How an investor watches a book it does not own

An investor never sees the book. Between audits it receives four signed documents on a contractual cadence, and its craft is making them checkable: shorten the interval, demand the raw records behind each figure, and tie consequences to arithmetic. One filed agreement requires a borrowing base report, that is how much may be drawn, monthly within 7 days.

An investor never sees the book, it sees a stream of assertions about the book

An investor watching a portfolio it does not own has no connection into the lender's system of record, no authority to change a limit, and no way to open a credit file. What arrives instead is a sequence of documents in which the other party states what happened. The Financial Stability Board took up the same asymmetry for the asset class as a whole in its report on vulnerabilities in private credit of 6 May 2026.

That is not a smaller version of the problem a bank has with its own book. A bank's problem is deciding what to do about what it can see. An observer's problem is that seeing is not available at all, so every technique it has is a technique for making somebody else's statement verifiable.

The software a search returns for this question is built for the other side of the table

Look for portfolio monitoring and the results are vendor pages for systems a bank runs over its own exposures: automatic collection of borrower financials, covenant tracking, coverage ratio thresholds, a dashboard sorting loans into at risk, moderate and healthy, early warning signals. Every one of those functions assumes the operator can reach into the loan system, change something, and telephone the borrower. An outside investor can do none of the three, so none of it transfers.

The audit is annual and the book is not, so the interval is the whole problem

On the anonymised book we read in July 2026, the average tenor was about seven months and the average portfolio grew from MXN 34.3 million to MXN 82.0 million inside a single year, per our own published credit memo. Those two facts together mean the book turned over more than once and roughly doubled between two audited statements. An annual figure does not describe that portfolio. It describes two different portfolios averaged into one number.

Nothing there is a criticism of a particular lender. An annual audit is the standard revaluation cycle for a private asset, and the mismatch is structural: the observation cycle was designed for things that change slowly, and a consumer or small business loan book does not.

Three things move inside the interval and none of them shows up in a balance sheet. Mix moves, because new origination at a different ticket, term or product changes the risk of the book without changing its size. Migration moves, because loans travel between arrears buckets continuously and a month-end snapshot hides the direction. And funding callability moves: on the same anonymised book, about 30 individual lenders held 64.2 per cent of the debt with no stated maturity, per our own published credit memo, which is a run risk the balance sheet cannot express because it records the amount and not the notice period.

Between audits the investor receives four artefacts, and every one of them is a signed claim

The borrowing base certificate carries most of the weight, that is the running arithmetic of how much may be drawn against eligible collateral today. A filed warehouse agreement defines it as «the certificate from the Servicer, executed and delivered by the Servicer, setting forth the calculation of the Class A Borrowing Base and Class B Borrowing Base ... and certifying as to the accuracy of such calculations».

Beside it sit the loan level file itself, the compliance certificate in which the covenant arithmetic is certified, and the financial statements. A published summary of what a warehouse actually demands lists them together: «1. Collateral Feed: Receivable file uploads required to Agent and Analytics Provider with each borrowing request. 2. Settlement Reporting: Variety of reports required monthly including settlement calculations, certificates, borrowing base updates and prepayment summaries prior to cash movements. 3. Borrowing Base Certificate: Primary collateral report with eligibility detail supporting borrowing availability. 4. Financial Statements: Unaudited annual/quarterly reports of Borrower, and consolidated audited financials for Parent/Servicer. 5. Compliance Certificate: Certification of financial covenant calculations.»

The cadence is contractual and faster than most lenders expect. One credit agreement filed with the SEC requires monthly financial statements with a compliance certificate within 30 days, a borrowing base report with every advance request and monthly within 7 days, and weekly reporting no later than Friday of each week at the lender's option.

In Mexican structured transactions the reporting party is an independent master servicer rather than the originator. Two 2025 transactions rated by HR Ratings report weekly and monthly, a third measures its collateral coverage on weekly cut dates, and late delivery of a master servicer report uncured for 10 business days is itself an accelerated amortisation event.

Recomputing beats reading, because a calculated field can only be believed

Receiving a report and checking it are different activities. Tape cracking is the market name for the second one, and the firm that does it at scale describes itself as one that «cleans and validates data tapes by recalculating values from raw inputs». The reason the step exists is that a calculated field carries no evidence of its own derivation, so it cannot be checked, only accepted.

The anonymised package we read shows what recomputation catches. A stated default rate of 0.02 per cent sat beside 1.41 and 1.34 per cent in the same workbook, the weighted average yield was a hardcoded formula on a base that did not equal the portfolio, and 37 of 42 funding facilities recorded collateral as not applicable, per our own published credit memo. Each of those is invisible in a summary and obvious in the raw columns.

The practical form of this rule is a delivery obligation rather than a piece of software. A loan tape, that is the loan level file with one row per loan at a stated cut-off date, has to arrive on the same cadence as the certificate that summarises it. A certificate without its underlying file is a number the observer has agreed in advance to take on trust.

A trigger is arithmetic so that nobody has to form a judgement under pressure

One Mexican transaction accelerates if minimum collateral coverage of 1.20 times is missed on three consecutive weekly cut dates or for 21 consecutive days, per the HR Ratings presale document, and a United States auto securitisation sets its delinquency trigger at four times the previous historical peak delinquency observed, per the prospectus filed with the SEC. Nobody votes and nobody has to be brave on the day.

Both designs put the burden in the same place. A trigger computed from the observed party's own history requires that history to exist and to be clean, so the observer's leverage between audits is built out of the other side's data quality rather than out of contractual language. A lender with no cohort series does not get a lenient covenant, it gets an arbitrary one.

Somebody has to open the credit files, and it is never the investor

File audit is delegated, and the sampling rule is written into the deal rather than left to judgement. In one 2025 transaction it is the lesser of 100 random files or a statistically significant sample at 95 per cent confidence and 5 per cent margin of error, within 20 business days, per HR Ratings. In another it is 100 per cent of the credit files reviewed twice a year, per Fitch, with the first review covering every file physically.

Continuity is the same question asked about the operator rather than the loans. S&P Global Ratings maintains an operational risk framework for structured finance servicers, republished on 28 October 2025, and the practical form of it is a question with a number in the answer: how many weeks would a replacement servicer need before borrowers noticed, and that number is a function of how often the loan level file was being delivered while everything was still fine.

A late report is treated as a credit event, because from outside it cannot be told apart from one

The document is the only surface, so the arrival of the document is itself information. An investor who cannot walk into the office and look has no way to separate a reporting failure from a portfolio failure, and pricing the ambiguous case at the better of the two readings is how an observer gets surprised. That asymmetry is not pessimism. It is the same rule that makes an undeclared blank cell in a data file get priced as the worst thing it could mean.

Which is why the whole observer toolkit reduces to three moves, none of which requires access to anybody's systems. Shorten the interval between assertions. Require the raw records each assertion was computed from, so it can be recomputed instead of believed. Attach the consequences to arithmetic rather than to a conversation. All three are agreed before the money moves, because afterwards there is nothing left to trade for them.

Effect is not a bank, a lender, a broker-dealer, an investment adviser or a credit rating agency. Nothing here is investment, legal or tax advice, and no outcome is promised. Contract levels are quoted from filed agreements and rated transactions named in the sources below. The anonymised case is our own reading of a package a lender supplied to us, at a 31 May 2026 cut-off.