# Editor notes — LumenDesk 2.0 article

All entities are fictional test fixtures. The article uses only the files in `inputs/`. No browsing and no hands-on use of the product took place (editorial-brief.md, "Assignment"), and the article says so in its second paragraph.

## Decisions for the editor

1. **Is the assistant included in the Team price?** The packet never says so outright. S1 says it is available to existing Team customers, and S2 lists only seats, extra uploads and OCR. The article says "the pricing page lists no separate fee for the assistant beyond extra uploads and OCR" and "a trial should cost you mainly time". If you want no inference at all, cut those two sentences.
2. **"Outside review" for DeskLab.** The filename calls S4 an independent review, but the text does not state DeskLab's relationship to the vendor. I called it an "outside review" and did not use the word "independent". Soften this if you want a stricter reading.
3. **No publish date is set.** "Available today" in S1 means October 16, 2026, so the article anchors access to "the vendor's October 16, 2026 announcement". The newest source is dated October 18, 2026.

## Corrections reconciled

| Issue | Resolution | Source |
|---|---|---|
| Launch screenshot showed USD 19 per seat | The article uses USD 29 and tells readers the USD 19 figure was wrong. The screenshot itself is not in the packet; it is known only from the correction. | S2-pricing.md, "Correction" |
| Annual pricing | The article says the packet lists no annual offer. | S2-pricing.md, "Correction" |
| Upload allowance | 500 uploads are per workspace, not per seat, with no rollover. | S2-pricing.md, "Current plans" |
| Available now versus planned | Available: browser app for existing Team customers in Canada and the US. Waitlist: new customers, with no date given. Planned for Q1 2027, with no exact day and no guarantee: API and EU availability. | S1-announcement.md, "Access" |

## Claims kept out of the article

All six come from S5-internal-notes.txt, which describes itself as unverified leads. None appears in the article or metadata.

| Claim in S5 | Why it is excluded |
|---|---|
| "Eliminates hallucinations" | Nothing in the packet supports it. DeskLab found one wrong page citation in six documents (S4-independent-review.md, "Method"), and the vendor says the tool drafts answers and that its test excludes human verification (S1-announcement.md, "Features" and "Vendor test"). |
| "GDPR certified" | Nothing in the packet supports it. The policy "does not claim … guaranteed regulatory compliance" (S3-data-policy.md, "Processing"), and EU availability is only planned (S1-announcement.md, "Access"). |
| "Runs fully offline" | Contradicted. An external model provider processes requests and there is no on-device mode (S3-data-policy.md, "Processing"); the product is a browser app only (S1-announcement.md, "Access"). |
| "Saves every team 60% of their work time" | Unsupported. It probably comes from the vendor's 50-to-20-minute median, which is a 60% drop in first-draft time. That test covers 40 procurement packets, excludes verification and does not establish end-to-end time saved (S1-announcement.md, "Vendor test"). The article gives the minutes and caveats and omits the percentage. |
| "Lead with USD 19" | Contradicted by the vendor's correction (S2-pricing.md, "Correction"). |
| Customer quote ("We fired our paperwork department") | S5 says there is no named source and no permission. Not used or paraphrased. |

I also left out three things the sources do not support:

- **A DeskLab accuracy rate.** Five of six is reported as a count only, because DeskLab says the review is not a benchmark.
- **Any mention of a free trial.** The packet does not mention one.
- **Pricing for waitlisted customers.** The packet does not say what they will pay when admitted.

## Attribution of performance claims

- **Vendor test (50 to 20 minutes):** attributed to the vendor as an internal test, with all three of its own caveats (S1-announcement.md, "Vendor test").
- **DeskLab (five of six correct pages):** attributed to DeskLab, with its stated limits (S4-independent-review.md, "Method").
- **Author's inference:** "Six documents cannot give you an error rate" and the "Should you try it now?" section are my reading, not source claims. The advice to test with documents whose answers you already know follows voice-sample.md and is worded as a recommendation, not as something done.

## Author calculations

These are not in the sources, and the article labels them "by my arithmetic".

- Minimum monthly cost: 3 seats × USD 29 = USD 87 before tax (S2-pricing.md, "Current plans").
- With OCR: USD 87 + USD 12 = USD 99 before tax.

## Checks performed

I ran `output/check_package.py` (Python standard library) against the final article and metadata.

- **Word count:** 749 words for the whole file including headline, subheads and inline citations. It is 707 without headings and 684 without headings or citations, so it stays within 600–800 on each count.
- **Headline:** 62 characters, identical to the article's H1.
- **Excerpt:** 39 words.
- **Search description:** 154 characters.
- **Citations:** all 10 inline citations resolve to a real input filename and an exact section heading. S1 to S4 are cited; S5 is not.
- **Excluded claims:** a text scan found no occurrence of "GDPR", "offline", "60%", "hallucinat", "certified", "fired", "free trial" or testing phrases. "USD 19" appears once, in the correction paragraph.
- **Not checked:** I compared every factual sentence to its source by reading; that is not an automated check. No style or plagiarism tooling was available, and no fact was checked beyond the packet.
