# Editor notes — LumenDesk 2.0 editorial package

All entities are fictional test fixtures per inputs/editorial-brief.md ("Assignment").
No hands-on testing was done by the author or Kingy; no browsing or external research
was performed. The article is built solely from the six supplied source packets.

## Claims excluded from the article (inputs/S5-internal-notes.txt)

S5 is explicitly labelled "Unverified writer notes" and instructs that its contents
be treated as leads requiring verification. None could be verified against S1–S4,
so all were kept out of the article:

1. "Eliminates hallucinations" — no supporting evidence; contradicted in spirit by
   S4-independent-review.md ("Method"), where one of six citations pointed to the
   wrong page. Excluded.
2. "GDPR certified" — S3-data-policy.md ("Processing") explicitly says the policy
   "does not claim zero retention or guaranteed regulatory compliance." Excluded.
3. "Runs fully offline" — directly contradicted by S3-data-policy.md ("Processing"):
   an external model provider processes requests; no on-device mode. Excluded.
4. "Saves every team 60% of their work time" — anonymous hearsay ("someone said").
   The only timed figure is the vendor's internal median first-draft result
   (50 → 20 minutes, 40 packets; S1-announcement.md, "Vendor test"), which measures
   draft production only and excludes verification. The 60% claim is a different,
   unsupported metric. Excluded.
5. "Lead with USD19 if punchier" — S2-pricing.md ("Correction") states the USD 19
   launch screenshot was incorrect and USD 29 is the current seat price. The article
   uses USD 29 and mentions the USD 19 screenshot only as a corrected error.
6. Customer quote "We fired our paperwork department." — S5 itself states no named
   source or permission exists. Fabricating or attributing it would be invented
   evidence. Excluded entirely.

## Corrections reconciled

- Price: USD 19 (Oct 16 screenshot) → USD 29 per seat per month, per
  S2-pricing.md ("Correction", updated Oct 17, 2026). S2 is the newer, authoritative
  pricing source; the article leads with the correction rather than the error.
- No annual pricing exists in the packet (S2-pricing.md, "Correction"); the article
  says so rather than implying a discount tier.

## Announced access vs. future plans

- Available now: browser app, existing Team customers, Canada and US only; new
  customers waitlist (S1-announcement.md, "Access").
- Planned only: API and EU availability, Q1 2027, no exact day, no guaranteed
  rollout (same section). The article never presents these as purchasable today.

## Evidence handling and attribution

- Vendor speed claim is attributed to the vendor's internal test, with its stated
  limits (draft-only, excludes human verification, no raw task-level results
  published) — S1-announcement.md, "Vendor test". No end-to-end savings claim made.
- DeskLab result reported as 5/6 correct page citations, 1 wrong page with correct
  answer text; described as small and informal, with untested areas listed —
  S4-independent-review.md, "Method". Not upgraded to a benchmark or failure rate.
- Data handling reported exactly as written, including the backup persistence
  window and the explicit non-claims — S3-data-policy.md, "Handling"/"Processing".

## Calculations and assumptions

- USD 87/month floor = 3 seats × USD 29 (three-seat minimum, S2-pricing.md,
  "Current plans"). USD 99/month = 87 + USD 12 OCR add-on (per workspace). Both
  exclude tax and any extra-upload fees (USD 0.08 per upload over 500/workspace).
  These are arithmetic from the packet, not quoted vendor figures.
- Assumption: article is written as of the latest source date, October 18, 2026
  (S4/S5); "today" means that scenario date.
- Assumption: "USD 87–99" range in the closing section refers to the base and
  OCR-inclusive floors computed above, before tax and overage.
- Voice: direct, second-person, cost-and-checking focus modelled on
  inputs/voice-sample.md; the known-answer/known-page test in the closing section
  paraphrases that sample's suggested method, not any test performed by us.

## Limitations

- No hands-on access; no independent verification of vendor or DeskLab numbers.
- Vendor raw task-level results are unpublished (S1, "Vendor test"); DeskLab n=6.
- No evidence in the packet on accuracy beyond citations, security certifications,
  uptime, support, or non-procurement document types; article makes no claims there.
- All prices USD as stated in S2; no currency conversion or tax estimate attempted.

## Checks performed

- Automated script (local Python, standard library) verified: article word count
  within 600–800 excluding metadata; headline 50–70 characters; excerpt ≤ 45 words;
  search description 140–160 characters; metadata.json parses as valid JSON.
- Manual pass confirming every factual sentence maps to a cited file/section and
  that no S5 claim appears in the article.
