# Repeated-brief evaluation proposal

Status: proposed, no customer or savings result. Revised 3 October 2026.

Intended buyer: research director/partner at a small geopolitical-risk or policy consultancy. Test a repeated briefing job with an independent domain reviewer, equivalent sources and at least two assignments and two analysts.

Compare direct documents, capable chat/retrieval and a pinned structured packet. Keep source access, assignment/audience, reviewer standard and total budget comparable; record baseline strengths and failures fairly.

Record source/citation entailment, omitted caveats, wrong actor/date/modality, source-finding time, representation-switching cost, setup/elicitation time, licensing, maintenance, reviewer time, repeat usage, and change-impact false positives/missed dependencies.

| Measure | First task | Second task | Second analyst | Evidence |
|---|---|---|---|---|
| Source/citation entailment | | | | |
| Missing caveats / wrong actor, date, modality | | | | |
| Source-finding / representation-switching time | | | | |
| Elicitation, setup and licensing effort | | | | |
| Maintenance and reviewer effort | | | | |
| Repeat use and handover completeness | | | | |
| Impact false positives / missed dependencies | | | | |
| Net effort including all preparation/review/upkeep | | | | |

Leave unmeasured outcomes blank. Generation speed is not net savings. The hypothesis weakens if preparation outweighs repeat benefits, review remains equally burdensome, imported expertise creates unsupported confidence, caveats vanish or users abandon the capsule. Synthetic demo polish, successful runtime operations and traceable answers do not establish customer value.

Before recruitment, define permitted non-sensitive source access, consent, independent reviewer criteria, task equivalence and failure reporting. This website task sends no invitations and creates no external evaluation state.