How We Verify

One Question. 3 Pipelines. 5 Gates.

Most sites read the abstract and hit publish. Here is what happens when you refuse to cut corners — from raw research paper, through multi-study synthesis, to the flagship guide you read.

52 Studies verified
63 Claims grounded
1,456 Verification checks
0 Corrections published

It Starts With a Paper Nobody Reads

Somewhere in PubMed, a team of researchers just published a meta-analysis. Forty-nine randomized controlled trials. Three years of work. Hundreds of pages of methodology, statistics, and carefully hedged conclusions. It sits behind a paywall, written in language only other researchers understand.

This paper contains something genuinely valuable — a finding that could change how you think about the protein on your plate. But between that finding and your breakfast, there is a gap. That gap is where misinformation lives. Where "suggests" becomes "proves." Where "in trained athletes" becomes "for everyone."

Most nutrition sites bridge that gap with a single writer, a single draft, and a publish button. FitChef bridges it with 3 dedicated pipelines, 5 verification gates, and a system designed to catch its own mistakes before you ever see them.

Here is exactly what happens to that paper. Every step. Nothing hidden.

Phase 1

Find

Before anyone touches a study, the pipeline determines if it belongs.

Agent 01 Cluster Scout

Searches PubMed, Google Scholar, and journal databases for every relevant study on a nutrition topic. Not the top 5. Not the first page. Every peer-reviewed study that exists. If a major paper on this topic was published and we missed it — our credibility takes a hit. So we don't miss it.

Output Complete study inventory with DOIs, abstracts, and metadata
Human gate Paper Handoff

A human provides the full paper. Not the abstract. Not a summary. The complete study — methods, results, discussion, limitations, funding disclosures. The pipeline cannot proceed without it. This is the first of three human checkpoints.

Gate Human approval required — pipeline pauses until full paper is provided
Phase 2

Extract

Read the full paper. Capture every data point. Miss nothing.

Agent 02 Study Extractor

Reads the entire paper and produces a machine-readable extraction of every number, finding, limitation, and methodological detail. This is not a summary. It is a complete, faithful capture of what the study contains — structured so every downstream agent can use it.

Captured Sample sizes, p-values, effect sizes, confidence intervals, population details, funding sources, every direct quote
Rule If it is not in the extraction, it cannot appear in the content. No exceptions.

The extraction produces a structured JSON file — the Golden JSON. This single file grows as each agent adds its work. It is the chain of custody for every fact on FitChef. Previous agents' data is never overwritten. Only added to.

Phase 3

Verify the Data

Gate 1 — Did we read the paper correctly?

Before a single word of content is written, the Truth Map Skeptic takes every extracted data point and checks it against the original paper. Not a summary. Not the abstract. The actual paper.

The Skeptic assumes everything is wrong until proven right. In health content, "90% correct" is "100% dangerous." Every number. Every quote. Every claim of causation. Every population qualifier.

Numbers match the paper If the paper says 47 studies and we wrote "about 50" — killed. Exact means exact.
Framing matches the researchers "Small benefit in trained athletes" cannot become "everyone needs more protein." The Discussion section decides, not us.
Quotes are word-for-word Changed even one word — even to "sound better"? Caught.
All warnings preserved If the researchers said "but keep in mind..." and we left it out — flagged. Every limitation they noted, we must note.
No semantic drift "Correlation" stayed "correlation." "Statistically significant" didn't become "clinically significant." Population limits respected.

Only data that survives Gate 1 enters the Grounded Truth Map — the verified foundation that all content is built on. Bad data here means bad content everywhere downstream.

Phase 4

Enrich

Find the angles nobody else has. Build the FitChef lens. Plan the story.

Agent 04 Study Enricher

Takes the verified findings and enriches them with cultural context, myth-busting angles, industry gaps, and real-world implications. Searches Reddit, TikTok debates, and forums to find what people actually argue about. Builds the FitChef lens — how this research connects to your actual meals. Identifies the genuinely novel angles that make our content worth reading.

Target 10–40% genuinely novel content per page — angles the competition doesn't have
Agents 06 + 06B SEO + Story Architect

Two agents working in parallel. The SEO Agent maps what real people are searching for — the exact questions, the intent behind them. The Story Architect designs the emotional arc: where the tension builds, where the revelation lands, where the reader screenshots and sends to a friend. Together they produce the blueprint that the writer follows.

Phase 5

Write

Transform verified data into content people fall in love with reading.

This is the moment where verified data becomes content. The Study Writer is not a summarizer — it is a world-class magazine editor who takes the verified facts, the novel angles, and the story blueprint and builds a page that is simultaneously scientifically rigorous and impossible to stop reading.

Every sentence pulls to the next. The data is there — exact numbers, real findings — but it is woven into something the reader cares about. Something she screenshots and texts to her training partner.

Agent 07 Study Writer

Writes the full study page — the magazine-quality article that makes the research accessible, vivid, and practical. Every number comes from the verified extraction. Every angle comes from the information gain research. Nothing is invented.

Agent 08 Audio Agent

A dedicated trillion-dollar audio producer. Does NOT read the article aloud — rewrites the content entirely for the human voice. Pacing, tension, rhythm, and "wait, what?" moments designed for someone with earbuds in. A separate craft from writing.

At this point, one study has produced a full editorial analysis page, a dedicated audio script, structured findings for future claim synthesis, and citation-ready structured data. Multiple formats from one paper — each built from the same verified foundation.

