Editorial review: candidates.4 through .6
September 24, 2026 · Protocol 0.1-rc.4-candidate.6 · MCP 0.2.0rc6 · Skill/contract 0.2.0-rc.6
Scope and method
Candidate.4 applied the actual Academic Humanize v2.0.0 skill (academic-humanize) to the current protocol, reading guides, website copy, MCP guidance, report text and ai-ready Skill. The skill's integrity lock requires preservation of citations, reported values, technical terminology, scientific meaning and uncertainty.
The skill file used has SHA-256 859f9a7a6d739dbf6fce054c10360654fecfd9036262e0e88085ab980d622bfc. It was read in full before editing. No publication venue was specified, so the revision uses general scientific-writing conventions for the research material and plain instructions for practitioner guides.
The scope excludes private correspondence, the supplied topic-approval form, synced reference files, recorded feedback and frozen releases. Candidate.5 normalizes punctuation in the current reading copy. Candidate.6 corrects an inaccurate external-standard comparison in the current materials and earlier update entries at the author's request. Frozen originals remain unchanged. These records must not be rewritten to fit the current author's voice.
What changed
The previous text repeatedly used short slogans and contrast formulas. The revision replaces those patterns with explanations of the task, evidence and next action. It also reduces repeated warnings where the same boundary can be stated clearly once within a section. Necessary warnings remain in each document that may be read on its own.
| Area | Editorial change | Preserved content |
|---|---|---|
| Website and starting guide | Explain who can use the review, what to bring and what it produces. Reduce promotional headings. | Engineering-only scope, voluntary participation and evidence limits. |
| Protocol, QUICK-6 and full profile | Use direct instructions and clearer transitions between requirements. | Six gates, risk thresholds, required fields, formulas and stop rules. |
| Criteria and scenarios | Clarify checks and examples without altering their obligations. | Fifteen criterion IDs, gate mappings and constructed-example labels. |
| MCP, reports and Skill | Ask concrete questions and explain results without compressed slogans. | Tool names, data keys, deterministic decisions, privacy and authority limits. |
| Research and audit documents | Separate findings, interpretation and proposed evaluation more explicitly. | Sources, causal limits, untested assumptions and the distinction between study and practice. |
One cost paragraph could be read as subtracting residual manual work twice. It now follows the existing full-profile formula explicitly: gross savings already deduct residual manual work; net savings also deduct review, correction, escalation and rework. The formulas and engine have not changed.
Integrity checks
Candidate.4 froze the candidate.3 documents, packages and published web assets. Candidate.5 freezes candidate.4; candidate.6 freezes candidate.5. Each has its own download path. Tests compare the decision engine, contracts and record logic with the frozen distribution and compare all nine assessment fixtures across versions, allowing only version/provenance identity differences. Criterion IDs and gate mappings are checked separately. Candidate.6 deliberately removes the unrelated accessibility-standard references; other external citations remain checked against the frozen baseline. This requested framing correction is distinct from the candidate.4 prose edit.
An internal prose audit reviewed generic transitions, inflated claims, repeated sentence patterns, mechanical contrasts, paragraph structure and study-specific detail. Candidate.4 removed stylistic em dashes but retained range notation. At the author's request, candidate.5 also removes every remaining em dash and en dash from current reader-facing text, including ranges. Range endpoints, ordinary hyphens and mathematical minus signs are preserved. Precise requirement lists were retained where their repetition serves the review.
No supported external AI detector was available, so no detector result is reported. This revision makes no claim that the text is “undetectable” and assigns no human-likeness score. AI assistance remains disclosed.
Software checks cannot determine whether readers prefer the new wording or understand it better. Those questions require actual use. No new pilot, endorsement, measured usability result or effectiveness evidence is added by this revision.
See the test record, release status and update log.