About In[Neo]Sight
In[Neo]Sight is a neonatology knowledge platform that makes landmark research accessible to everyone involved in newborn care — from investigators and clinicians to the families at the bedside. Every article cluster is written, reviewed, and audited before it reaches you.
The Cluster Model
Each cluster publishes four documents covering the same source paper(s). Three are audience-facing articles; the fourth is a Source Transparency Report disclosing exactly how the cluster was produced.
Researchers & academic clinicians
- ·Full methodological detail
- ·Effect sizes, CIs, and p-values
- ·Critical appraisal and limitations
- ·Historical and mechanistic context
- ·Future research directions
Neonatologists, nurses & trainees
- ·Actionable bedside interpretation
- ·Guideline context
- ·Implementation considerations
- ·Subgroup nuances
- ·Clinical future directions
Families of NICU patients
- ·Plain-language explanation
- ·No unexplained jargon
- ·Findings in concrete terms
- ·What it means for your baby
- ·Warm, empowering tone
All audiences
- ·How the full text was retrieved
- ·Supplementary materials logged (or noted if inaccessible)
- ·Extracted data verbatim from the paper
- ·Study snapshot in plain language
- ·Corrections and revisions log
- ·AI model and version disclosed
Articles 1–3 each target 1,500–1,800 words. All four documents are published together as a unit — the cluster is not complete until all four are present and approved.
How Articles Are Written
Every cluster goes through a three-agent pipeline before publication. The roles are strictly separated — the same agent may not perform more than one role on the same cluster.
Pipeline: Researcher → Writer → Reviewer
Researcher
Retrieves and structures the source paper before any prose is written. Produces the Source Transparency Report — a published fourth document and the sole numerical reference for the entire cluster.
- → Fetches full text via PMC, Europe PMC XML, or Unpaywall (in that order)
- → Retrieves and screens the paper's supplementary materials — appendices, data supplements, and supplementary tables and figures — when accessible, logging each by label (or noting when they exist but cannot be accessed)
- → Transcribes study design, population, outcomes, and results verbatim into the report
- → Writes a plain-language Study Snapshot and "About This Report" for general readers
- → Verifies every reference at the point of capture against CrossRef or the canonical PubMed/PMC record — never from a search snippet alone — and records the verifying URL and date
- → Assembles the verified reference list once as a single block that every tier draws from
- → Places the cluster on hold if full text cannot be retrieved — no writing begins
Writer
Writes all three tiers from the Source Data Sheet exclusively. Does not fetch papers or supply numerical values from memory.
- → Produces all three tiers at 1,500–1,800 words each
- → Includes ≥ 10 references with hyperlinked DOIs, numbered in order of appearance
- → Generates each tier's reference list from the Researcher's single verified reference block — never hand-maintained separately
- → Adds bidirectional links between inline citations and reference entries
- → Discloses the AI model and version used in the article header
- → Performs a pre-submission citation self-check: confirms each cited source contains the specific finding attached to it, that the cited paper predates the claim, and that it is the primary (not secondary) source
- → Addresses all Reviewer revisions specifically before resubmitting
Reviewer
Uses the Source Transparency Report as the primary reference — not memory, not a fresh fetch. Performs two mandatory passes in sequence; Pass 2 does not begin until Pass 1 is complete and all resulting revisions are verified.
Pass 1 — Citation Integrity
- → Checks every citation individually for placement accuracy
- → Flags named failure modes: secondary-source displacement, temporal impossibility, correct-paper/wrong-finding, methods paper cited for outcome data, weak or imprecise source
- → Audits reference numbering, order, completeness, and bidirectional links
- → Re-confirms every cited reference against CrossRef/PMC — a full check, not a sample — building on the Researcher's verify-at-capture record
- → After any correction, runs a cluster-wide stale-citation sweep to confirm the superseded value appears nowhere outside the Transparency Report's audit trail
Pass 2 — Content & Coherence
- → Checks flow, coherence, register, and title/subtitle accuracy
- → Cross-checks every numerical value and outcome against the Source Transparency Report
- → Confirms the Transparency Report is publication-ready for a general audience
No cluster is published until the Reviewer has approved all three tiers. Mandatory revisions — factual errors, coherence failures, citation errors — must be resolved before approval.
Editorial Standards
Before every cluster reaches publication it passes an audit checklist. Structural checks run automatically; judgment checks require the Reviewer.
Source Transparency Report
- ✓Source Transparency Report present with correct filename (_4_Transparency.md)Script
- ✓Generated-with line presentScript
- ✓Fetch record documents method used, URL retrieved, and dateReviewer
- ✓Supplementary materials screened and logged, or documented as inaccessibleReviewer
- ✓Full text successfully obtained (no HOLD status)Reviewer
- ✓No Extracted Data fields marked [NOT IN FULL TEXT] remain unresolvedReviewer
- ✓"About This Report" section written in accessible plain languageReviewer
- ✓"Study Snapshot" accurately summarises source paper(s) without jargonReviewer
- ✓Corrections and Revisions section presentReviewer
Per-Article (all three tiers)
- ✓Word count within 1,500–1,800 wordsScript
- ✓AI model name and version disclosed in the article headerScript
- ✓Title accurately reflects content and does not name the tierReviewer
- ✓Subtitle identifies the source paper(s) and angle without naming the tierReviewer
- ✓All numerical values verified against the Source Data SheetReviewer
- ✓≥ 10 references in the References sectionScript
- ✓Every [N] in the body has a matching numbered entry in ReferencesScript
- ✓Every numbered entry in References is cited at least once in the bodyScript
- ✓References numbered 1–N sequentially by order of first appearanceScript
- ✓No duplicate reference numbersScript
- ✓No citation gaps (e.g. [1], [2], [4] with no [3])Script
- ✓Every inline [N] marker is a hyperlink anchored to its reference entryScript
- ✓Every reference entry has a ↩ back-link anchored to the first body occurrenceScript
- ✓Every DOI is a clickable hyperlink (https://doi.org/…)Script
- ✓No floating paragraphs outside the section structureReviewer
- ✓All citations verified against the source paper's actual findingsReviewer
- ✓Systematic per-citation review completed: no secondary-source displacement, temporal impossibility, correct-paper/wrong-finding, methods-paper-for-outcome, or weak-source errorsReviewer
- ✓Every reference's core fields verified at capture against CrossRef/PMC, with verifying URL recordedReviewer
- ✓Each tier's reference list generated from the single verified data-sheet block (no separately hand-maintained lists)Reviewer
- ✓Every cited reference (not a sample) verified via CrossRef/PMCReviewer
- ✓Stale-citation sweep clean: no superseded citation field appears anywhere except the Transparency Corrections trailScript
- ✓Reviewer approval documented before publicationReviewer
Reference Standards
Citations follow standard journal format. DOIs are rendered as clickable links. References are numbered in order of first appearance in the body. Every inline citation links down to its reference entry, and every entry links back to its first use in the body.
Style Principles
Bold used sparingly
Only for key findings or critical warnings, never decoratively.
Headers throughout
All articles over 1,000 words use section headings for navigation.
No unexplained jargon
Abbreviations are spelled out on first use in every tier.
Prose over bullet lists
Lists only where structure genuinely aids comprehension.
Consistent English register
British or American English used consistently within each article.
Warm, not alarming
The Layperson tier is informative and empowering, never frightening.
These standards are documented in the InNeoSight_Article_Requirements.md specification, last updated July 2026. If you are interested in contributing or have a paper to suggest, submit it here.