Accept-Reject Matrix: The Comparable-by-Comparable Decision Record

Definition

The accept-reject matrix is the study’s record of the comparable-by-comparable decision — the table in which every candidate comparable that survived the quantitative screening appears as a row, with the columns that say what happened to it: the company (identifier, segment, the screening results), the [qualitative screening] (/docs/benchmarking/qualitative-screening) verdict, and the decision (accept, reject, or flag — held for review) with the recorded reason for that decision (the product mismatch, the revenue-mix gap, the asset difference, the extraordinary item, the [size] (/docs/glossary/size-filter) outlier, the source the check came from). The matrix is the boundary between the search population (the database results, the candidates) and the comparable set (the companies that actually carry the range) — every company that drops out leaves the matrix with a reason on file, and every company that stays has its inclusion basis on file.

The accept-reject matrix, in one row:
  1. The candidate (the company, the segment, the NIC code, the FY data)
  2. The screens passed (the quantitative filters it cleared)
  3. The qualitative verdict (what the company-by-company review found)
  4. The decision (accept / reject / flag)
  5. The reason (the specific fact that drove it — the recorded rationale)
  6. The evidence (the filing, the website, the GLEIF record it rests on)

The column	The content
The identifier	The company (name, ticker/LEI, the segment used, the NIC code)
The screens	The quantitative results it carried in (turnover, the PLI value, the filters it cleared)
The verdict	The qualitative finding (product, revenue mix, assets, geography, the mismatch found — if any)
The decision	Accept / Reject / Flag (the held-for-review state, not a silent drop)
The reason	The specific fact (the “70% of revenue from a product the tested party does not trade” — not “not comparable”)
The evidence	The source the reason rests on (the annual filing, the website capture, the database record the extract came from)

The working read (the defending the matrix guide): the TPO and the IRS do not re-run the search — they read the matrix. The examination question is “why did you keep this one and drop that one?”, and the matrix is the answer in tabular form. The discipline is three-way: no silent exclusions (a company that disappears between the search and the set is a red flag), no unreasoned acceptances (an accept without a stated basis is the same gap on the other side), and the reasons are specific (the fact, the number, the source — not the category). Where the screening is AI-assisted, the matrix is more important, not less: the AI’s per-company rationale becomes the recorded reason, and the human reviewer’s accept/reject becomes the decision on top of it — the matrix is where the two meet, and where the evidence ledger gets its rows.

Example

A TNMM study on a routine distributor: the search returns 41 candidates; the quantitative screens (the size filter, the profitability filter, the asset filter) leave 19. The matrix for the 19:

Company	Qualitative finding	Decision	Recorded reason
C1	Same product line, similar margins	Accept	Products aligned (trading, no services); assets comparable (working-capital intensive, no major PP&E)
C2	68% of revenue from services	Reject	Revenue mix — services dominant; tested party is 97% trading (the mix breaks comparability)
C3	One-off spectrum charge in FY24	Flag	Extraordinary item — adjusted per the extraordinary-events analysis; re-screen after adjustment
C4	Segmented filing unavailable	Reject	No segment data; consolidated results include a manufacturing division (the segment mismatch is unresolvable)
…	…	…	…

The result: 12 accepted (the comparable set), 6 rejected (each with its specific reason and source), 1 flagged then accepted (post-adjustment value inside the range, the adjustment documented). The matrix page in the Local File is this table — and the TPO’s follow-up question (“C2: show the services split”) is answered by the source column, without a rescreen.

See also

• Defending Your Accept-Reject Matrix Before the TPO
• Qualitative Screening
• AI-Assisted Comparable Screening
• Comparable Set

FAQ

What is the difference between the accept-reject matrix and the comparables table? The comparables table (the final screen of the study) lists the accepted companies and their PLI values — the input to the range. The accept-reject matrix covers every screened candidate — the accepted, the rejected, the flagged — with the reason for each. The comparables table is the matrix’s accept rows; the matrix is the decision history around them, and it is the matrix the examination reads, because it shows both sides of every decision.

Do rejected comparables affect the arm’s length range? No — the range is computed on the accepted set only (the comparable set). The rejected companies’ values never enter the IQR; what they do affect is the defensibility of the selection — an unexplained rejection (a company with a better margin dropped without a reason) reads as cherry-picking, and the TPO’s challenge to the range starts with the matrix, not the percentiles. The reason column is the protection.

How detailed must the recorded reason be? Specific enough that a reviewer who has never seen the company can verify it: the fact (the product, the revenue line, the asset), the number (the mix percentage, the margin, the days), and the source (the filing page, the website capture, the database field). “Not comparable” is a category, not a reason. The test: could the reason be checked against a document? If yes, it is recorded; if it can only be felt, it belongs in the reviewer’s margin of judgment — and the decision should be the flag, not the silent reject.