Recruitment / Field notes
Read the evidence before ranking the person
A recruiter note can contain the right words for the wrong reason. Someone may have discussed a role with budget responsibility without ever holding it. Put that phrase into a candidate summary and a possibility can quietly become a qualification.
The research sheet below follows one fictional note through that problem. The brief asks for cross-functional leadership and budget ownership. The note supports only part of the requirement. A useful shortlist needs to preserve that gap all the way to the person deciding what to do next.
The word “and” creates two questions
Cross-functional leadership and budget ownership are separate claims. A source that supports the first does not settle the second. Before researching candidates, break the brief into requirements that someone can check against a record. Keep a hard requirement distinct from a preference or a suggestion about where to look.
In the recruitment workflow behind this note, the system extracts criteria from the brief and finds comparable earlier searches. An optional plan-review step, off by default, lets a recruiter edit both before deeper screening begins.
If budget ownership is a preference, the recruiter can say so while reviewing the plan. If it is mandatory, the missing evidence remains a reason to investigate. Quietly relaxing the requirement later would change the search without giving the recruiter that choice.
Illustrative example · fictional research note
One requirement, two claims
Select a requirement to trace its support through a fictional note, then make a separate review decision.
Led cross-functional work and owned its budget.
In the previous role, led weekly delivery reviews across engineering and operations.
We discussed a programme director opening with budget responsibility.
The second sentence describes the opening. It does not describe a role already held.
Trace the evidence
The brief asks for both leadership and budget ownership. The first has support; the second still needs evidence.
This finding stays attached to the record.
No recruiter decision yet. The generated assessment is partial.
Read the sentence around the match
The implementation combines structured records and keyword retrieval across work history, prior searches, notes, and resumes. A matching phrase gets a record into the research pool. The assessment still needs to establish who the passage describes, when it applies, and whether it describes work performed or an opportunity discussed.
Our fictional note makes that distinction visible. “Led weekly delivery reviews” describes previous work. “An opening with budget responsibility” describes a proposed role. Both passages may be useful to a recruiter, but they support different conclusions. A summary that merges them would give the person experience the source never established.
Give unresolved evidence a place in the result
The longlist artifact uses three criterion states: met, partial, and not evidenced. “Not evidenced” describes the research available to the reviewer. It leaves room for a missing document, an incomplete note, or a question that needs to be asked directly.
Citations make those states inspectable. The longlist drops a citation when its quoted text cannot be found in the source. It changes a “met” finding to “not evidenced” when no citation survives for that criterion. A second question remains for the reader: does the exact passage justify this particular conclusion? Correct quotation and sound interpretation need separate attention. The chat interface works differently. Its final checks run after the answer is already shown, so they raise quality flags and do not block every inaccurate statement.
Coverage belongs nearby. The artifact records limits such as unread material and bounded research scope, and it can return a shortfall explanation when the list is smaller than requested. A recruiter can then choose between more reading, a broader search, or a change to the brief. Filling an arbitrary number of slots would conceal those choices.
Keep the decision next to the assessment
A recruiter may retain a promising person while asking for evidence of budget ownership. That decision should leave the unresolved criterion visible. In the implemented workflow, accepted, rejected, and deferred decisions are stored separately from generated assessments, with a place for the reason.
The example lets you make that distinction yourself. Keeping the person for review records a decision, while the assessment stays partial. Deferring the decision records the missing evidence as the next task. Neither button rewrites the note. When a colleague picks up the search, they need to understand both what the source established and what the recruiter chose.