Methodology & Editorial Process

Critical Care Vault is built and maintained by a single CVICU registered nurse with a BSN. There is no team, no investors, no parent company, no marketing department. The project started as a personal collection of study references and bedside notes, grew into something larger than originally planned, and ships now as a free iOS app. The Android build has completed closed testing and is currently in Google Play's production-access review, which is taking a little longer than originally projected — apologies to anyone waiting on it.

It is not a business. There is no revenue, no paid tier, no advertising, no subscription, and no tracking. The developer's name and credentials are intentionally not displayed prominently in the app because the goal is a useful tool, not a personal brand. To contact the developer: support@criticalcarevault.com.

The question bank is the largest body of clinical content in the app and is also the most important to be honest about. The pipeline:

1. Source material begins with the developer's own clinical study references — a multi-year accumulation of bedside observations, textbook excerpts, society guideline summaries, and primary-literature notes maintained while working in critical care.
2. Question generation is AI-assisted. Large language models are used to scaffold question stems, distractors, and rationale prose from those source notes. The AI is not making up the underlying clinical material — it is rewriting human-curated study references into the structured question format the app uses.
3. Citation review is performed manually by the developer. Each rationale on newer questions includes one or more references to published literature; those references are checked for existence and for whether they actually support the claim being made. Hallucinated citations (a known LLM failure mode) are removed.
4. Releases ship the questions that have passed that review. Earlier-batch questions that predate the structured citation pass remain in the app while the back-catalog citation review continues.

This is a defensible pipeline for a solo-built educational tool, but it is not the same as physician peer review or institutional editorial oversight. We say so plainly because the alternative — pretending otherwise — would be both dishonest and discoverable.

• Has: developer-led citation review against primary literature and standard critical-care references.
• Has: structural QA against the in-app question schema (4-part rationale on newer questions, two cited sources where applicable).
• Has: user-reported corrections from clinicians who flag specific items via Settings → Report an Issue. Every flagged item is reviewed and either rewritten, retired, or kept with a brief justification.
• Has not: formal peer review by an external physician panel.
• Has not: sign-off from a hospital, residency program, professional society, or accreditation body.
• Has not: CME / CE accreditation for continuing-education credit.
• Has not: regulatory clearance as a medical device — and is not seeking it, because the app is explicitly not a medical device (see Intended Use).

The app is best used the way you'd use a textbook chapter, a colleague's well-organized notes, or a high-quality podcast: as one input to your reasoning, alongside the references you already trust, the institutional protocols you work within, and the judgment of the providers and pharmacists on your team.

The Intended Use page describes the boundaries in formal language: the app is an educational and reference tool, not a clinical decision-support system, not a substitute for clinical judgment, not appropriate during active resuscitation. None of that changes because the methodology is documented here. We're documenting the methodology so that the limits are visible from the start, not buried in a footer.

To be specific about a question that comes up often:

• The AI is rewriting human-curated study references into the structured question format, suggesting plausible-sounding distractors, drafting initial rationale prose, and helping format references.
• The AI is not generating clinical content from nothing, deciding which guidelines or studies to cite, or making the final call on which questions ship. Those decisions sit with the developer.
• The AI is not running on patient data, in the app at runtime, or making any decision the user sees — there is no LLM inside the shipping app. Everything LLM-assisted is in the offline content pipeline before release.

If you find an item that reads as machine-generated nonsense, factually wrong, or that cites a source that does not say what the rationale claims, please flag it (see below). That is the QA loop the app depends on.

The clinical content is grounded in standard critical-care textbooks, society guidelines, and primary literature. The full bibliography is accessible inside the app under Settings → References & Citations, and a representative list of the main sources by domain is published on the Intended Use page.

If you find a clinically incorrect question, a rationale that does not match the cited reference, a typo, an outdated guideline reference, or anything else that needs correction, the fastest path is in-app:

• In the app: Settings → Report an Issue. The form captures the screen / item context automatically.
• By email: support@criticalcarevault.com — please include the question text or item identifier so the report can be triaged quickly.

Reports route directly to the developer's inbox. There is no support team between you and the person who can fix the issue. Most clinically-substantive reports are reviewed within 24–48 hours and either rolled into the next release or, for safety-relevant corrections, into a point release.

If the project ever adds a formal physician review pipeline, an external editorial board, or any other layer of oversight, this page will be updated in the same release that introduces it — and the in-app About & Legal copy will be updated to match. We will not retroactively claim review processes that did not exist when a given question was written.

The Intended Use page describes the regulatory framing; this page describes the editorial framing. Both are kept current with whatever the methodology actually is at any given release.