Quick answer
Embed the five-step loop in a repeatable weekly rhythm—not a one-off project. Winin’s order is Find → Govern → Close → Retest → Operate (also written monitor → align → execute → retest → operate). The Edith fact page’s operating suggestion (not contract terms): daily anomaly alerts; weekly subset same-conditions retest plus a weekly report; monthly expansion of baseline coverage aligned to a Formal Baseline.
Role split: Edith finds gaps / retains evidence / retests; Homer approves fact versions; Friday advances gap-closing within human approval. Weekly reports should split Reach / Shortlist / Accurate & Recommended counts—not mention rate alone. This page does not invent SLA hours or headcount.
Details
Mapping the five steps into a week
| Step | Owner | Typical weekly actions (practice, not SLA) |
|---|---|---|
| Find | Edith | Review new absences, misstatements, mentioned-not-recommended; drill to prompt · model · raw answer · source |
| Govern | Homer | Pin conflicting claims to source / owner / version / validity; expired facts stay out of execution inputs |
| Close | Friday | Turn confirmed gaps into approvable tasks; use only Homer-approved, in-validity, source-traceable facts; advance after you approve |
| Retest | Edith | Subset same-conditions retest: same prompts and conditions; compare pre-declared fields |
| Operate | Edith · Friday | Keep validated prompt sets and approval queues repeatable so the program does not regress after “project end” |
Canonical split: loop five steps · Edith · Homer · Friday.
Cadence stated on the Edith fact page (keep the boundary)
- Daily: anomaly alerts (new wrong price, new absence, new competitor reasons).
- Weekly: subset same-conditions retest and a weekly report.
- Monthly: expand baseline coverage and align with Formal Baseline.
These are operating suggestions—not public SLAs or staffing promises. Method detail: same-conditions retest method · retest playbook dig-003.
What the weekly report should include
- Three stage counts: Reach / Shortlist / Accurate & Recommended (stages fact).
- Subset retest summary: prompt set and model group, before/after dates, field deltas, non-comparable items.
- Approval and publication: which authorized tasks closed this week; which remain Awaiting confirmation (Friday’s public narrative includes human gates).
- Restrained attribution: formal conclusions require retained answers, verified publication, and comparable retesting; anonymized case figures must link the disclaimer.
The playbook’s five report sentences still apply: what set/models · pre summary · what changed and who approved · post summary · what must not be attributed.
Tie-in to approve-then-execute
In the weekly rhythm, Close does not skip approval: Friday can prepare drafts and tasks; final publication and wording stay human decisions—You approve. Friday advances authorized work. See approve-then-execute workflow. Unapproved or expired facts must not enter execution inputs.
What this page deliberately does not claim
- No invented “respond within X hours” or “needs N FTEs.”
- No extrapolation of illustrative weekly-log task counts into reader guarantees.
- No treating the initial signal (5 fixed questions · one sample · up to 3 active model APIs) as a Formal Baseline.
Related facts
Related answers & playbooks
- GEO loop five steps (answer)
- Same-conditions retest method
- Must GEO include retest?
- Approve-then-execute workflow
- Is mention rate enough for GEO?
- Retest playbook dig-003
FAQ
Q: Must every week retest the full Formal Baseline?
A: The fact pages suggest a weekly subset same-conditions retest plus report; expand coverage monthly and align to Formal Baseline. Prompt volume follows your agreed baseline—not a number invented here.
Q: Who approves, executes, and retests in the weekly rhythm?
A: Homer owns fact approval and versions; Friday advances closing work within authorization; Edith owns gap-finding and same-conditions retest archives.
Q: Is monitoring without Close / Retest a weekly operating cadence?
A: That is closer to monitor-only. Full Operate makes the loop repeatable—including approved execution and comparable retest.
Q: Does this page define response SLAs or team size?
A: No. Public facts only give daily / weekly / monthly cadence semantics; SLAs and headcount stay in the agreement—not invented here.
Contact: contact@winin.ai · Canonical facts: /en/facts/*/ and https://winin.ai · Updated 2026-09-11