Researching one prospect, end to end

One candidate, end to end#

1. Trigger — a candidate from a discovery run, an imported row, or a person asking for one organisation to be researched.
2. User event — for a manual request, the person's instruction is itself the reason and is recorded as one.
3. Authentication and workspace resolution — the candidate already belongs to a workspace; research inherits it and cannot read outside it.
4. Ingest — the candidate's name and location are resolved into the sources worth reading.
5. Canonical record — the prospect row exists before the reading starts, so a partial pass leaves a partial record rather than nothing.
6. Research and reasoning — sources are read and claims extracted; prospect_evidence stores each claim with its source, and refuses the claim if there is none.
7. Knowledge, memory and rules — the workspace's knowledge decides what a good fit means here; standing instructions and blocks are honoured at this point, not after.
8. Autonomy and approval — prospect_cost_router decides whether a deeper pass could change a decision; the AI budget can refuse further paid work outright.
9. Action through a provider — public-source fetches and model calls. Nothing is sent to the organisation being researched.
10. Result — claims with evidence, a score from prospect_scoring, a written fit reason, and a contactability state.
11. Relationship and memory — if an address resolves to a Person you already have, the prospect becomes that relationship instead.
12. Audit and usage — the spend is booked, and a refusal (budget, identity, no sources) is recorded as the decision it is.

How a claim earns its place#

The rule at the bottom of that table is the one that shapes the whole flow. Because a pattern-generated address is never produced, the research pass has no fallback and no reason to pretend. Either a source showed the address or the record honestly says there is none.

Depth is decided, not assumed#

1. The first pass reads the obvious sources and extracts what they say.

   Result Most prospects are decided here — clearly in, clearly out, or clearly unreachable.

2. The router asks whether more reading could move this particular prospect across a line that matters.

   Result Where it could not, the pass stops and the prospect keeps the evidence it has.

3. Where it could, a deeper pass runs and its claims join the record with their own sources.

   Result The score and the fit reason are recalculated, and the change is visible rather than silent.

What failure looks like#

Very few claims

The organisation has a small public footprint. The record is thin and honest; the score reflects the evidence rather than the intuition.

Claims that contradict each other

Both are kept with their sources, and the contradiction is visible. Resolving it is a person's judgement, not a coin toss.

Research stopped early

The budget refused, or the router judged further work pointless. The two look different in the decision trail.

The prospect vanished

It resolved to an existing relationship. Look for it under people or companies rather than under prospects.

Questions#

Related#

• Researching a candidate
• Evidence on a prospect
• Deep research on a prospect
• Capturing evidence for a prospect, end to end
• Connect does not guess email addresses
• Discovering prospects, end to end