Sample First-Verdict Report

1 · Criterion, fixed before evidence

The numeric criterion above was agreed in writing — the criterion memo — before any evidence was examined. This ordering is the discipline of the engagement: the target cannot drift toward whatever the evidence later shows. The criterion memo defines “resolved correctly,” the sampling frame, and the exclusion rules, and is attached to the report in a real engagement.

2 · Evidence examined

All evidence was material the client already held. Nothing was generated for this engagement, and no access to the vendor’s system was requested or granted. The inventory (fictional throughout):

• E-01 — Vendor monthly performance reports for the review period [3 documents]
• E-02 — Interaction logs already exported to the client’s quality team [sample of 400 interactions]
• E-03 — The vendor contract and SLA, accuracy representations highlighted
• E-04 — The client’s internal QA scorecards for the same period
• E-05 — Escalation and reopen records from the client’s ticketing system

Each item is cited by identifier wherever it supports a finding, so every statement in the report traces to a specific document.

3 · Method summary

A stratified random sample of AI-resolved inquiries was drawn from E-02 under the sampling frame in the criterion memo. Each sampled interaction was scored against the written resolution definition, using E-04 and E-05 to check whether “resolved” interactions were later reopened or escalated. Vendor-reported figures in E-01 were then compared against the independently scored sample. The method is specified step-by-step in an appendix so that a third party given the same evidence would reach the same numbers.

4 · Findings

All figures fictional; shown to illustrate how findings are stated and attributed.

• F-01. Independently scored resolution accuracy in the sample: 89.4% [fictional] — below the pre-agreed 92.0% criterion.
• F-02. The vendor’s reported figure for the same period was 94.1% [fictional]. The gap is attributable to a definitional difference: the vendor counts an interaction as resolved at close; the criterion memo counts it as resolved only if not reopened within the defined window. Attribution: measurement definition, not model performance.
• F-03. Reopen records (E-05) were complete for the period; no evidence gaps prevented scoring. Attribution notes name a cause — data, model, operator, or handoff — for each discrepancy found.

5 · Verdict

The verdict is stated in one sentence, against one criterion, with the cause named. A real report says “met” or “not met” — it does not grade on a curve, and it does not editorialize.

6 · Reproducibility statement

Given evidence items E-01 through E-05, the criterion memo, and the method appendix, a competent third party can re-run this verification and reach the same result without any input from MarginSignal OS. Same inputs, same method, same answer — that property, not our say-so, is what makes the report evidence.

7 · Limitations

• The verdict speaks to the named deployment, the stated criterion, and the review period — nothing broader.
• Desk-based verification is bounded by the evidence provided; the report lists what was requested and not available, if anything.
• This report is not legal advice and not a compliance opinion; where it supports a regulatory filing, it does so as the factual basis, through counsel where engaged.