Evidence and inference
Evidence is something Connect observed and can point at: a message that arrived, a page that was read, a value in a record. An inference is a conclusion drawn from evidence. Connect is allowed to infer, and is not allowed to present an inference as an observation — which is why a prospect has a source you can open and why an email address is never invented to fill a gap.
The distinction, and why it is not pedantry#
| Evidence | Inference | |
|---|---|---|
| What it is | Something observed, with a source | Something concluded from what was observed |
| Can be shown | Yes — the message, the page, the field | Only its reasoning and its inputs |
| Can be wrong | By being stale or misread | By being a reasonable conclusion from true facts |
| What it licenses | Acting on the fact itself | Deciding what to do next, and asking |
| How it should be worded | Plainly, with the source named | As a judgement, with the uncertainty kept |
The reason to hold them apart is that they fail differently. Bad evidence is usually visible — the address bounces, the page is gone. A bad inference looks exactly like a good one, because both are fluent sentences built from true inputs. A system that blurs the two makes its confident mistakes indistinguishable from its correct answers.
Prospecting, where the line is hardest#
Research is inference-shaped work: you are trying to establish who might want something, from public signals that were not written for you. Connect keeps the line by recording what it found alongside what it concluded, so a qualification is always traceable to something you can read.
- A discovered organisation carries its evidence — what was found, and where. A prospect with no source is not a prospect worth working.
- A qualification score is an inference over that evidence, and is presented as a judgement rather than as a fact about the business.
- An email address is evidence or it is nothing. Connect does not guess one from a name and a domain, because a guessed address is an inference wearing the clothes of a fact.
- A contact who cannot be reached honestly stays unreachable. That is a smaller outcome than a plausible address, and a truthful one.
Voice, where the line is easiest to see#
A call produces both kinds of thing at once, which makes it the clearest worked example in the product. The transcript is evidence: it is what was said. The summary is inference: it is what somebody concluded the call was about. The outcome is a stronger inference still, because it is a judgement about what the call achieved.
- The transcript
- Evidence. Ambiguous in places, incomplete where audio was poor, but a record of words rather than a reading of them.
- The summary
- Inference. Useful precisely because it compresses, and wrong in the way compressions are wrong — by dropping the qualification somebody added at the end.
- The outcome
- Inference with consequences. It is the thing a pipeline or a follow-up may act on, so it deserves a glance at the transcript before it does.
- A follow-up promise
- Evidence when the caller said it, inference when it was understood. This is why a commitment is recorded with the reason it exists.
The practical rule that falls out of this: when a decision matters, read the evidence rather than the inference built on it. A record and a transcript takes that further.
What this means for an answer you are given#
Notice whether the answer names a source.
Result A cited answer is checkable in seconds; an uncited one is a conclusion you are being asked to accept.
Treat *not sure* as a real answer rather than a failure.
Result An assistant that says it does not know is telling you where the evidence stopped, which is information.
When two sources disagree, expect to be told so rather than shown a blend.
Result A merged answer hides the disagreement, which is usually the most useful thing in the room.
Before acting on a judgement, look at one piece of the evidence under it.
Result One check is cheap. Discovering after the fact that a chain of reasonable inferences started from a misread field is not.
None of this makes inference suspect. A business runs on judgement, and an assistant that only ever repeated observations would be a search box. The requirement is that the two are labelled, so you know which one you are reading.
Questions#
Is a summary evidence of what was said?
No, and the difference matters when somebody disputes a call. A summary is a reading of the conversation; the transcript is the record of it. Quote the transcript when the wording is the point, and use the summary for what it is good at — remembering quickly what a call was about.
Why will Connect not guess an email address?
Because a guessed address is a claim about a person that nothing supports. It also produces the worst kind of failure: a message that reaches somebody it was never meant for, at an organisation you were trying to impress. A prospect without a verified address is honest about what is known.
Can I tell Connect something as a fact?
Yes. A memory or a standing instruction is evidence of what you said, and it is treated as such — durable, attributable and editable. What it is not is a conclusion Connect reached, which is why corrections you make behave differently from conclusions it drew.