# Editor notes: LumenDesk 2.0 article package (revision 2)

## Deliverables
- `article.md`: 778 words, not counting the H1, or 790 words including it. Both counts are from `check.py` and include subheads and inline citations.
- `metadata.json`: the headline is 61 characters and matches the article H1. The excerpt is 38 words. The search description is 159 characters. The JSON loads and has exactly three keys.
- `CHANGELOG.md`: lists the changes made in this revision.
- `v1-original/`: the first submission, kept unchanged.
- Helper scripts, not for publication:
  - `check.py` checks length, arithmetic, banned phrases and citations. It can be rerun.
  - `trim.py` and `trim2.py` made one-time text edits. They'll stop with an error if run again, because the text they replaced is gone.
  - `preserve.py` copied the first submission into `v1-original/`.

## Sources used
| Claim area | Source |
|---|---|
| Who can use it, browser only, waitlist, API and EU planned for Q1 2027 with no date or guarantee | S1-announcement.md, "Access" |
| File types, 50 MB limit, OCR needed for scans, drafts only (doesn't approve payments, send emails or execute contracts) | S1-announcement.md, "Features" |
| 50 → 20 minute median first-draft time from 40 packets, with the vendor's caveats | S1-announcement.md, "Vendor test" |
| USD 29/seat/month, billed monthly, 3-seat minimum, 500 uploads per workspace, no rollover, USD 0.08 overage, OCR USD 12/workspace/month, tax excluded | S2-pricing.md, "Current plans" |
| USD 19 screenshot was wrong; USD 29 is current; no annual offer | S2-pricing.md, "Correction" |
| No base-model training, 30-day default storage, early deletion by admins, 24 h active deletion, backups up to 30 more days | S3-data-policy.md, "Handling" |
| External model provider, no on-device mode, no claim of zero retention or compliance | S3-data-policy.md, "Processing" |
| 6 PDFs, 5 correct page citations, 1 wrong page with correct text, informal, not a benchmark, untested areas | S4-independent-review.md, "Method" |
| Audience, no hands-on work, don't turn internal notes into facts, avoid slogans | editorial-brief.md, "Assignment" |

## How the corrections and claims were handled
- **Price:** The article uses USD 29 per seat per month (S2, "Correction"). USD 19 appears only in the sentence explaining that the launch screenshot was wrong. Ignore the S5 suggestion to lead with USD 19.
- **Pricing calculations:** I calculated these from S2, and `check.py` asserts them. Three seats × USD 29 = USD 87. Adding OCR at USD 12 brings it to USD 99. For 750 uploads, (750 − 500) × USD 0.08 = USD 20. All amounts are before tax. The article labels the 750-upload case as an example, not a vendor figure.
- **Vendor's timing claim:** The article attributes this to LumenDesk and includes the vendor's own caveats. It covers drafting only, leaves out human checking, doesn't establish end-to-end time saved and has no task-level results published.
- **Independent review:** The article describes it as small and informal, which is how DeskLab describes it. It doesn't turn the result into an accuracy rate.
- **Available now vs. planned:** What's available now is limited to existing Team customers in Canada and the US, in the browser app. The API and EU access are planned for Q1 2027, with no date or guarantee. New customers can only join the waitlist, and no access date is given.
- **No hands-on testing:** The article says neither the writer nor anyone else at Kingy has tested the product (editorial-brief.md). The testing steps are offered as suggestions for readers.
- **Voice:** editorial-brief.md asks for a direct Kingy-style voice. The article borrows the sample's direct style and its practical advice, which is to test with known answers and count the time spent checking. Revision 2 removed the opening and closing lines that copied sentences from voice-sample.md almost word for word.

## S5 claims left out, with reasons
S5-internal-notes.txt describes itself as "unverified writer notes" and "leads requiring verification." None of its claims are in the article.
| S5 claim | Why it's excluded |
|---|---|
| "Eliminates hallucinations" | No source supports it. S4 found one wrong page citation in six, and S1 says the tool only drafts answers. |
| "GDPR certified" | No source supports it. S3, "Processing," says the policy doesn't claim guaranteed regulatory compliance. EU availability is only planned (S1, "Access"). |
| "Runs fully offline" | S3, "Processing," contradicts it. Requests go to an external model provider, and there's no on-device mode. |
| "Saves every team 60% of their work time" | The source is unnamed ("someone said"). S1, "Vendor test," says the vendor's figure doesn't establish end-to-end time saved. |
| "Lead with USD19" | S2, "Correction," says this price was wrong. |
| Customer quote "We fired our paperwork department" | It has no named source or permission. It's also a slogan, which editorial-brief.md says to avoid. |

## Assumptions and limitations
- I treated the dates and policies in the fixtures as authoritative within the scenario. I didn't do any outside research.
- **Needs an editor to confirm:** No source says whether the assistant changes the bill for current Team customers. The article now says this directly. It no longer implies that trying the assistant costs USD 87 on top of what they already pay.
- No source covers when waitlisted customers will get access, whether Team customers outside Canada and the US are included, or how DOCX files count toward uploads. The article doesn't speculate on these points.
- Word counts depend on how you count. My script counts citations like "S1-announcement.md," as one word each. A different counter could give a slightly different total. Both of my counts are within 600–800.
- Checks are text-based only. I didn't test how the Markdown renders or run a style or fact-checking tool.
