# Editor notes — LumenDesk 2.0

## Scope and editorial approach

This is a fictional exercise, assessed as of October 18, 2026, the latest dated source observation in the packet. “Now” refers to this launch window, not the real-world date. The available sources do not establish later changes. The article addresses small teams deciding whether to try a document assistant, as requested in `inputs/editorial-brief.md` — Assignment.

Read the full supplied brief and voice sample. Used the sample's direct, practical approach: check known answers, inspect citations and count verification work. Did not reproduce its sentences or imply firsthand experience. Neither the author nor Kingy conducted hands-on testing; the article states this explicitly. [inputs/editorial-brief.md — Assignment; inputs/voice-sample.md — Original voice sample written for this exercise]

“Trial” means a recommended evaluation by an eligible customer, not an advertised free trial. No free-trial offer, waitlist timing, cancellation terms or refund rights are established by the packet. The proposed evaluation uses non-sensitive documents and is advice, not an action performed or a tested outcome.

## Source and correction decisions

| Source and section | Treatment |
| --- | --- |
| `inputs/S1-announcement.md` — Access | Existing Team customers in Canada and the US have announced browser access. New customers have a waitlist, not immediate access. API and EU availability are Q1 2027 plans only, with no exact or guaranteed rollout date. |
| `inputs/S1-announcement.md` — Features | Included supported formats, 50 MB per-file limit, separate OCR requirement and limits on actions. Did not describe the assistant as autonomous. |
| `inputs/S2-pricing.md` — Current plans; Correction | October 17 correction controls: USD 29 per seat monthly, not the erroneous USD 19 launch screenshot. Three seats minimum; allowance belongs to the workspace, not each seat. No annual offer inferred. |
| `inputs/S1-announcement.md` — Vendor test | Attributed the 40-packet result to LumenDesk. Restricted the 60% calculation to reported median first-draft time; verification is excluded and raw task-level results are unavailable. No end-to-end savings claim. |
| `inputs/S4-independent-review.md` — Method | Attributed six-PDF observation to DeskLab, not the author. Reported five correct page citations and one wrong page despite correct answer text. Did not convert this into a general accuracy percentage or imply OCR, privacy or workflow testing. |
| `inputs/S3-data-policy.md` — Handling; Processing | Preserved the narrow base-model training statement, default retention, active-storage deletion timing, possible further backup retention and external processing. Did not equate “not used to train” with no storage or no third-party processing. |

The pricing correction replaces an earlier erroneous price; it is not evidence of a price increase from USD 19. The independent observation supplements the announcement but does not validate the vendor's speed result: its method addresses different questions. Inline citations identify local filenames and exact section headings. All source files were left unchanged.

## Calculations and metadata provenance

- Minimum monthly subscription: 3 seats × USD 29 = **USD 87**, before tax and overages.
- Minimum with OCR: USD 87 + USD 12 per workspace = **USD 99**, before tax and overages.
- Draft-time change: 50 − 20 = **30 minutes**; 30 ÷ 50 × 100 = **60%** reduction in the reported median. This is not an average saving across individuals or an estimate of total working time saved.
- Arithmetic inputs: `inputs/S2-pricing.md` — Current plans; `inputs/S1-announcement.md` — Vendor test.
- The excerpt's price comes from that three-seat calculation; its access caveat comes from `inputs/S1-announcement.md` — Access. The search description summarizes topics, not additional product claims.

## Unsupported leads excluded

All leads below come from `inputs/S5-internal-notes.txt` — “Unverified writer notes, October 18” (the file has no formal section headings).

| Lead | Why it is not used as a claim |
| --- | --- |
| “Eliminates hallucinations” | No supplied evidence establishes this. DeskLab's page-citation failure also makes an error-free framing inappropriate, though it is not itself evidence of wrong answer text. [S4 — Method] |
| “GDPR certified” | No certificate or supporting evidence is supplied. The policy expressly does not guarantee regulatory compliance. [S3 — Processing] |
| “Runs fully offline” | Unsupported and inconsistent with the browser-only access, external model processing and absence of on-device mode described in the packet. [S1 — Access; S3 — Processing] |
| “Saves every team 60% of their work time” | Improper generalization from a vendor median first-draft measurement that excludes checking. The article includes only the qualified calculation, not this assertion. [S1 — Vendor test] |
| “Lead with USD19” | Superseded by the explicit October 17 correction to USD 29. [S2 — Correction] |
| “We fired our paperwork department” | No named source or permission; no evidence for the implied staffing outcome. No customer quote was fabricated or included. [S5 — Unverified writer notes, October 18] |

S1–S5 in this table refer to the full corresponding `inputs/` filenames listed above, including `inputs/S5-internal-notes.txt`.

## Checks actually completed

Executed `output/check_package.py` successfully using the local Python standard library. Results:

- Article body: **622 whitespace-delimited words excluding headline and inline citations; 668 including inline citations**. Both counts exclude metadata and meet the 600–800 requirement.
- Headline: **61 characters**, within 50–70; matches the article heading.
- Excerpt: **31 words**, within 45.
- Search description: **151 characters**, within 140–160.
- Metadata parses as JSON and contains exactly `headline`, `excerpt`, `search_description`.
- All **13 citation groups** in the article resolve to supplied files and section headings.
- Price and percentage calculations pass explicit arithmetic checks.

Also compared the article's claims against the source passages while editing, including the corrected price, workspace upload allowance, future-versus-current access and deletion timelines. The automated citation check verifies headings exist; it does not independently prove that prose is supported. No browser use, product testing, external research, publication or image work was performed.

## Remaining limitations and handoff

The package cannot establish real-world time savings, general answer accuracy, OCR quality, privacy implementation or suitability for a particular business's regulatory obligations. Vendor raw results are absent and DeskLab's review is explicitly small and informal. These are evidence limitations, not missing tests secretly performed. [inputs/S1-announcement.md — Vendor test; inputs/S4-independent-review.md — Method; inputs/S3-data-policy.md — Processing]

Deliverables: `output/article.md`, `output/metadata.json`, `output/editor-notes.md`. Supplemental reproducible checks: `output/check_package.py`.

Estimated human finishing time: **5–10 minutes** for final copy review and house-style handling of local citations. This is an estimate, not measured work time; additional reporting or product evaluation would require a separate scope. No further work is required to meet this exercise's stated deliverables.
