Defending Your Accept-Reject Matrix Before the TPO (and IRS)

The Accept-Reject matrix is where the benchmarking study stands or falls in examination. The range is arithmetic; the pool is judgment. When the TPO or the IRS challenges the study, the challenge is almost always the pool — and the pool is the matrix: every company in it, every company out of it, and the reason for each. This guide is the preparation: what is scrutinized, how the evidence file is structured, how the overrides are reasoned, and the questions to rehearse before the proceeding.

What the authority scrutinizes

The examination does not read the matrix for the range — it reads it for the method. The recurring scrutiny points:

Scrutiny point	What is actually being tested
The rejects, one by one	Is each rejection grounded in evidence that exists for that company — or is it a label? “Not comparable” with no substance is the fastest way to lose the whole matrix’s credibility
The accepts, one by one	The accepts get examined too — a company with a size gap, related-party revenue or an adjacent function that was kept is an admission that the screen was selective
Consistency across companies	The same factor (say, a 2× size gap) rejected in company A and accepted in company B without a distinguishing reason is an inconsistency the authority will tabulate
Consistency across years	The pool and the reasons moving year to year in step with the tested party’s position is range shopping
The override trail	Where a company was kept against the screen’s recommendation (or dropped against it), the rationale must exist and must be the professional’s — not the model’s default
Reconstructability	Can the authority rebuild the pool from the stated source, code, string and filters — and get the same companies? A pool that only exists in the consultant’s working file is a pool the authority will replace with its own

The through-line: the matrix is examined as a method, and a method is defensible when every step is stated, every decision is reasoned, and the whole thing is reconstructable.

The evidence file structure

The defence is a file per company, not a spreadsheet of conclusions. The standard structure:

Study / pool
├── 00  Search design        code, string, size bounds, candidate counts per stage
├── 01  Candidate list        every company at every stage, with its stage outcome
└── per company
    ├── identity              legal name, registration/LEI, source of the data
    ├── financials            the lines used, the period, the source (filing/database)
    ├── profile evidence      business description, product/revenue mix, the
    │                         comparability-factor assessment
    ├── disposition           accept / reject / flagged-resolved
    ├── reason                the specific, evidence-grounded rationale
    └── override              if the final disposition differs from the screen's
                              recommendation: who, what, and why

Two properties make the file work in a proceeding:

1. Every company is present — accepted, rejected or excluded for data reasons. The absence of a company from the file is read as the absence of a decision.
2. Every number has a source — the filing page, the database extract, the registry record. The reason refers to evidence in the file; the evidence is in the file.

Where AI-assisted screening is part of the method, the file carries the ground-truth layer too: what the screen recommended, what the professional decided, and the reason for each divergence. That layer is the answer to the question “who decided this, and on what” — the question the matrix defence cannot dodge.

The override rationale

Overrides are not a weakness — a matrix with zero overrides is itself suspicious (it implies either a perfect screen or a rubber stamp). The discipline is in the record:

• Keep-against-recommendation (the professional retains a company the screen rejected): the reason is the specific distinguishing fact — “the screen rejected on product-line similarity, but the segment disclosure shows the product line is 8% of revenue; the function is comparable” — with the segment evidence attached.
• Reject-against-recommendation (the professional drops a company the screen accepted): the reason is the specific disqualifier the screen missed — the related-party revenue in the notes, the customer concentration, the risk the website description did not capture.
• No silent overrides. An override without a recorded rationale is the single most damaging line in the file — it converts a defensible method into an unexplained choice.

The proportion of overrides is itself a review metric: a healthy file has overrides where the judgment genuinely diverges from the screen — and a pattern of one-directional overrides (always keeping, always dropping) is a signal that the screen’s input (the tested party profile) was wrong, which is a different fix: correct the profile, not the dispositions.

The questions to rehearse

The TPO’s and the IRS’s matrix questions run a predictable script:

1. Walk me through how the pool was built. — code, string, size, thresholds, counts at each stage. The answer is the search design, stated fluently.
2. Why was company X rejected? — the specific reason, the specific evidence. Practice the three rejects the file is weakest on; they will be asked.
3. Why was company Y kept? — the accepts with the gaps (size, related-party, adjacent function) are the real questions. The reason is prepared, not improvised.
4. Company Z was rejected this year and accepted last year — what changed? — the year-on-year consistency answer, with the factual change identified.
5. What did your screening tool recommend, and where did you disagree? — the override trail, company by company, with the rationales.
6. Show me the source for this number. — the filing page. If the answer is “the database”, the follow-up is “and the database’s source”, and the database’s source is the filing — so the file should point at the filing.
7. Rebuild your pool from the stated filters. Same companies? — the reconstructability test. If the answer is “approximately”, the file has a problem the authority will find.

The position in the proceeding

The matrix defence sets the frame for everything downstream. Where the pool survives scrutiny, the range stands, and the examination moves to the adjustments and the tested party’s position — a fight the file is built to win. Where the pool falls, the authority rebuilds it on its own filters (which will be narrower, and will price the tested party differently), and the study becomes an adjustment with a method attached that was not chosen. The preparation above — the file structure, the reasoned dispositions, the override trail, the rehearsed answers — is what keeps the fight on the first frame.

See also

• Qualitative screening
• AI-assisted screening and human control
• 12 common benchmarking mistakes
• TP audit defense