Extended partnership K-1s are due September 15. Somewhere between now and then, a CFO at a fund will pull up a K-1, compare the ending capital to the partner's capital account, and find a number that doesn't tie.

The instinct is to re-add the column. Don't. In our experience reconciling K-1 batches, the arithmetic is almost never the problem. The mismatch is structural, and it comes from one of three places.

1. Book basis vs. tax basis

The partner capital account your fund admin maintains is usually a book (GAAP) number. It moves with unrealized gains. The K-1's capital account, since the IRS mandate, is tax basis — unrealized appreciation isn't in it.

If your fund marked up a portfolio company this year, the two numbers should disagree, by roughly the partner's share of the unrealized movement. A K-1 that ties exactly to a GAAP capital account in an up year is the suspicious one.

The check isn't "do they match." It's "does the difference equal the partner's share of unrealized gain/loss, within rounding?" That's a computable tie-out, not a judgment call.

2. A mid-year transfer you booked once

A partner transfers their interest in June. The transferor gets a K-1 for January through the transfer date; the transferee gets one for the rest of the year. The fund's books, meanwhile, often carry the position under a single account that was renamed rather than split.

Now the transferee's K-1 shows a beginning capital of zero — or worse, the transferor's full beginning balance — and neither ties to anything in your ledger. The allocation is right. The identity plumbing is wrong. Every check downstream fails until the transfer is represented as two K-1s against two capital account segments.

3. Bundling: the wrong year, the wrong entity, one PDF

Fund admins and accountants frequently deliver K-1s as bundles — multiple partners, multiple vehicles, sometimes multiple tax years in one file. When a 2024 K-1 gets booked against a 2025 capital account, or an SPV K-1 lands on the main fund, the variance report lights up with differences that look enormous and mean nothing.

The tell: variances that approximate an entire year's activity, or a partner who appears twice. If the variance is huge and round-ish, check the identity of the document before checking any number on it.

The checklist beats the line-by-line review

Every one of these is catchable with a mechanical tie-out:

Ending capital = beginning + contributions + allocated income/(loss) − distributions (the box L rollforward)
Partner allocation ≈ fund total × ownership %, flagged when the ratio drifts
Book-to-tax difference ≈ partner's share of unrealized movement
One partner, one vehicle, one tax year per K-1 — bundles split before anything is compared
Beginning capital = last year's ending capital, unless there's a documented transfer

None of this requires re-reading the K-1 line by line. It requires running the same ties on every K-1 in the batch, every year — which is exactly the kind of work a person does carefully for the first five documents and quickly for the last forty.

That's the job we built the Tax Review agent for: it parses each K-1, runs these reconciliations against the partner capital accounts, and puts the variances — with the reason it suspects — into a review queue. It drafts; you decide. Nothing is filed, sent, or changed without a human approving it.

If your extension-season plan is "we'll eyeball them again this year," run the checklist first. The variances you find in ten minutes are the ones that would otherwise surface in an LP phone call.