# Editor notes — LumenDesk 2.0 editorial package (revised)

## Scope and method
- Deliverables: `article.md`, `metadata.json`, this file, plus `change-log.md`.
  The first submission is preserved unchanged in `output/original-submission/`.
  All entities are fictional test fixtures per `inputs/editorial-brief.md`
  ("Assignment"); dates and policy versions in the packets are treated as
  authoritative within the scenario.
- Sources used: S1-announcement.md (Oct 16), S2-pricing.md (Oct 17), S3-data-policy.md
  (Oct 17), S4-independent-review.md (Oct 18), plus editorial-brief.md and
  voice-sample.md for audience and tone. S5-internal-notes.txt was read but treated
  strictly as unverified leads, per its own header and the brief ("Do not turn
  internal notes into established facts").
- No hands-on testing by the author or Kingy; the article says so explicitly and
  attributes all performance figures to the vendor or DeskLab. No browsing, no
  external research, per the brief.

## Unsupported claims kept out of the article (all from S5-internal-notes.txt)
1. "Eliminates hallucinations" — no supporting evidence in any packet. S4 ("Method")
   documents a wrong-page citation in a six-file informal review; nothing establishes
   error-free output. Excluded.
2. "GDPR certified" — directly unconfirmed: S3-data-policy.md ("Processing") states
   the policy "does not claim zero retention or guaranteed regulatory compliance."
   Excluded; article instead reports what the policy actually says.
3. "Runs fully offline" — contradicted by S3 ("Processing"): an external model
   provider processes requests; no on-device mode. Excluded.
4. "Saves every team 60% of their work time" — unsupported generalization. The only
   quantitative claim is S1 ("Vendor test"): median first-draft time 50→20 minutes on
   40 procurement packets, excluding human verification, no raw task-level results.
   The article carries that claim with all its caveats and attribution, not the 60%
   figure. Excluded.
5. "Lead with USD19 if punchier" — rejected. S2-pricing.md ("Correction") states the
   Oct 16 launch screenshot showing USD 19 was incorrect; USD 29 is current. The
   article uses USD 29 and explicitly flags the screenshot error so readers who saw
   the launch image are not misled.
6. Quote "We fired our paperwork department." — no named source or permission
   exists (S5 itself says so). Fabricating or attributing it would violate the brief.
   Excluded entirely; not paraphrased either.

## Corrections reconciled
- Price: S2 (Oct 17) supersedes the Oct 16 launch screenshot. Article states USD 29
  per seat/month, monthly billing, three-seat minimum, no annual offer in the packet
  (S2, "Current plans" and "Correction").
- Access vs plans: article separates today's access (existing Team customers,
  Canada/US, browser only; waitlist for new customers) from Q1 2027 plans (API, EU)
  and notes no exact date or guaranteed rollout (S1, "Access"). The closing section
  repeats that Q1 2027 "is a plan, not a promise."

## Review-pass corrections (this revision)
- Removed "The waitlist costs nothing but patience": S1 ("Access") states a waitlist
  exists but says nothing about its cost; "costs nothing" was an unsupported claim.
  Replaced with "Joining the waitlist is the only step available to you today."
- Changed "the only hands-on evidence" to "the only independent hands-on evidence":
  the vendor's internal test (S1, "Vendor test") is also hands-on; DeskLab's
  distinction is independence, not exclusivity of hands-on testing.
- Details and re-verification in `change-log.md`.

## Calculations shown in the article (derived, not source claims)
- Minimum spend: 3 seats × USD 29 = USD 87/month, before tax (S2, "Current plans";
  prices exclude tax).
- With OCR add-on: 87 + 12 = USD 99/month per workspace, before tax.
- Overage example: 100 extra uploads × USD 0.08 = USD 8. Uploads are 500 per
  workspace per month (not per seat), no rollover.

## Evidence handling
- Vendor test (S1, "Vendor test"): attributed to the vendor, labeled internal,
  median not mean, draft-only, verification excluded, no raw data published.
- DeskLab (S4, "Method"): labeled small/informal, not a benchmark; 5 of 6 correct
  page citations, 1 wrong page with correct answer text; scans, OCR, privacy and
  team workflow untested. Article does not extrapolate an error rate from n=6.
- Voice: direct, second-person, cost-and-checking focus modeled on
  inputs/voice-sample.md (start with a document whose answer you know; check the
  citation; count the checking left behind). No slogans; script confirmed no
  verbatim sentence is shared with the sample.

## Assumptions recorded
- Word count "excluding metadata" interpreted as excluding the H1 headline (which
  duplicates metadata.json). Body = 744 words; including H1 = 754. Both within
  600–800, so the interpretation does not affect compliance.
- Currency stated as USD and all prices flagged "before tax" because S2 says prices
  exclude tax; no tax rate is assumed.
- "Existing Team customers" assumed to mean customers already on the Team plan as
  named in S1/S2; the packets do not define migration from other plans, so the
  article does not speculate about them.
- Waitlist cost is not stated in any packet; the article makes no claim about it.
- No annual pricing discussed beyond noting none exists in the packet (S2).

## Checks actually performed (local scripts: check_package.py, review_checks.py)
Run on the revised deliverables:
- Article word count: 744 excluding H1 / 754 including — within 600–800. PASS
- Headline length: 52 characters — within 50–70. PASS
- Excerpt: 37 words — at most 45. PASS
- Search description: 154 characters — within 140–160. PASS
- All 9 inline citations (filename + quoted section heading) resolve to actual
  "##" headings in inputs/. PASS
- Arithmetic: 3×29=87; 87+12=99; 100×0.08=8.00. PASS
- Article contains no "GDPR", "offline", "hallucination", "60%", or fabricated
  customer quote; the only "USD 19" mention is the explicit correction. PASS
- metadata.json parses as valid JSON with exactly the three required keys
  (explicit key-set check added in the review pass). PASS
- Article H1 matches metadata headline exactly. PASS
- No verbatim sentence overlap with inputs/voice-sample.md. PASS
- Original submission preserved byte-for-byte via file copy before revision. PASS
- Line-by-line re-read of article against S1–S4 found and fixed the two issues
  listed under "Review-pass corrections"; no other factual mismatches found.

## Limitations
- No hands-on testing was done or possible; all product behavior claims rest on the
  vendor packet and one six-file informal third-party review.
- Vendor raw task-level results are not published (S1), so the 50→20 minute figure
  cannot be independently checked.
- OCR quality, privacy behavior in practice, team workflow, and non-procurement
  document types are untested by any source (S4, "Method").
- No information in the packets on support, uptime, security certifications,
  waitlist cost, or contract terms beyond S3; the article does not speculate.

## Estimated human finish time
Roughly 15–25 minutes for an editor: fact-check pass against the packets, tone
polish, and headline A/B preference. This is an estimate, not a measured result.
