Field Notes
Notes from the slow work.
Shorter than the Library and less finished, these are the things we say to one another between visits. We set them down here in case they are useful to someone tending a system of their own.
The first thing we tell a new associate is that speed is rarely the constraint. A system that must stand for thirty years is not improved by being finished in three weeks rather than five. The clients who stay with us have learned this too; they no longer ask when it will be done, but whether it will hold.
Patience is not slowness for its own sake. It is the refusal to ship a decision we have not yet understood. We would rather sit with a problem an extra day than spend a decade explaining the shortcut we took.
A figure on its own is a rumour. We have spent years teaching systems to keep, beside every number, the small history of where it came from, who entered it, from what source, and when. The cost is a little more storage. The reward is that no record ever has to be taken on faith.
When an examiner asks a hard question, we do not reconstruct the answer from memory. We ask the system, and the system, having remembered, replies.
This is the unglamorous heart of document work. The visible part, the form, the statement, the report, is the smaller half. The larger half is the quiet ledger underneath, recording how each visible thing came to be true.
We have never regretted storing too much provenance. We have, more than once, regretted storing too little.
Every system must eventually move, to new hardware, a new format, a new rule from the state. The temptation is to switch over on a chosen night and hope. We do not hope. We run the old system and the new one together, in parallel, feeding both the same work until their figures agree to the row.
Only when a full season has passed without a single disagreement do we quietly retire the old. The best migrations are the ones no one outside the firm ever notices.
A surprising amount of our value is simply that a partner picks up. When the auditor calls, or the board, or a successor who inherited a system they do not yet understand, they reach a person who can read the work and explain it in plain language.
We consider reachability a feature, and we maintain it as deliberately as any line of code. A system that cannot be explained is not finished, however well it runs. The explaining is part of the building.