The schools that actually graduate therapists

Hand & Stone runs 651 massage and facial spas across the United States, and every one runs on licensed massage therapists. Getting enough good ones in the door is a big problem. The most direct source is the schools that train them.

Its team had built a list of massage and esthetics schools by hand, thin on detail and probably missing schools. They wanted it longer and deeper: the schools they had missed, eleven researched fields for each one and a priority score that says where to start.

That is research at a scale no team does by hand. And a confident wrong answer sends recruiters to the wrong door.

We paired research agents with a deterministic spine. Agents search and read. Plain code does everything that has to be exact, so no model grades its own work.

• Discovery agents work state by state, 51 of them across all 50 states and D.C., each shown the schools already held so it hunts for new ones.
• Federal completions data and state training provider lists add schools vetted by someone else. Code checks every find against Hand & Stone’s list by fuzzy name.
• A light screening agent asks one question in two searches: does this school teach massage at all? It ruled out 835 candidates before any deep research.
• Research agents investigate each school that passes, four at once, returning a strict structured record with a source for each figure. Finished schools save instantly.
• A priority score in ordinary code: graduates 35%, licensing exam pass rates 20%, geography near Hand & Stone’s spas 20%, reachability 15% and accreditation 10%.
• Missing data is never scored as zero. Scores rescale over what is known. Below 45% of the weight known, a school goes unranked instead of guessed.

We checked the agents without asking a model. A separate step fetched every page each record cited and looked for the school’s name and the figures themselves. The right school was named on 1,230 of 1,239 checkable records. 71% of the figures appeared on the page too.

Tuition is the weak spot, with 367 of the 453 misses. Schools quote cost per credit, per clock hour or per program. The workbook says so plainly and tells the team to open the source before quoting it.

Where research disagreed with a federal filing on graduates, the filing won. Usually the agent had taken a lifetime total for a yearly rate. Rereading the page behind each screening verdict reinstated three schools the agent had wrongly ruled out.

• 2,424 agent runs and 11,155 web searches, with every call metered by stage.
• 421 tests, including ones that kill a run mid-write and confirm finished work survives.

• Only 74 schools have a licensing exam pass rate and 408 have a graduate count. 135 of the 201 top-tier schools rank without a graduate count, each flagged in its row.
• Discovery only finds what search can reach. A state list published as a PDF behind a form will not be fully harvested in one pass.
• Distances use the center of each zip code, a mile or two off. That does not matter at a 30-mile radius.

• An agentic research system: discovery, screening and research agents on a deterministic spine
• Hand & Stone’s own spreadsheet, extended with 29 new columns and 572 new schools
• A program check on every school, separating real massage pipelines from schools with nothing to hire from
• Model-free verification of every cited page, plus separate confidence and applicability scores
• A priority score in plain code, with the rubric written out and every weight adjustable

Technologies

• Claude with web search
• