How We Research

The 8 phases

Every condition we cover follows the same eight-phase research process. No phase is skipped; nothing advances without pair verification.

1. Phase 1

   Disease Landscape Triage

   Every disease assessed by complexity tier; simplest researched first.

   Before any research begins, every condition we plan to cover is assessed and scored by complexity tier (1–5) based on the expected depth of evidence across the research sections we publish — Guide, Evidence, Comparison, Practice Gaps, Situation, Kit, and Story. Tier 1 (simplest) conditions are researched first; higher tiers require progressively more time and resources.

   Scoring is performed by the research pair using domain knowledge and the available landscape of published literature. The triage produces a prioritized queue so the most accessible evidence reaches patients first while deeper, more complex conditions get the time they need.

2. Phase 2

   Systematic Query Gathering

   Two researchers independently design queries, compare, execute, deduplicate.

   For each condition, two research agents independently design and execute structured PubMed queries. The pair starts with a knowledge inventory: each researcher drafts what findings they expect to exist, then the lists are compared. Gaps between their expectations become additional search targets.

   Query design covers every research section through 7–8 structured topic groups. All queries are executed and results deduplicated across topic groups. Papers are ranked by cross-query hit frequency — a paper appearing in 6 topic groups is more likely relevant than one appearing in 1.

   Every retrieved title is then reviewed at three evidence tiers:

   • L1 (5+ query hits): high-confidence relevance — every title read and decided
   • L2 (2–4 hits): moderate confidence — every title read and decided
   • L3 (1 hit): targeted keyword rescue probes through single-hit papers to catch anything missed by ranked queries

   The second researcher reviews all query designs before execution and spot-checks title screening decisions. All exclusions include documented rationale, and the full evidence chain is preserved on disk.

3. Phase 3

   Abstract Screening + Evidence Tiering

   Every abstract read; papers classified Essential / Confirmatory / Nice-to-have; citation mining + gap queries fill thin spots.

   Every paper on the master read list undergoes abstract-level review. The primary researcher reads each abstract in full, with spot-checking by the second researcher. Each paper is then classified into one of three tiers:

   • Essential — foundational to the condition page; must be read in full text
   • Confirmatory — supports or validates Essential findings; read in full where available
   • Nice-to-have — peripheral relevance; available for reference but not required

   Reference lists of Essential papers are then mined for studies not captured by the original queries, and citation graphs are walked to identify additional relevant work. Where evidence is thin or absent for a particular section, targeted gap queries are designed and run; results undergo the same screening.

   After all screening is complete, the pair jointly re-evaluates the Confirmatory tier. Enrichment findings from deeper reading sometimes reveal that a Confirmatory paper is actually Essential. Promotions are documented.

4. Phase 4

   Article Procurement + Registry Building

   Every Essential and Confirmatory paper procured in full text; structured registry built.

   Every Essential and Confirmatory paper is procured in full text where accessible. Articles are sourced from PubMed Central, open-access publisher sites, and institutional access; multiple download methods are attempted for each article.

   A structured source registry is built for each condition listing every article with citation details, PMID, evidence classification, section mapping, replication status, and access level (full text vs. abstract only). The second researcher performs targeted full-text reads of key articles to verify that abstract-level classifications are correct and to capture structural detail (tables, supplementary data) not visible in abstracts.

   A formal gap report identifies sections with missing coverage, abstract-only key claims, unreplicated findings, contradictions between studies, and unused articles. The manager independently verifies that every article marked "abstract only" was genuinely inaccessible after all download methods were exhausted.

5. Phase 5

   Disease Page Writing

   Writer reads every Essential article in full and builds the page DURING reading; pair-discussed throughout.

   A dedicated writer agent reads every Essential article in full and builds the disease page during reading — not after. Articles are delivered in topic-grouped batches; the writer reads each batch and immediately writes the corresponding page sections, building the page as understanding develops rather than synthesizing from memory after reading everything.

   The writer applies evidence classification markers throughout: established knowledge (verified across multiple sources), expert assessment (reasonable inference from available data), and expert inference (single-source or interpretive claims requiring verification). These markers make the evidence basis of every claim visible.

   The writer works with a second writer agent. Both read the same articles. They discuss interpretations, flag disagreements, and resolve ambiguities together. Neither writer works alone. Every factual claim must trace to a specific article in the source registry. Claims from abstract-only articles are constrained to what the abstract explicitly states — no elaboration beyond the abstract ceiling. Unreplicated findings from single studies are explicitly caveated.

   When the writer encounters claims worth including but not traceable to the research package, the research pair is asked to investigate and either provide new citations or confirm the gap.

6. Phase 6

   Four-Pass Pair Review

   Original research pair reviews across four dimensions: evidence gaps / classification / prose density / content placement.

   After writing is complete, the original research pair conducts a structured four-pass review. Each pass addresses a different dimension of quality:

   • Pass 1 — Evidence Gap Resolution: every evidence marker is evaluated. Claims marked as expert inference are investigated — can a citation be found? If yes, the claim is upgraded with a source. If no source exists, the claim is removed or explicitly caveated. This pass runs first so all unsourced claims are resolved before content review begins.
   • Pass 2 — Evidence Classification Verification: Core vs. Emerging classifications are verified against the actual source documents. Same-cohort independence is checked — findings from the same study population are not double-counted as independent evidence.
   • Pass 3 — Prose Density Evaluation: every sentence is evaluated for whether it earns its place. The manager routes each section through editorial for density assessment, then back through the research pair for accuracy verification of proposed compressions, then to the writer for implementation. One section at a time, sequentially — no batching.
   • Pass 4 — Content Placement Audit: every piece of content is checked for correct structural location. Mechanism descriptions belong in the pathways section, not in mutation table notes; prognostic data belongs in clinical context, not diagnostic testing. This pass catches content that migrated during the intensity of writing.

   Both researchers form independent assessments before discussing. Full section text is shared — not summaries. Each section receives its own dedicated discussion. The writer is engaged as a partner, not directed — the writer has deeper article knowledge than the reviewers and may have valid reasons for choices that look unusual from the outside.

7. Phase 7

   Independent Editorial Verification

   Separate editorial pair (no prior exposure) verifies every PMID and every factual claim against source documents.

   After all four review passes are complete, a separate editorial pair — agents who did NOT participate in the research or writing — independently verifies every factual claim:

   • PMID Verification: every cited PMID is checked against PubMed to confirm it resolves to the claimed paper.
   • Claim-to-Source Verification: each factual claim is compared against the actual source article on disk. Study design, sample size, key findings, statistical values, journal, and year are all verified.
   • Pair Verification: both editors split the work evenly, then cross-check each other's findings. Neither editor works alone.
   • Error Documentation: every discrepancy is documented with severity classification, the exact claim, what the source actually says, and which editor found it.

   The editorial pair has no prior exposure to the research or writing — verification happens with fresh eyes against primary sources, not against the writer's memory of those sources.

8. Phase 8

   Manager Board-View Review and Delivery

   The manager agent — who has overseen the entire process — conducts a final board-level review and assembles the deliverable.

   The manager agent — who has overseen the entire process from condition assignment through review — conducts a final board-level review:

   • Scope Alignment: does the research package address the original assignment? Is anything out of scope or missing?
   • Template Compliance: all sections present, tables complete, gaps documented with honest explanations.
   • Notation Cleanup: internal working markers are converted to clean, professional output.
   • Delivery: final package assembled — research content, source registry, gap report, and source article archive.

Quality architecture: five layers of trust

Five built-in layers reinforce each other across every phase. Nothing gets presented as fact until it has been independently confirmed.

1. Pair collaboration — No agent works alone at any phase; every research, writing, and review step is pair-verified.
2. Epistemological markers (C2C) — A formal notation system tracks uncertainty so unverified claims are structurally visible, not hidden.
3. Manager oversight — A dedicated manager AI ensures every phase receives full rigor and process discipline.
4. Independent editorial verification — A separate pair, with no prior exposure to the research, verifies every factual claim against primary sources.
5. Evidence-to-disk — All research evidence is saved to disk before any data enters a deliverable; editorial verifies against files, not memory.

What this produces

For each condition, the pipeline delivers:

• Research package— a full evidence-classified write-up across BringThisIn’s research sections (Guide, Evidence, Comparison, Practice Gaps, Situation, Kit, Story), with verified PubMed citations behind every claim
• Source registry— a structured catalog of every article evaluated, with PMID, evidence tier, section mapping, and access status
• Gap report— honest documentation of what evidence exists, what is missing, and why
• Source article archive— every full-text article used, preserved on disk
• Audit trail— the complete evidence chain from initial query design through final editorial verification

What this does not do

• No human medical professional reviews this content for clinical accuracy
• AI outputs may contain inaccuracies; always confirm with your healthcare provider
• The Concord Method improves AI accuracy through cooperative pair verification but does not replace clinical judgment

Read the full methodology

Want a printable version for your appointments or advocacy work?

The Concord Method badge

Wherever you see the Pair-Reviewed · Concord Method badge on BringThisIn, it means the content was produced and verified using this methodology.

How it works

Multiple independent AI agents work in coordinated teams, each with a specific role:

• Research Pair— Two specialized research agents independently analyze published medical literature. They compare interpretations, flag uncertainty, and only agree when the evidence supports it.
• Writing Team— Separate agents translate research findings into patient-accessible language, maintaining citation accuracy throughout.
• Editorial Review— Independent editorial agents verify that the written content accurately represents the source research — no claims without citations, no softening of inconvenient findings.
• Verification Swarm— A final pass by multiple independent agents, each checking specific claims against source papers. Every citation is traced back to its PubMed source.

What makes it different

The Concord Method is built on cooperative agent pairs, not adversarial review. Our agents work in cooperative pairs through every research phase: no agent works alone. They share the same goal — accurate, evidence-grounded research — and they hold each other accountable through a formal epistemological marker system (C2C).

The markers make uncertainty structurally visible: every claim must expose whether it’s verified against a primary source, inferred from secondary signals, or still open as a question. Because the markers are visible to every agent in the pair, no agent can present an uncertain claim as a verified one without the other agent seeing the gap. This is cooperation with built-in honesty tools, not adversarial review.