BEN Review Engine
Architecture & FAQ

When a CE submission doesn't match anything in the Known Programs database (Tier 4 — Unknown), BEN runs a structured research sweep before flagging it for human review. The goal is to exhaust the automated research so the Director and Committee are spending their time on genuine judgment calls, not Googling.

The sweep works in layers:

• Layer 1 — Direct web search: BEN searches for the program/event name, year, and location. It's looking for an official event website, registration page, agenda PDF, press coverage, or a sponsoring organization's page. If it finds an official agenda, it cross-references session titles, speakers, hours, and dates against what the holder submitted.
• Layer 2 — LinkedIn: Conference organizers and speakers almost always post about their events. Attendees post recaps. This surfaces legitimacy signals even when there's no formal website.
• Layer 3 — YouTube: Many industry conferences post recorded sessions. If BEN finds a recording of a claimed session, that's strong corroboration — and it can pull title, speaker, and approximate length from video metadata.
• Layer 4 — Facebook / other social: Less reliable, but useful for smaller regional events. Event pages, attendee check-ins, and post-event recap posts can confirm an event existed even without a formal web presence.

BEN doesn't make the approval decision — it produces a research summary card for each Unknown submission with whatever it found: source URLs, event description, agenda match confidence, and a plain-language note. The Director and Committee make the final call.

After each cycle, every newly-approved Tier 4 program is automatically promoted to Tier 3 (Known), so the Unknown pool shrinks year over year.

Every CE submission gets classified into one of four tiers. The tier determines how BEN handles the submission and how much human review it needs.

• Tier 1 — Auto-ApprovedATFA-affiliated programs (TAF, TAI). BEN recommends approval automatically. No individual Committee review required — ratified in bulk.
• Tier 2 — ReputableNationally recognized industry organizations (ABA, Cannon Financial Institute, FIRMA). BEN matches the submission against known session metadata and recommends approval when verified.
• Tier 3 — KnownPrograms previously approved by ATFA in prior cycles. BEN matches against historical records and recommends approval. This database grows automatically after every cycle.
• Tier 4 — UnknownNot in any existing database. BEN runs the web research sweep and produces a summary card. Director and Committee review individually.

Classification is based on program name matching — exact match first, then fuzzy match for typos and variations. BEN flags low-confidence matches for Director review rather than auto-classifying incorrectly.

The v1 stack was chosen for simplicity, low cost, and ease of handoff — not for technical elegance. The goal is a working system in Year 1, not an over-engineered one.

• CE App (myceapp.com): Retained for holder-facing submission. Has bulk CSV export. API capabilities are being assessed — if available, the manual export step could be eliminated.
• Google Sheets: Serves three roles: staging area for submissions, the Known Programs database, and the Committee review surface. Chosen because it's already familiar, supports View+Comment sharing, and creates an audit trail via comments.
• Claude (Anthropic) with web search: Powers the BEN Review Engine. Classification logic, data cleanup, Tier 4 research, and recommendation report generation all run through Claude. Configured as a persistent Claude Project so the Known Programs database doesn't need to be re-uploaded each run.
• Gmail: Outbound holder communications in Year 1 are Director-reviewed drafts. Future state: fully automated sends triggered by BEN after Committee approval.
• Netlify + atfacertification.com subdomains: Static HTML pages for internal documentation and published read-only views of key data. No CMS, no backend — just files.

Each layer can be upgraded independently. The Google Sheets database could migrate to Airtable or Supabase. The CE App dependency could be replaced with a lightweight custom intake form. None of those upgrades require rebuilding the others.

The Committee review is designed to be asynchronous — no meeting required unless a submission warrants real discussion.

• What Committee members receive: A View + Comment link to a Google Sheet containing BEN's recommendation report. The sheet is color-coded: Tier 1 auto-approvals are pre-marked, verified Tier 2/3 matches have a recommended approval status, and Tier 4 unknowns have BEN's research summary card attached.
• How to respond: Committee members add a comment to any row where they want to flag a concern, ask a question, or formally approve/deny. No edits to the sheet itself — comments only.
• What constitutes approval: A formal approval decision is recorded when the Director confirms consensus from the Committee comments. The Director has final authority to close any open items.
• When a call is needed: If a submission is genuinely contested — unusual program, disputed hours, credibility question — the Director escalates to a brief synchronous call. These should be rare.
• The audit trail: All Committee comments are timestamped and attributed in Google Sheets. This creates a defensible governance record without any additional documentation overhead.

The Committee's workload should decrease each year as the Known Programs database grows and the Tier 4 Unknown pool shrinks.