Cataloguing an Exhibit Is Not Reading It

2026-09-08 · Philip Choo · AI9OS

Eleven exhibits arrive from a client. Over four days each one is given a reference, described in a register, checked for provenance against the signed intake, graded for use, hashed and placed in the evidence locker.

Every control the firm has is applied. The file looks immaculate.

None of them has been read.

When they finally were, line by line, while mapping files to references for counsel's bundle, three material facts came out at once:

• A second bank account nobody knew existed, named as the destination for profit payments on a deposit slip.
• A utility account in the subject's own name at the property in dispute. The first independent record tying that person to that address, and it was not even quarantined.
• A bill being relied on for current occupancy was six years old.

All three had been in custody since day one, while the matter spent money searching public registries for facts already sitting in the firm's own locker.

Why a well-run file makes this more likely, not less

Cataloguing answers may we use this? Reading answers what does it say? Those are different questions, and on that file only the first had ever been asked.

The register entry becomes the exhibit. Term deposit slip, bank, amount is a true and useful description, and it quietly stands in for the document, which is never opened again. Because the cataloguing was rigorous, the file feels thoroughly worked. Nothing prompts anyone to go back, and the better the process, the stronger that false sense of completion.

The bias is worse for anything quarantined. A held exhibit gets filed mentally as cannot use, which slides into need not read. Those are not the same. A quarantined document's contents still tell you what to ask for, as long as the request is worded so it does not reveal where the question came from.

Make the read a separate step with its own output

Not a habit, and not part of cataloguing. A scheduled step that produces its own artefact.

For each exhibit, transcribe every field: account numbers, reference numbers, dates, addresses, names, amounts, effective and expiry dates, payment destinations, meter readings, issuing office, and the document's own period or edition date.

Then ask three questions of the transcription:

1. What is here that is on no list we hold? Second accounts, referenced third parties, destination accounts, guarantors, agents. This is where the unknown account came from.
2. What date does this document actually speak to? A bill has a billing period. A report has an edition. A register has an as at. A document is evidence of its own date, not of today, and the six-year-old bill was being read as current.
3. Whose name is on it, and in what capacity? Account holder, ratepayer, addressee and owner are four different things, and they are routinely collapsed into one.

Where a read produces a lead from a quarantined exhibit, draft the resulting request generically, so it captures the target without naming it, and record explicitly that the detail must not be used until it has been disclosed properly.

The family this belongs to

It is the same failure as a summary that drops the fields before the analysis runs, and the same as trusting a filename to tell you what a file contains. A description of a thing is being trusted in place of the thing.

The description is usually accurate. That is what makes it work.

If you are the one paying for the work

"Which exhibits have been read line by line, and where is that transcription?"

If the answer describes the exhibit register, you have been told what the firm may use. You have not been told what it says.