# Editor notes

## Scope and editorial decisions

- This is an unpublished editorial package about fictional fixtures, assessed against the supplied October 16–18, 2026 packet. “Now” means the access described in that packet, not the real-world date. No external research, product access, personal testing or customer interviews were used.
- Followed `inputs/editorial-brief.md` — “Assignment”: explain availability, cost, evidence limits and suitability for small teams. Used the direct, checking-first approach in `inputs/voice-sample.md` — “Original voice sample written for this exercise”, without presenting its hypothetical test as completed work.
- The article recommends only a controlled evaluation for eligible existing customers. It does not imply a free trial or a new-customer signup path beyond the waitlist. The packet provides no trial, refund or waitlist timing terms (`inputs/S1-announcement.md` — “Access”).
- Article source markers resolve to local filenames and exact section headings. Recommendations are editorial judgments, not additional product findings. Metadata summarizes sourced article points; the USD 87 figure is a derived minimum, not a quoted seat price.

## Corrections and calculations

- Use the October 17 correction, not the October 16 screenshot: USD 29/seat/month, three-seat minimum; no annual offer supplied (`inputs/S2-pricing.md` — “Current plans”; “Correction”). Minimum: 3 × USD 29 = USD 87/month. With OCR: USD 87 + USD 12 = USD 99/month. Both exclude tax and upload overages.
- The 500-upload monthly allowance belongs to the workspace, not each seat; it does not roll over. Overage is USD 0.08/upload (`inputs/S2-pricing.md` — “Current plans”). No unsupported definition of what counts as an upload was added.
- Vendor test calculation: (50 − 20) / 50 = 60% reduction in median first-draft time; absolute reduction is 30 minutes. It is not a measured 30-minute saving on every packet, a guarantee, or an end-to-end productivity result. Human verification is excluded and raw task-level results are absent (`inputs/S1-announcement.md` — “Vendor test”).
- Current access is existing Team customers in Canada and the United States, browser only. API and EU access are Q1 2027 plans, not guaranteed releases (`inputs/S1-announcement.md` — “Access”). No precise future launch date was inferred.

## Unsupported leads excluded

All leads below come from `inputs/S5-internal-notes.txt` — “Unverified writer notes, October 18”; they are not established facts.

| Lead | Treatment and evidence |
| --- | --- |
| “Eliminates hallucinations” | Omitted: no supporting evaluation. DeskLab's small review reports one incorrect page citation and cannot establish absence of hallucinations (`inputs/S4-independent-review.md` — “Method”). |
| “GDPR certified” | Omitted: no certification evidence. The policy makes no guaranteed regulatory-compliance claim (`inputs/S3-data-policy.md` — “Processing”). This is not a finding of noncompliance either. |
| Runs fully offline | Omitted as contradicted by browser-only access and external model processing with no on-device mode (`inputs/S1-announcement.md` — “Access”; `inputs/S3-data-policy.md` — “Processing”). |
| Saves every team 60% of work time | Omitted: improperly generalizes an internal first-draft result. The article explains the narrower calculation and excluded verification (`inputs/S1-announcement.md` — “Vendor test”). |
| Lead with USD 19 | Rejected: explicitly corrected to USD 29 (`inputs/S2-pricing.md` — “Correction”). |
| Alleged customer quote about firing a paperwork department | Omitted entirely from article and metadata: no named source or permission, as the internal notes themselves acknowledge. No interview or employment outcome can be inferred. |

## Evidence limits and remaining editorial work

- DeskLab's six-PDF observation is attributed as informal, not converted into a generalized accuracy percentage. It tested neither scans/OCR nor privacy or team workflow (`inputs/S4-independent-review.md` — “Method”).
- The vendor's no-base-model-training statement is not expanded into zero retention, local processing or compliance assurance. Active-storage deletion and possible backup persistence remain distinct (`inputs/S3-data-policy.md` — “Handling”; “Processing”).
- There is no author hands-on evidence and no independent end-to-end time/cost study in the packet. A real procurement decision would need a buyer's own verification-time and document-policy assessment. Those are suggested next steps, not completed checks.
- Final human copy review is estimated at 5–10 minutes (an estimate, not measured work). Product evaluation or obtaining additional evidence would be separate work of unknown duration. No publication, images or external research are required for this assignment.

## Checks performed

Read every supplied input and the shared instructions. Reconciled the article's factual statements against the cited sections and reviewed the excluded internal-note leads. A local standard-library script checked JSON parsing and required keys, matching article/metadata headline, headline/excerpt/search-description limits, article word counts and the stated price/time calculations; recorded results appear below.

## Revision review

- Preserved the complete initial submission in `output/original-submission/`; its `manifest.json` records SHA-256 digests. Revised deliverables remain at the original top-level output paths.
- Re-read all seven supplied inputs. The source conflict remains correctly resolved in favor of the October 17 USD 29 correction. The access, feature, retention and review facts required no substantive correction.
- Tightened the standalone excerpt to include Canada/United States eligibility. Replaced “trial” with “evaluation” to avoid suggesting a trial offer and “promise” with “plan” to avoid overstating the roadmap. Added the scenario year in the lead, made the absence of a documented annual offer explicit, and stated the pre-tax qualifier in the search description and closing recommendation.
- Expanded the proposed buyer evaluation with checking-time and upload-cost criteria. These are recommendations only; no new tests or product findings are asserted. The closing now asks whether enough checked drafts justify the recurring cost, rather than implying one draft is the evaluation unit.
- Rechecked USD 87 base minimum, USD 99 with OCR and the narrowly scoped 60% draft-stage reduction. No arithmetic corrections were needed. No new research, product tests or interviews were performed.
- Validation now also counts prose without headings or source markers, checks that every article source marker has a definition, verifies cited source filenames/headings exist, and verifies the preserved originals against their manifest. Automated checks validate structure and arithmetic, not factual truth; source reconciliation was an editorial review.

### Automated check results

- article prose words excluding headings and citation markers: 630
- article body words including section headings: 671
- article all words including title and sources: 708
- headline characters: 58
- excerpt words: 31
- search description characters: 151

All automated assertions passed, including JSON fields, matching headline, length limits, citation definitions and source headings, arithmetic, and original-submission manifest verification. Word counts use whitespace separation. These are format/arithmetic/archive-integrity checks, not product tests.