Phase 6

Verify the Content

Gate 2 — Does what we wrote still match what the paper says?

This is the step that matters most. Most verification systems check once. We check three times — once for the raw data, once for the finished content, and once for the final import package.

Why? Because writing changes things. A "modest benefit in trained athletes" quietly becomes "everyone should do this." A "suggests" turns into a "proves." It is not even intentional. It just happens when you turn data into a story. So the Skeptic runs again — this time against the finished content.

I don't care how good the writing is. If the paper says "small benefit in trained athletes eating 2g/kg" and our page says "everyone needs more protein" — I kill it. The writing can be beautiful. The facts must be right.

— The Skeptic
Agent 03 (Gate 2) Content Skeptic

Takes every sentence of finished content and compares it to the verified claims from Gate 1. Did the meaning shift? Did we add hype? Did we present a guess as a fact? Would the original researcher read our page and nod — or cringe?

Agent 12 Quality Auditor

Accuracy is necessary but not sufficient. The Quality Auditor checks that the content is genuinely world-class — not just correct, but the single best page on the internet for this topic. It reads the top 10 Google results for the same keyword and ensures our page makes every competitor feel like a downgrade.

Nothing passes without sign-off from both the Content Skeptic (truth) and the Quality Auditor (excellence). Accurate but boring? Rejected. Beautiful but inaccurate? Killed.

Phase 7

Ship

Final approval, system sync, and publication.

Human gate Final Approval

A human reviews the complete package before import. This is the third and final human checkpoint — editorial sign-off on everything the pipeline produced. The machine does the work. A human gives the green light.

Agent 13 System Sync

Ensures that every field, every relationship, and every schema mapping from the Golden JSON can be fully handled by our WordPress infrastructure. If a study introduces a new data type the system hasn't seen before, this agent flags it before import — not after.

The study is now live. Two layers, simultaneously:

Layer 1 — The Magazine

What you read. Magazine-quality narrative. Vivid, practical, impossible to stop reading. The reader forgets she is reading about a scientific study.

Layer 2 — The Whisper

What machines read. Schema markup. JSON-LD. Verification data. DOIs. Citation-ready answer capsules. Everything Google, ChatGPT, and researchers need to trust and cite us.

The Chain of Custody

Every step in the pipeline reads from and writes to a single file: the Golden JSON. It grows with each agent. Previous data is never overwritten — only added to. If the pipeline crashes at step 11, the log tells us exactly where to resume. If a claim on our site is questioned, we can trace it back through every agent that touched it — all the way to the original paper.

Raw extraction — the full paper, faithfully captured
Gate 1 verification — every number checked against the original
Structured findings — every data point from the paper, ready for synthesis
Information gain — novel angles the competition doesn't have
SEO + story blueprint — intent mapping and emotional architecture
Written content — study analysis page, audio script, Netflix-style card hooks
Gate 2 verification — every sentence checked against the evidence
Quality audit — world-class or rejected, no middle ground
Gate 3 verification — import data, infrastructure, and trust layer consistency checked
Human sign-off — editorial approval before import
Published — both layers live, fully traceable

This is not a flowchart on a slide deck. This is a living system that runs on every single study we publish. No study enters FitChef without completing this chain. No exceptions.

But the Journey Doesn't End With One Study

Everything above describes how one paper becomes a verified study page. But the real power of FitChef comes from what happens next — when findings from multiple studies are combined into something no single paper could ever provide.

Pipeline 2

Findings Become Evidence-Based Answers

When enough studies are verified, a second pipeline begins.

A Social Demand Scout scans TikTok, Instagram, YouTube, and Reddit to find what questions people actually care about. A Demand-Evidence Mapper cross-references those questions with all verified findings across all studies. For each question with strong evidence behind it — a claim is synthesized.

Not one study's opinion. The weight of all evidence. How many studies support this answer? How strong are they? What diverges? Every claim gets an Evidence Consistency Index and a Certainty Tier — High, Moderate, or Low — visible on the page so you can see exactly how consistent the evidence is.

A dedicated Skeptic gate verifies the synthesis is honest (no cherry-picking) and the writing matches the evidence (no drift). A dedicated Audio Agent rewrites the answer for your earbuds. That's 5 verification gates deep.

Pipeline 3

Answers Become Flagship Guides

The flagship resource. The page that replaces everything else.

Verified answers are woven into flagship guides — the kind of page you bookmark and come back to. Not a list of claims. A flowing editorial that uses evidence the way a great magazine article uses sources. You start curious and finish feeling like an expert.

A final Skeptic gate verifies that every claim in the guide is accurately represented. No overstatement. No creative drift. Every fact traces back through 5 verification layers to the original paper.

Three types per topic: The Complete Guide (the answer page), The Myth Buster (everything you thought was wrong), and The Skeptic Review (what we know and don't know). Together they cover every angle — for searchers, sharers, and skeptics.

After Publication, the Work Continues

Publishing is not the end. Science changes. New studies get published. Old studies get retracted. Findings that were "emerging" become "verified." Things we thought were true turn out to be incomplete.

When that happens, we update — and we do it publicly. Every correction is timestamped, permanent, and visible. The old version stays on record. Nothing gets quietly edited. You always know what changed and why. That is not a feature. That is the entire point.

Go Deeper
The Skeptic Protocol The gate system that enforces verification Methodology The full pipeline from paper to page Corrections Log When verification catches errors Trust Dashboard Live verification stats