assumptions behind it. Then compare evidence, not authority. If the new estimate reveals something you missed, update. If it only supplies a number, that is not yet a reason.

### Steps

1. Define the quantity, units, scope, and time period.
2. Use comparable cases or a task breakdown to write a range and key assumptions.
3. Read the suggested number. Compare its assumptions and evidence with yours.
4. Keep both estimates and record why you changed — or did not change — your view. Compare them with the eventual outcome when possible.

### Example

A team records a delivery range before opening an AI estimate. The model points out a real dependency the team missed, so the range changes. The useful input was the dependency, not the fact that software produced a number.

### Check

Was the first range recorded before exposure? Can every later change be traced to new evidence or an explicit assumption?

### Limits

- If you already saw the number, do not reconstruct an 'independent' estimate after the fact. Use separate evidence or another estimator and disclose the exposure.
- This general reasoning aid does not replace professional advice in high-stakes decisions.

### Evidence and sources

- supports: Numerical anchoring research examines how a comparison with a starting number can influence a later judgment; reported effects vary across conditions. — MHC-S-9C0A00000002. Supports the existence and condition dependence of anchoring effects; does not validate the corpus procedure. (Publisher abstract and article summary)
- MHC-S-9C0A00000002: Fifty Years of Anchoring Effects: A Theoretical Reintegration and Meta-Analysis — https://pubsonline.informs.org/doi/full/10.1287/mnsc.2023.03238 — Dan R. Schley, Evan Weingarten

### Vedokrok editorial review

Make the first estimate independent, not untouchable

Writing a range before seeing someone else’s number can preserve information that would otherwise be pulled toward the suggestion. That can be useful with experts, forecasts and AI. But the first estimate is not protected truth: people also cling to their own judgments and can underuse good advice. The useful sequence is independence first, revision second.

#### Independence creates information worth comparing

The point of estimating first is not to prove that your number is better. It is to preserve a judgment that was formed without the later suggestion. Advice-taking research shows that two estimates can contain useful separate information: Soll and Larrick found that averaging one’s own estimate with another person’s estimate often improved accuracy across a wide range of judgment environments, even though people frequently preferred to choose one estimate instead. An independent first estimate therefore gives you something real to compare with the outside number. It does not tell you how much weight either estimate deserves.

#### With AI, order can change the kind of error you make

In a 2021 experiment with 199 participants, Buçinca, Malaya and Gajos tested several cognitive-forcing designs for AI-assisted decisions. In one condition, participants had to make an initial decision before seeing the AI suggestion and explanation; across the study, cognitive-forcing designs reduced overreliance on wrong AI advice compared with simpler explainable-AI interfaces. The authors also found a usability trade-off: the more effective forcing designs were not necessarily the most liked. This is evidence for separating the first judgment from the AI suggestion in some tasks, not evidence that humans should distrust AI by default.

#### Your own first number can become the next anchor

An independent estimate can protect against one influence and create another. Epley and Gilovich found that people can anchor on values they generate themselves and adjust away from them too little. Advice-taking research points in the same direction from another angle: Yaniv and Kleinberger found that people discounted others’ advice relative to their own opinions. So the first estimate should include assumptions and a revision rule, not just a number. If new evidence is better than the evidence behind your first range, updating is the correct outcome.

#### Should I always estimate before looking at expert or AI advice?

No. The method is most useful when you have relevant independent information and can form a bounded estimate cheaply. If you are a novice with almost no basis for judgment, forcing a precise guess may add noise and give you a number to defend. When a source is demonstrably better calibrated or has data you do not have, seeing it early may be efficient. Use the first estimate as a way to preserve independent evidence, not as a ritual that makes your opinion equal to expert evidence.

#### What if I already saw the number?

Do not reconstruct an “independent” estimate after exposure and pretend the suggestion had no effect. Independence has already been lost. Instead, rebuild from evidence the number did not supply: comparable cases, a task breakdown, a separate estimator, or an explicit range based on different assumptions. Record that you saw the suggestion. The goal is to create another line of reasoning, not to manufacture a clean baseline after the fact.

