# Editor notes — LumenDesk 2.0 article

## Basis
- The article uses only inputs/S1–S4, the editorial brief and the voice sample. I did no browsing or outside research.
- Neither the author nor Kingy did any hands-on testing (editorial-brief.md, "Assignment"). The article says this in the first paragraph and never claims personal use. The trial advice in the last section is conditional guidance adapted from voice-sample.md. It does not report any result.

## Corrections reconciled
- **Price:** USD 29 per seat per month is current. The USD 19 in the October 16 launch screenshot was wrong (S2-pricing.md, "Correction"). The article reports the correction and does not lead with USD 19, even though S5 suggests doing so.
- **Dates:** The announcement is dated October 16, 2026, and the pricing and data policy are dated October 17. The only conflict between sources is the price, and the article follows the later correction.

## Calculations (from S2-pricing.md, "Current plans")
- Minimum monthly cost: 3 seats × USD 29 = USD 87 before tax.
- With OCR: USD 87 + USD 12 = USD 99 before tax.
- Overage example (my own illustration, not a vendor figure): 700 uploads − 500 included = 200 × USD 0.08 = USD 16.
- The upload allowance is per workspace, so adding seats does not add uploads.

## Available now vs planned
- **Available now:** existing Team customers in Canada and the US, browser app only. Everyone else can join the waitlist (S1, "Access").
- **Planned:** an API and EU availability in Q1 2027, with no date or guarantee. The article describes these only as plans.

## Performance claims and how they are attributed
- **50→20 minute median:** this is the vendor's internal test of 40 packets. It covers producing a draft only, leaves out human checking, does not show end-to-end time saved, and has no raw data (S1, "Vendor test"). The article attributes it to LumenDesk and includes all of these conditions.
- **DeskLab, 5 of 6 correct pages:** this is an informal, small review, not a benchmark. It did not test scans, OCR, privacy or workflow (S4, "Method"). The article attributes it and includes these limits.

## Claims excluded from the article (S5-internal-notes.txt)
S5 describes itself as "unverified writer notes" to be treated as leads. None of these claims appears in the article.

| Claim in S5 | Why it was excluded |
|---|---|
| "Eliminates hallucinations" | No source supports it. S4 found one wrong page citation out of six. |
| "GDPR certified" | S3 "Processing" says the policy does not claim guaranteed regulatory compliance. EU availability is only planned (S1). |
| "Runs fully offline" | S3 "Processing" contradicts it: an external model provider processes requests, and there is no on-device mode. |
| "Saves every team 60% of their work time" | The source is "someone said". The vendor's own test does not establish end-to-end time saved (S1 "Vendor test"). |
| "Lead with USD19" | The price was corrected to USD 29 (S2 "Correction"). |
| Quote: "We fired our paperwork department" | It has no named source or permission (S5). It is also a slogan, which the brief says to avoid. |

## Assumptions
- "Kingy" refers to the publication or voice owner named in the brief. Following the brief's instruction, the article states that neither the author nor Kingy did hands-on work.
- The headline is the H1 in article.md. The 600–800 word count excludes the H1. The body is 780 words counting subheadings and 749 words without them, so it is in range either way.
- The article cites filenames with section headings inline. It has no separate reference list.

## Open questions for the desk
- Is there a date or confirmed scope for the API and EU rollout? (Not in S1.)
- Can LumenDesk provide raw task-level results from its test? Does it have any measurement that includes time spent checking?
- Do sources include an OCR accuracy figure or a sub-processor name? (Neither is supplied.)

## Checks performed (output/check.py, run during revision)
- Body is 780 words counting subheadings and 749 without them (limit 600–800).
- Metadata limits: the headline is 61 characters (limit 50–70), the excerpt is 40 words (limit 45) and the search description is 153 characters (limit 140–160). The H1 matches the metadata headline.
- metadata.json parses as JSON and has exactly the keys headline, excerpt and search_description.
- Each of the 8 inline citation pairs (file plus section) matches an actual "##" heading in inputs/. The article does not cite S5.
- Arithmetic: 3 × 29 = 87; 87 + 12 = 99; (700 − 500) × 0.08 = 16.
- Every number in the article body either appears in S1–S4 or is one of the stated calculations (87, 99, 16, plus the illustrative 700). No other numbers were flagged.
- Searching the article for the excluded S5 claims and for personal-testing wording finds no matches. "USD 19" appears only in the sentence that reports the correction.
- I did not run a separate check for meaning or tone. The comparison of each claim against its source sentence was done by reading.
