Methodology
How to trust, or distrust, what you read here.
Clarity is only as good as its method. This page is the method, in full: how literature is found and deduplicated, how each claim is graded, how systems map to the IFM Matrix, the limitations we deliberately accept, and the human gate every page clears before it publishes.
Every claim on this site is a hypothesis to discuss with a clinician, never a diagnosis or a treatment. What follows is not a promise that the evidence is strong — it is a promise that the evidence is labelled honestly so you can judge it yourself.
1 · The evidence grades
What A, B, C and D actually mean.
Every claim and every test finding carries one badge. The grade describes how strong the underlying literature is — never whether something is good or bad for you. Saturation, not colour, carries the signal, and the letter is always present.
Strongest
Meta-analysis / RCT
Solid
Cohort / observational
Emerging
Mechanistic / animal
Early
Hypothesis / opinion
- A Meta-analyses and randomised controlled trials — the strongest causal evidence.
- B Cohort and other well-controlled observational studies — consistent association, weaker on cause.
- C Mechanistic, in vitro, or animal work — a plausible pathway, not yet shown in people.
- D Hypothesis-level — reasoned, cited, and explicitly early.
An abstract-level tag means no open-access full text underpins the claim — it rests on an abstract alone, so read it more cautiously regardless of letter. The grading rubric is versioned; its current revision ships in the corpus metadata so a grade can always be traced to the rule that produced it.
2 · The IFM Matrix mapping
Where each system sits on the Functional Medicine Matrix.
The eight systems are not ad hoc — each anchors to a node on the Institute for Functional Medicine (IFM) Matrix. This is the exact mapping the deep-dive pages cite, kept in one place so the site can never disagree with itself.
- 01
Gut / Microbiome
AssimilationAssimilation covers how the body breaks down, absorbs, and tolerates what enters the gut — digestion, the microbiome, and barrier integrity. It is the IFM node where the gut–skin axis is described.
- 02
Nutrient Status
Assimilation + EnergyNutrient status spans two IFM nodes: Assimilation (whether nutrients are absorbed) and Energy (whether the cell can use them). Skin-relevant micronutrients sit across both.
- 03
Hormones
CommunicationCommunication is the IFM node for hormonal and neurotransmitter signalling — the messengers that coordinate tissues, including the skin, end to end.
- 04
Stress / HPA
CommunicationThe HPA (hypothalamic–pituitary–adrenal) axis is part of the Communication node: the stress-signalling pathway that, via cortisol, reaches the skin through the brain–skin axis.
- 05
Detoxification / Liver
Biotransformation & EliminationBiotransformation & Elimination is the IFM node for how the body chemically processes and clears compounds — largely hepatic. The Transport node is folded in here (disclosed on the methodology page).
- 06
Immune / Inflammation
Defense & RepairDefense & Repair is the IFM node for immune activation, inflammation, and tissue repair — where Th2 skew and barrier–immune crosstalk in eczema are described.
- 07
Skin Barrier
Structural IntegrityStructural Integrity is the IFM node for the integrity of membranes and barriers — at the skin, the filaggrin–lipid envelope whose breakdown raises transepidermal water loss.
- 08
Energy / Mitochondrial
EnergyEnergy is the IFM node for mitochondrial function and oxidative balance — the cellular energy supply that keratinocyte turnover and barrier repair depend on.
Two deliberate simplifications, disclosed here rather than buried: the IFM Transport node is folded into Detoxification / Liver and Immune / Inflammation rather than given its own system; and a few systems span two nodes (noted inline above). The Matrix is a lens for organising mechanism, not a clinical instrument.
3 · How papers are identified & deduplicated
One work, one identity — even without a PubMed ID.
Much of the integrative and nutrition literature — preprints, specialist journals, books — has a DOI but no PMID. A PubMed-keyed corpus would silently drop the core of that evidence base, so a paper’s canonical identity is its DOI, falling back to PMID, then OpenAlex Work ID only when no DOI exists.
Every paper carries an alias set
(doi · pmid · pmcid · openalex_id) and deduplication matches
on any alias, with OpenAlex as the authority for clustering
preprint and published versions of the same work. The internal corpus
filename is a stable slug allocated once and never changed — the
citation references you see (e.g. corpus slugs) stay stable
even if a DOI or PMID is discovered later and added to the alias set.
4 · Sourcing — coverage over a tidy reading list
What we deliberately include, and the cost we accept.
Per condition×system×subtype, the corpus is seeded with the top systematic reviews, meta-analyses and RCTs, then expanded by walking the citation graph two hops from those seeds. There are deliberately no depth-1 caps, no per-topic volume caps, and no recency or citation-count thresholds — that would quietly exclude foundational older and mechanistic work the whole premise depends on.
The honest cost: a relevance screen is the only thing between that wide net and a junk corpus. So the screen is specified strictly — an LLM classifier against a versioned condition×system×subtype rubric with a tunable score threshold, a random audit sample of both kept and rejected decisions on every refresh, and every rejection logged (not deleted) so the noise is auditable. Review load grows as conditions are added; it is managed by screen precision and the audit, not by silently capping coverage.
5 · The human review gate
Nothing publishes itself.
The corpus refreshes on a monthly schedule. A refresh never edits the live site directly — it opens a diff pull request, and merging that PR is the human review gate. A person reads the proposed claims, grades, quotes and citations in context before anything reaches a reader. Per-system sign-off is recorded before publish; until then a system honestly reads “coming soon” rather than showing unreviewed output.
The literature pipeline lands in Epic C; current claims are reviewed fixtures and are labelled as such on each page. The method above is the durable contract — it does not change when the content does.
6 · The testing page
How we chose the tests we list.
The functional testing guide at /testing is hand-curated, not pipeline-emitted. We use an LLM to draft a broad list of commercially-available tests per system, then a human reviewer narrows to two or three per system based on cross-provider coverage, manufacturer credibility, and documented clinical use. Test cards link to the manufacturer’s product page and carry no per-test citations — the evidence for each system lives on its condition deep-dive page. The biomarker registry was expanded the same way to cover all eight systems for eczema.