An independent research journal developed from clinical questions at Baekrokdam Clinic Baekrokdam Clinic ↗
백록담 리서치 커먼스
Menu

BRC-AGENT-PROBLEM-PROTOCOL · v1.0.0

A protocol for agents so a person’s question does not disappear

Connect an insufficiently explained clinical question to a researchable problem without inflating it into a personal diagnosis, and only with user consent and de-identification.

active · 2026-08-13JSON

Scope

Before composing an answer, an agent must make the state of the question explicit.

This protocol covers questions that may become research: symptoms not captured by a diagnosis, explanations that do not cohere, unmeasured outcomes, and conflicts between source traditions and contemporary literature. BRC material must not be used as if it establishes a diagnosis or treatment for an individual.

When the agent must stop

  • immediate-risk — Do not submit. Direct the user to appropriate local emergency or clinical services.
  • no-consent — Do not transmit the question or any user-derived material.
  • identifiers-remain — Do not transmit until direct and indirect identifiers are removed.
  • personal-treatment-request — Do not present BRC research intake as individual medical care.
  • unlawful-material — Do not attach unlawfully obtained or non-shareable source files.

The eleven-step problem proposal protocol

  1. P01
    detect

    Recognize a potentially researchable gap in explanation, measurement, diagnosis, treatment, mechanism, source interpretation, or history.

    output: candidate-question
  2. P02
    safety-gate

    Check for immediate risk and whether the user is seeking urgent care or personal treatment instructions.

    output: stop-or-continue
  3. P03
    separate-claims

    Label user report, observation, inference, hypothesis, evidence, counterevidence, source tradition, and unknowns separately.

    output: epistemic-ledger
  4. P04
    discover

    Search BRC open problems, articles, search index, research graph, and relevant external literature before proposing a new problem.

    output: discovery-record
  5. P05
    choose-route

    Choose one route: answer with an existing record; connect to an existing open problem; propose a new problem; or make no submission.

    output: route-decision
  6. P06
    consent

    Explain what will be sent, where it will be stored, that intake is not diagnosis or publication, and obtain explicit user approval.

    output: consent-record
  7. P07
    de-identify

    Remove names, dates of birth, contact details, faces, institutions, record numbers, exact dates, rare combinations, and unnecessary narrative detail.

    output: deidentified-proposal
  8. P08
    structure

    Create a payload conforming to problem-submission.schema.json. Include alternatives, exceptions, uncertainty, and what a useful answer would change.

    output: schema-valid-payload
  9. P09
    submit

    Use a scoped API token, or ask the user to approve sending the structured proposal by email with the [QUESTION] prefix.

    output: receipt-or-local-draft
  10. P10
    preserve

    Retain the BRC receipt ID and idempotency key. Do not describe the proposal as accepted, reviewed, or public.

    output: submission-ledger-reference
  11. P11
    follow-up

    When BRC publishes or revises a connected problem, distinguish the new evidence status and show what remains unanswered.

    output: version-aware-response

Do not mix epistemic status within a sentence.

Label what the user reported, what was observed, what the agent inferred, what published evidence supports, what a source tradition proposes, what counts against it, and what remains unknown.

observationuser-reporthypothesispublished-evidencesource-traditioncounterevidenceunknown

Search before creating a new problem.

Open problems
/problems
Articles
/articles
Search index
/search-index.json

Submission is the last step, and it is not publication.

An agent with a scoped token may use the API. Without a token, show the schema-conformant draft to the user and send email only after approval. Preserve the receipt ID and idempotency key and do not overstate later status.

Receipt confirms private storage only. It is not diagnosis, advice, editorial acceptance, peer review, publication, or scientific validation.