Provenance
Provenance is the record of where a piece of information came from, kept alongside the information itself. It matters most at the moment somebody edits the statement: without provenance the edit produces a claim with no origin, indistinguishable from something the system invented, and nobody downstream can tell which they are reading.
Origin is a property of the statement, not of the import#
The common mistake is to treat provenance as a loading detail — a filename kept on the batch, a timestamp on the sync. That survives exactly until someone corrects one row, at which point the corrected row is attributed to a document that no longer says what the row says.
Provenance that works is attached to the smallest unit anybody edits, and it changes when the statement changes. A corrected fact is a fact whose origin is now the person who corrected it, with the earlier origin preserved as history rather than overwritten. That is a different data shape from a field called source that is set once at import.
- Origin
- Where this statement came from: a document, a conversation, a person, a research pass
- Basis
- The specific passage, message or observation, not just the container
- Time
- When the origin asserted it, which is not the same as when it was recorded
- Supersession
- What this replaced, so a correction reads as a correction rather than as a contradiction
Why an unsourced claim is worse than a missing one#
In a system that drafts messages, an unsourced statement is not neutral. It is available to be repeated to a customer with the same confidence as a sourced one, and the reader has no way to weigh it. A missing fact produces a question; a fabricated one produces a commitment.
That asymmetry is why the rule in prospecting is stated as a hard gate: evidence is attached to claims, and a claim without evidence does not reach the record. It costs coverage — some prospects carry fewer attributes than a system willing to guess would show — and the trade is made deliberately in that direction.
Does Connect use provenance?#
Yes, in four places, each with its own shape.
| Record | What the origin is | What it is used for |
|---|---|---|
| A Fact | The Knowledge source it was grounded in | Answering only from what a workspace supplied, and showing why an answer says what it does |
| A Memory | The conversation, call or person it came from, at one of four tiers | Letting a person read what Connect knows and forget any of it |
| A prospect claim | The public source and the captured evidence | Refusing a claim that has none, and re-researching when it goes stale |
| A file version | Who or what changed it, and the audit entry for the change | Reading a document's history without leaving the workspace |
The memory viewer makes this concrete for a person rather than for a developer: what Connect knows is readable at every tier, from the mailbox row, the channel screen or the contact panel, and anything there can be forgotten. A memory whose origin cannot be shown is a memory nobody can sensibly decide about.
Provenance for actions, not just for facts#
The same idea applies to things Connect did. A message is reported as sent only when the provider has acknowledged it, and the acknowledgement is the evidence. There is a third state — uncertain — shown as uncertain rather than guessed in either direction, because re-sending on a maybe is how a customer receives the same reply twice.
Read that as provenance rather than as error handling: the state of the send is derived from what the provider actually said, and where the provider said nothing the record admits it instead of inventing an origin for a claim of success.
Questions#
If I correct something Connect learned, is the original lost?
No. A correction supersedes rather than erases: the earlier statement stays as history with its own origin, and the current statement carries yours. That is what lets somebody later see that a value changed and on whose authority, rather than finding a field that has always said this.
Why does a prospect record sometimes have fewer details than expected?
Because a claim with no evidence behind it does not reach the record. Research that found nothing usable produces a gap rather than a plausible guess. The same rule is why no email address is ever pattern-generated: a prospect with no discoverable address stays a researched prospect with a different next action.
Does provenance slow anything down?
It adds writing, not reading — the origin is stored with the statement, so showing it costs nothing extra at read time. What it does cost is coverage, in the places where the honest answer is that nothing supports the claim.