How "first licensed" is derived
Three chained public sources. No PHI.
1 · CredentialStream
Provider roster + NPI + primary office (cs_mirror.db sync).
2 · NPI Registry API
npiregistry.cms.hhs.gov → enumeration date + license # + state (primary taxonomy).
3 · State Dental Board
License # + state → board verification portal → original issue date = first licensed.
Coverage status. All 35 license states were scraped. 33 boards are live — 514 of 527 first-licensed dates pulled, each carrying the board's own literal date string as evidence. CAPTCHA-walled boards (GA, MS, CA, SC, UT, TN, MD) were unlocked via a paid 2Captcha solver or via residential-browser manual entry (UT, TN). Two boards remain effectively blocked: CT exposes no original-issue field (only renewal grant date), and OK's four are not on its public active-dentist list. No national source exists — every board is a separate portal (ASP.NET, Salesforce/Angular SPAs, Accela, Thentia, GLSuite, MyLicense, state open-data APIs), so each live board is its own adapter.
⚠ Known limitation — multi-state careers. The pipeline pulls the doctor's primary license state from NPI Registry, then asks only that state's board for the original issue date. If a doctor practiced for 20 years in one state, then moved and now flags the new state as their NPI primary, this dashboard will report the later date — understating actual clinical tenure. 75 of 538 providers practice in a different state than their primary license, so this risk is real and measurable. Treat "First Licensed" as a near-true floor, not a career-tenure score, for any provider whose office state differs from their license state (flagged on the Action Board and with a MEDIUM confidence chip in the detail rail).
Anti-fabrication guarantee. A wrong licensure date is worse than a blank in a credentialing file. Every shown date was parsed from the board's actual HTTP response and is stored with the literal board string + source URL; the aggregator drops any date lacking that evidence. Blocked or absent providers are shown blank, never guessed — and the NPI enumeration date is never used as a stand-in.