Disqualifier Meaning: Knockout Criteria in Deal Screening

What a disqualifier is, and what it is not

A disqualifier is a condition that, once established for a candidate, removes that candidate from consideration, however well it performs on everything else. In acquisition screening the terms "disqualifier", "knockout criterion" and "hard criterion" are used for the same thing: a rule that ends a company's eligibility under a given mandate, as opposed to a preference that merely makes it less attractive. Our glossary gives the short disqualifier definition. This article covers the drafting: how to write a knockout that is testable, sourced, checked in a sensible order, and honest about what is not yet known.

Three distinctions keep it from drifting into something else:

• A disqualifier is not a preference. "We prefer founder-owned businesses" is a preference. "Exclude any company that is a subsidiary of a listed parent" is a disqualifier. The first belongs on a scale; the second is a gate.
• A disqualifier is not a large score penalty. A minus-fifty weight can still be outvoted by enough positive points elsewhere. A knockout cannot.
• A disqualifier is not a suspicion. The rule fires on established facts. A weak indication that a company might fail the condition is an open question, not an exclusion.

The idea of choosing among options by removing those that fail one attribute at a time has a formal name in decision research. Amos Tversky's 1972 paper in Psychological Review is titled "Elimination by aspects: A theory of choice" ^[1]. This article uses the phrase in its plain sense, eliminating candidates aspect by aspect, and does not rely on the paper's formal model or results.

Knockout rules versus weighted scores

A compensatory rule lets strength on one attribute offset weakness on another; a non-compensatory rule does not. A weighted score is compensatory, a knockout is not, and most screens need both in sequence: knockouts first, scores on the survivors, judgement on the top of the ranking. Our page on screening criteria sets out that full framework, including how to build and weight the scored dimensions; weighted scoring has its own literature, including Robyn Dawes's 1979 paper "The robust beauty of improper linear models in decision making" ^[2]. The comparison below covers only whether a criterion should be a knockout at all.

The fourth row is the reason to be sparing: a bad score misranks a company, while a bad knockout removes it from view. Before promoting a criterion to a knockout, ask whether a partner shown an otherwise excellent company failing it would decline without discussion. If the honest answer is "it depends", the criterion belongs in the score.

Writing a knockout two reviewers will apply the same way

A knockout is only as good as its wording. If two reviewers reading the same evidence can reach different verdicts, the rule is producing what Kahneman, Sibony and Sunstein call noise; their publisher's description of Noise: A Flaw in Human Judgment defines it as "variability in judgments that should be identical", with examples such as two doctors giving different diagnoses to identical patients ^[3]. A knockout should leave as little room for that variability as possible.

Every knockout should state six things:

1. Measure. The observable quantity or fact, named precisely: "average annual receipts", not "size".
2. Threshold or condition. The number, list or state that triggers exclusion, with the direction of the inequality stated.
3. Scope. Which entity is measured: the operating company, the legal entity, or the group including affiliates.
4. Period or as-of date. The window over which the measure is taken, or the date at which the fact must hold.
5. Accepted source. The document or register whose content settles the question, and what is not accepted.
6. Result states. What counts as pass, fail and unknown.

The U.S. Small Business Administration (SBA) size standards are a useful public example of a criterion written this way, whatever one thinks of them as an acquisition filter. SBA states that size standards "vary by industry and are generally based on the number of employees or the amount of annual receipts", and that it "assigns a size standard to each NAICS code", meaning each North American Industry Classification System code ^[4]. The same page defines the scope: "You must include the employees or receipts of all affiliates when determining the size of a business." It defines the period: for federal contracting, receipts are averaged over the latest five complete fiscal years, and employees are averaged per pay period over the latest 24 calendar months. It also notes that applicants to certain other SBA programs may use three or five years, so the same measure can carry a different window in a different context.

