A partner at a mid-sized fund called me last spring. He had been staring at a figure in a Trace report for twenty minutes, trying to reconcile it against memory. The figure was right; his memory was wrong. But he could not prove it, because the report did not show him where the number came from. That conversation changed how we ship.
The Trace team now builds to a single rule, applied without exception: every figure we show is traceable to its source. Every source is dated. Every derivation is recoverable. When a client asks where a number came from, the answer is a link, a timestamp, and a name — not a shrug, not a re-computation, and not a conversation that requires two analysts and a spreadsheet.
What "a witness" means
In a courtroom, a witness does not assert something and ask to be trusted. A witness identifies themselves, describes where they were, and tells you what they saw. If the account needs to be verified, there is a clear path to verification. Data, in our view, should behave the same way.
A Trace figure is attached to its source record, the timestamp at which that record was ingested, the pipeline version that processed it, and the analyst who approved the method. Any figure can be expanded into its provenance, in the tool itself, in two clicks. The client does not have to email us to find out where a number came from. The number tells them.
What we do not do
We do not synthesize. When a data point is missing, the report says it is missing. We do not interpolate, smooth, or round to flatter a narrative the client seems to want. If the underlying series has a gap, the chart has a gap. The client can decide how to treat it; we will not decide for them.
We do not quietly correct the historical record. When we discover that a past figure was wrong — it happens, it will happen again — we publish the correction, dated, with a note. Clients who cite the old figure should know they were citing something we no longer stand behind. They get an email from a person, not a silent overwrite in the database.
Seven years, minimum
We retain every source file, every pipeline version, and every report output for at least seven years. This is longer than any of our clients have contracted us for, and deliberately so. If a client needs to reconstruct a decision they made three years ago, the materials are there, in the state they were in when the decision was made.
Seven years is not a number we picked because it sounded defensible. It is the horizon over which most of our clients' decisions are evaluated — by their investors, by their boards, by their own second-guessing. It is also the horizon at which memory stops being reliable, even for the people who were in the room. A traceable archive is, in effect, a second memory, and it does not embellish.
An inconvenient standard
This standard is inconvenient. It is more expensive to build against, it is slower to ship, and it requires a culture of admitting error that not every firm can sustain. It also produces a product that, in the long run, is the only kind of data product worth paying for: one that can be cited, defended, and built upon.
Our clients price deals on the figures we show them. They argue cases. They take positions. They deserve to know that when they say, out loud, "according to Trace," the next question — "and where does Trace get that?" — has a real answer.
— Anya