# Control 1

- Host: Microsoft 365 Copilot Cowork
- Model shown by host: Claude Opus 4.8
- Condition: custom Marketing Claims Review skill absent
- Attachment: `campaign-brief.md` only
- Prompt: `evaluation/chat-only-prompt.md`
- Run ID: `control-1`
- Run started: 2026-08-04 11:13 (Cowork UI)
- Artifact creation: prohibited by prompt; none created

## Raw response

**Scope & basis (supplied).** Advisory only; not legal advice. Copilot cannot approve, publish, pause, correct, or withdraw. This covers only Brief MC-0204. The only performance evidence referenced anywhere in the brief is the pilot described in the evidence-bounded option: [1](coworker-citation://1) a Lumena-sponsored 8-week pilot of 120 self-selected eligible US homes across four climate regions, with a median reduction in HVAC electricity use of 12% versus a weather-adjusted prior-year baseline, and the brief states the pilot was company-sponsored and has no independent replication.

**Recommendation (judgement).**

- **Decision class:** Revise before release — do not release the original as presented.
- **Campaign option:** `evidence-bounded-campaign`.

**1–2. Claim register + evidence fit (supplied claims; fit is judgement).**

| # | Channel | Express claim | Implied/material | Evidence fit |
|---|---|---|---|---|
| 1 | Hero, paid social | Save 30% on home energy bills | Typical result; whole-bill (not just HVAC) | Not supported — evidence shows 12% median HVAC-electricity, not 30% of the bill |
| 2 | Landing page | Pays for itself in six months | ROI/payback; price & savings math | Not supported — no cost/savings/price data supplied |
| 3 | Display, product page | Works in every home | Universal compatibility | Not supported — contradicted by supported-HVAC, Wi-Fi, installation-check requirements and named exclusions |
| 4 | Video, landing page | Independent pilot proves the savings | Independence; causation ("proves") | Not supported — pilot was company-sponsored with no independent replication |
| 5 | Ava Reed influencer post | My energy bill dropped 35% in eight weeks with Lumena Sense | Typical result; material-connection disclosure | Not supported — exceeds 12% median; only disclosure is #LumenaPartner after a collapsed "more" boundary, with no on-image or in-video disclosure |
| 6 | Landing-page footnote | Results vary. Based on an 8-week pilot | Disclaimer placement/proximity | Partial — footnote sits at page bottom, remote from the claims it qualifies |

**3. Required edits/disclosures verifiable before release (supplied via evidence-bounded option).**

- Replace the 30% claim with the exact pilot-bounded wording (12% median HVAC-electricity, 120 self-selected homes, four regions, weather-adjusted baseline).
- Remove the six-month payback and every-home claims.
- State that the pilot was company-sponsored and has no independent replication (drop "independent" / "proves").
- Show supported-HVAC, Wi-Fi, installation-check requirements and named exclusions close to the call to action.
- Remove Ava Reed's performance endorsement because no typical-results analysis is supplied.
- Place "Paid partnership with Lumena; device provided" clearly at the start and in the image or video content.
- Maintain an approved claim register and verify final creative matches it.

Each edit is verifiable by inspecting final creative against the register before release.

**4. Missing/conflicting information & required reviewers (judgement — gaps flagged, nothing invented).**

- Missing: any price/cost basis for payback; any typical-results (substantiation) analysis for the 30%/35% figures; the compatibility and exclusion list; any independent verification.
- Conflicting: "works in every home" vs. the stated eligibility limits; "independent" vs. company-sponsored; 35% testimonial vs. 12% median.
- Reviewers: the brief names only the Lumena Marketing Director as campaign launch owner. It does **not** name legal, regulatory, or substantiation reviewers — I cannot supply that authority. Route claim substantiation, disclosure adequacy, and the influencer post to qualified human legal/regulatory review before release.

**5. Release controls, monitoring, timing, triggers.**

- Supplied controls: the approved claim register and final-creative match check above; requested launch date 2026-10-01.
- Not supplied — must be set by human owners, not by me: monitoring ownership, review cadence, and correction/stop-trigger thresholds. The brief only defines the fallback: no-launch means no further release until required human reviewers determine correction, pause, or withdrawal actions. If any live asset is later found off-register or a claim loses its evidence basis, escalate to those human reviewers for correction/pause/withdrawal.

**Bottom line (judgement):** move to `evidence-bounded-campaign`; the original's core claims lack supplied substantiation.