The thresholds themselves are code-specific. The SBA page says most manufacturing companies with 500 employees or fewer, and most non-manufacturing businesses with average annual receipts under $7.5 million, will qualify as small, and immediately adds that "there are exceptions by industry" ^[4]. One such exception is directly relevant to the worked example below: SBA's table of size standards, in the version effective March 17, 2023, lists NAICS 238220, Plumbing, Heating, and Air Conditioning Contractors, at $19.0 million in average annual receipts ^[5]. No single SBA figure is a universal "small business" line, and a screen that copies one figure across sectors will misclassify companies.

The version matters too. On August 20, 2026, SBA published a proposed rule proposing new size standards for 338 industry groups and industries ^[6], and on September 24, 2026 it extended the comment period to November 20, 2026 ^[7]. Until a final rule issues, the March 17, 2023 figures remain in force: the codified table at 13 CFR 121.201, checked on October 28, 2026, still lists NAICS 238220 at $19.0 million, with no amendment since that date. A knockout that references a size standard should therefore name the table version it uses. This is a description of published rules, not legal or regulatory advice.

Compare a weak and a testable knockout:

• Weak: "Exclude companies that are too big."
• Testable: "Fail if the company's most recent reported annual revenue, for the operating group including majority-owned subsidiaries, exceeds $40 million, as stated in a filed document or in the company's own published materials dated within the last 24 months. Unknown if no such statement exists."

Choosing the source for each knockout

The accepted source is part of the rule. A knockout that names no source is answered from whatever the reviewer finds first, and different reviewers find different things. For each criterion, name the record that settles it, the record that only suggests it, and how old it may be.

Record the source date and the date it was read, and retain the supporting passage rather than only a link. Our evidence standards page sets out why a quoted, dated passage is easier to review than a URL.

Ordering knockouts by cost to check

The order in which knockouts are checked does not change who survives. It changes what the screen costs: every record put through an expensive check that a cheap check would have eliminated is wasted time.

A simple heuristic is to order knockouts by the cost of a check divided by the share of records the check is expected to eliminate, lowest first. Cheap structured checks that remove many records go first; expensive judgement checks that remove few go last.

The worked example below is fictional. The minutes per record and the pass rates are assumptions chosen for illustration, not measured figures, and the arithmetic ignores unknowns for simplicity.

Applied to a fictional list of 1,000 companies in the order shown, the checks consume 1,000 × 1 + 400 × 1 + 360 × 4 + 216 × 8 + 173 × 15 minutes, about 7,160 minutes, or roughly 119 hours, leaving about 121 companies. Run in reverse order, from K5 to K1, the same assumptions give 1,000 × 15 + 700 × 8 + 560 × 4 + 336 × 1 + 302 × 1, about 23,480 minutes, or roughly 391 hours, for the same survivors. Under these assumptions, the reverse order costs about three times as much.

Real screens will not match these figures, and the point does not depend on them. Time your own checks on a sample before fixing the order, and revisit it when the list source changes, since pass rates depend on how the list was built.

What to do with "unknown"

Every knockout has three possible results, not two: pass, fail and unknown. Unknown means that no accepted source settled the question within the effort allowed for it. It is a fact about the research, not about the company.

Missing data has a formal literature in statistics; Donald Rubin's 1976 paper "Inference and missing data" in Biometrika is one example ^[8]. A screen does not need that apparatus to adopt the practical rule: absence of evidence is recorded as absence, and it is never silently converted into a verdict.

Both silent conversions are costly:

• Unknown treated as pass. The company reaches scoring and partner review on an unverified assumption, and the error surfaces after the expensive work.
• Unknown treated as fail. The company leaves the list with no defensible reason. If unknowns cluster where public information is thin, the screen quietly removes that whole segment.

The workable alternative is a research queue. Records with one or more unknown knockouts and no failed knockout stay visible, with the open questions listed. The mandate owner decides how much effort the queue receives, and that policy is recorded separately from evidenced exclusions. When reporting pass rates, state the denominator and how many records remain unknown.

