# Editor notes — LumenDesk 2.0

## Scope and editorial decisions

- This is an unpublished editorial package for the fictional exercise, assessed as of October 18, 2026, the latest dated supplied observation. “Now” means the access described in this packet, not a live availability check. No external research, product access or personal testing was performed.
- The audience is a small team deciding whether to try the assistant. The recommendation is conditional: eligible existing customers can run a controlled evaluation; new customers have a waitlist, not an immediate deployment option. Recommendations about trial design are editorial advice, not reported tests. Sources: `inputs/editorial-brief.md`, “Assignment”; `inputs/S1-announcement.md`, “Access”.
- Used direct language, concrete costs and attention to verification burden, following `inputs/voice-sample.md`, “Original voice sample written for this exercise”. No sample sentence or invented testimonial is presented as a customer quote.
- “Trial” means a suggested evaluation by an eligible customer, not an advertised free trial. The packet gives no free-trial terms, waitlist timing or guaranteed future rollout. No such terms are assumed. Source: `inputs/S1-announcement.md`, “Access”.

## Evidence and version reconciliation

| Article issue | Authority and treatment |
| --- | --- |
| Present versus planned access | `inputs/S1-announcement.md`, “Access”: existing Team customers in Canada/US, browser only; new customers waitlist. API and EU availability are Q1 2027 plans, not promises. |
| Supported files and actions | `inputs/S1-announcement.md`, “Features”: searchable PDF/DOCX, page-linked citations, 50 MB per file, scans require OCR; drafts do not execute transactions or communications. |
| Price correction | `inputs/S2-pricing.md`, “Correction” supersedes the October 16 screenshot: USD 29, not USD 19. “Current plans” governs monthly billing, minimum seats, pooled uploads, OCR and overages. No annual discount is inferred. |
| Performance | `inputs/S1-announcement.md`, “Vendor test”: attributed to LumenDesk, 40 procurement packets, medians of 50 and 20 minutes, draft stage only. Human checking is excluded and raw task-level results are absent. |
| Independent observation | `inputs/S4-independent-review.md`, “Method”: six searchable procurement PDFs; five correct page citations and one wrong page with correct text. No generalized accuracy rate, benchmark result or validation of vendor speed claims is inferred. Scans, OCR, privacy and workflow were not tested. |
| Data handling | `inputs/S3-data-policy.md`, “Handling” and “Processing”: base-model training restriction is not expanded into a blanket data-use promise. Default storage, deletion lag, backup persistence, external processing and no on-device mode are stated separately. |

### Derived figures, not additional source claims

Using `inputs/S2-pricing.md`, “Current plans”:

- Minimum monthly subscription: 3 × USD 29 = USD 87 before tax.
- Minimum with OCR: USD 87 + USD 12 = USD 99 before tax.
- Illustrative three-seat workspace with 750 uploads, without OCR: USD 87 + (750 − 500) × USD 0.08 = USD 107 before tax. This is a hypothetical bill, not observed customer spending.

Using `inputs/S1-announcement.md`, “Vendor test”: 50 − 20 = 30 minutes; 30 ÷ 50 = 60%. These describe the change between reported first-draft medians, not individual task savings, total staff time or a guaranteed outcome.

## Unverified notes excluded from article

All leads below come from `inputs/S5-internal-notes.txt`, “Unverified writer notes, October 18”. They are not independent evidence.

- **“Eliminates hallucinations”:** no supporting evaluation. DeskLab's limited citation observation cannot establish error-free performance (`inputs/S4-independent-review.md`, “Method”). Omitted rather than repeated as a launch claim.
- **“GDPR certified”:** no certification evidence in the packet. The data policy does not guarantee regulatory compliance (`inputs/S3-data-policy.md`, “Processing”). Omitted; no legal compliance conclusion is offered.
- **“Runs fully offline”:** incompatible with the supplied browser-only, external-processing description and absence of an on-device mode (`inputs/S1-announcement.md`, “Access”; `inputs/S3-data-policy.md`, “Processing”). Omitted.
- **“Saves every team 60% of their work time”:** unsupported generalization of vendor draft-stage medians. Only the bounded, attributed calculation is retained (`inputs/S1-announcement.md`, “Vendor test”).
- **“Lead with USD19”:** rejected because the later pricing source explicitly corrects that figure (`inputs/S2-pricing.md`, “Correction”). The article mentions it only to resolve the discrepancy.
- **Proposed customer quote about firing a department:** no named source or permission; not used, paraphrased or attributed to an invented customer.

## Checks actually performed

Read all supplied files and manually compared article claims with the cited sections. Reconciled the dated pricing correction, separated access from plans, and reviewed the article for unsupported claims and implied hands-on testing.

Ran `output/check_package.py` with Python's standard library. Results:

- Article body: **685 whitespace-delimited words**, including section headings and citation text, excluding the headline and metadata; **650 words** after removing parenthetical source citations. Both satisfy 600–800.
- Headline: **55 characters** (required 50–70).
- Excerpt: **30 words** (maximum 45).
- Search description: **150 characters** (required 140–160).
- Metadata parses as JSON with exactly the requested keys; article headline matches it.
- All cited article filenames exist; cost and percentage arithmetic assertions passed.

These are editorial and file checks, not product, privacy or workflow tests. Source-section alignment was checked manually, not by the script.

## Remaining limitations and handoff

The packet supplies no raw vendor results, end-to-end time measurement, broad independent evaluation, or tests of OCR/privacy/team workflow. Future availability remains uncertain. Those gaps prevent a general productivity or compliance endorsement, but do not block delivery of this qualified article. Sources: `inputs/S1-announcement.md`, “Access” and “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`. Reproducible check utility: `output/check_package.py`. Original inputs were not edited. No publication or images were produced, as requested.

Estimated human finishing time: **5–10 minutes** for a final copy edit and approval, not a measured result. Obtaining new product evidence would be a separate task with unknown timing.