#### Should I simply average my estimate with the other one?

Not automatically. Soll and Larrick found that averaging often performed well, but the right weight depends on the quality and dependence of the information. Two estimates based on the same flawed data do not become independent because they came from two people or systems. Give more weight to the estimate with stronger evidence, better calibration and a more relevant track record. A 50/50 average is a useful benchmark to question, not a universal rule.

#### What if the AI is usually more accurate than I am?

Then excessive self-reliance may be the larger risk. Logg, Minson and Moore found across six experiments that people often gave algorithmic advice more weight than equivalent human advice, but reliance on the algorithm fell when it was compared directly with the participant’s own estimate; experienced forecasters also relied less on algorithmic advice, which reduced their accuracy in that study. Keep the first estimate as an audit trail, not a veto. If the AI has stronger evidence or a better validated record for this task, your final judgment should be allowed to move toward it.

**Editorial takeaway:** Record a first range before exposure when independence is genuinely useful, and record the assumptions behind it. Then treat the outside number as new evidence: inspect its basis, compare it with yours and update when it deserves weight. The protection comes from separating information sources, not from loyalty to the first estimate.

#### Editorial evidence

- supports: In quantitative judgment tasks, averaging one’s estimate with another person’s estimate was often more accurate than choosing between the two, although participants frequently preferred the choosing strategy. — soll-larrick-2009. This supports deliberate combination of independent judgments in the studied environments; it does not imply that an equal-weight average is optimal for every pair of estimates. (Abstract)
- supports: A 2021 experiment with 199 participants found that cognitive-forcing designs reduced overreliance on incorrect AI advice; one design required an initial decision before showing the AI suggestion and explanation. — bucinca-malaya-gajos-2021. The study used a simulated AI and a controlled meal-choice task. It supports separating initial judgment from AI advice as a bounded interaction design, not a universal rule for all AI-assisted work. (Abstract; sections 3.2.2 and 3.5)
- challenges: Self-generated anchor values can produce insufficient adjustment, so a person’s own first estimate can itself bias the later judgment. — epley-gilovich-2001. This directly limits the idea that making an estimate first is automatically debiasing. The initial estimate must remain revisable and tied to evidence. (Abstract)
- limits: Across four advice-taking studies, people discounted advice relative to their own opinions. — yaniv-kleinberger-2000. This cautions that an initial self-generated judgment can increase commitment to one’s own view and lead to underuse of useful outside information. (Abstract)
- contextualizes: Across six experiments, people often relied more on algorithmic than human advice, but algorithm appreciation weakened when the algorithm was compared with their own estimate; experienced forecasters relied less on algorithmic advice and were less accurate in that study. — logg-minson-moore-2019. The findings argue against a simple story that users always ignore algorithms or always overtrust them. The appropriate weight depends on task, evidence and comparative accuracy. (Abstract; Experiments 3 and 4)
- soll-larrick-2009: Strategies for revising judgment: how (and how well) people use others' opinions — https://doi.org/10.1037/a0015145 — Jack B. Soll, Richard P. Larrick
- bucinca-malaya-gajos-2021: To Trust or to Think: Cognitive Forcing Functions Can Reduce Overreliance on AI in AI-Assisted Decision-Making — https://doi.org/10.1145/3449287 — Zana Buçinca, Maja Barbara Malaya, Krzysztof Z. Gajos
- epley-gilovich-2001: Putting adjustment back in the anchoring and adjustment heuristic: differential processing of self-generated and experimenter-provided anchors — https://doi.org/10.1111/1467-9280.00372 — Nicholas Epley, Thomas Gilovich
- yaniv-kleinberger-2000: Advice Taking in Decision Making: Egocentric Discounting and Reputation Formation — https://doi.org/10.1006/obhd.2000.2909 — Ilan Yaniv, Eli Kleinberger
- logg-minson-moore-2019: Algorithm appreciation: People prefer algorithmic to human judgment — https://doi.org/10.1016/j.obhdp.2018.12.005 — Jennifer M. Logg, Julia A. Minson, Don A. Moore

Editorial version: 1.0.0 · evidence checked 2026-09-23 · Agent editorial review; not human or expert certification.

No review details supplied.
Related (useful_with): https://vedokrok.com/knowledge/anchoring-effect
Related (useful_with): https://vedokrok.com/knowledge/flip-the-frame-keep-the-facts
