{
 "url": "https://agenticweb.wiki/methodology/verify-brief/",
 "title": "Verification brief · agenticweb.wiki",
 "description": "The fixed brief every research pass on this map is run against.",
 "markdown_url": "https://agenticweb.wiki/methodology/verify-brief/index.md",
 "markdown": "# Verification brief\n\nThe fixed brief every research pass on this map is run against, unedited.\n\nYou are verifying protocol records for agenticweb.wiki, a reference map of the agentic web's standards layer.\nStep: {STEP}. Records to verify (ids): {IDS}. Do NOT edit data/protocols.json or data/resources.json directly — nine agents run concurrently on this data. Instead write data/verify/{STEP}.json — a JSON array containing the FULL updated record for each of your ids (same shape as the matching entries in data/protocols.json, with your edits applied) — and data/verify/{STEP}-resources.json — an object keyed by id with the five resource lists (implementations, sdks, validators, registries, adopters) for your ids only. Vocab: data/vocab.json — every value you write for step/parties/form/track/scope/phase/implementation/adoption.level/evidence.label MUST be an id from it.\n\nFor EACH record:\n1. Open spec_url with firecrawl_scrape. If it is not the canonical spec, find the canonical spec (repo, datatracker, vendor doc) and replace spec_url. Record the version label and its date.\n2. Fill: question (one line, the reader's question), analog (a classic-web thing, or \"none\"), origin.org + origin.date (YYYY-MM or YYYY-MM-DD), governance (today's steward, exact name), native_status (the issuer's own words), phase, implementation (name the implementations in summary_md), collisions (other acronyms it clashes with), aliases.\n3. probe: method, path (with {origin} or {endpoint} placeholders), parse (what field/header proves presence), on_absent (what absence means, one line). Must be executable by curl or a JSON-RPC call.\n4. discovery.mechanism and discovery.path (fixed path, header, DNS name, or \"none\").\n5. adoption: level from vocab; if \"claimed\", by = who claims; if \"measured\", probe_run = a probe file name that exists; else null. Do NOT upgrade to measured without a probe file.\n6. evidence: at least two entries, each {claim, label, url, date}. label VERIFIED only if you read a primary source or two independent secondaries yourself. Prefer sources after 2026-03-01. Never invent a URL; every url must be one you fetched.\n7. summary_md: 200–600 words, neutral, no marketing language, in this order: what it is, what it solves, what it does not do, current state with dates, who implements it, what is disputed or unknown. Cite inline as [n] matching evidence order.\n8. last_verified: today (YYYY-MM-DD). history: append {date: today, note: \"verified against <source>\"}.\n9. resources data for this id: implementations, sdks, validators, registries, adopters — each {name,url,date,evidence} with a fetched URL. Empty lists are fine if nothing exists; say so in summary_md.\n10. If a record should not exist (not a distinct spec), do not delete it: set phase \"withdrawn\", explain in summary_md, and report it.\n\nBefore finishing, run uv run python scripts/merge_verified.py --check {STEP} and fix any error it prints.\nReport: ids done, ids with open questions, any records you believe belong on a different step.\n"
}