Knockouts that backfire

Several patterns turn a knockout into a source of error:

• Thresholds copied from a single source. A number lifted from one regime or one sector, such as a single SBA figure applied across industries, will misclassify companies whose rules differ. Name the source and version of every threshold.
• Proxies standing in for the real fact. A headcount range in a profile is not revenue. If the rule is about revenue, a proxy can produce "unknown" or a review flag, never a fail.
• Stale data. A two-year-old ownership record can remove a company since sold to a permitted owner, or keep one that has not been.
• Miscounted affiliates. SBA's own rule counts the employees or receipts of all affiliates, and treats control as existing when an external party has 50% or more ownership ^[4]. An acquisition screen needs its own scope statement; without one, one reviewer measures the operating subsidiary, another the group, and related entities' revenue may be counted twice.
• Rules that change without a version. When a mandate's knockouts change, earlier verdicts were made under different rules. See our article on re-screening when criteria change for handling that transition.
• Too many knockouts. Each additional gate removes companies from view. Matters of degree belong in the score.

Worksheet: five knockouts for a fictional HVAC-services thesis

The thesis and worksheet below are fictional and illustrative only. They describe no real fund, mandate or company, and the minutes per record are assumptions.

Fictional thesis. A fictional sponsor seeks a commercial heating, ventilation and air-conditioning (HVAC) service and maintenance business in three Midwestern states, independently owned, with recurring service revenue rather than new-construction installation as its main line.

A note on how the fictional sponsor might handle size. SBA's size standard for NAICS 238220 is a receipts figure for federal programs, not an acquisition size band ^[5]. The sponsor would write its own size rule with the same anatomy (measure, scope, period, source, version) and might keep it as a scored dimension, since private contractors often publish no revenue figure.

Before using a worksheet like this, have two reviewers apply it independently to the same sample, compare verdicts rule by rule, and rewrite any rule where they disagree for reasons other than missing evidence.

Docket's Triage agent asks a mandate's disqualifying questions before detailed research begins, and its Audit agent leaves unsupported answers open rather than filling them.

Sources

1. Size standards — U.S. Small Business Administration, undated page; size standards based on employees or annual receipts, assigned per NAICS code, affiliates included, 50% control test, averaging periods, the "most manufacturing / most non-manufacturing" generalisation and "exceptions by industry". Accessed 2026-10-28.
2. Table of Small Business Size Standards Matched to North American Industry Classification System Codes — U.S. Small Business Administration, version effective March 17, 2023; NAICS 238220 at $19.0 million average annual receipts. Accessed 2026-10-28.
3. Small Business Size Standards (proposed rule, document 2026-17042) — Federal Register, August 20, 2026; SBA proposes new size standards for 338 industry groups and industries. Accessed 2026-10-28.
4. Small Business Size Standards (extension of comment periods, document 2026-19546) — Federal Register, September 24, 2026; comment period extended to November 20, 2026. Accessed 2026-10-28.
5. Noise: A Flaw in Human Judgment — Hachette Book Group, publisher's description of the book by Daniel Kahneman, Olivier Sibony and Cass R. Sunstein; definition of noise as "variability in judgments that should be identical". Accessed 2026-10-28.
6. Tversky, A. (1972). Elimination by aspects: A theory of choice. Psychological Review, 79(4), 281-299 — American Psychological Association; cited for the existence and title of the paper only (metadata checked via Crossref). Accessed 2026-10-28.
7. Dawes, R. M. (1979). The robust beauty of improper linear models in decision making. American Psychologist, 34(7), 571-582 — American Psychological Association; cited for the existence and title of the paper only (metadata checked via Crossref). Accessed 2026-10-28.
8. Rubin, D. B. (1976). Inference and missing data. Biometrika, 63(3), 581-592 — Oxford University Press; cited for the existence and title of the paper only (metadata checked via Crossref). Accessed 2026-10-28.
9. 13 CFR 121.201, What size standards has SBA identified by North American Industry Classification System codes? — eCFR, current text checked 2026-10-28 (no amendment since March 17, 2023); NAICS 238220 at $19.0 million. Accessed 2026-10-28.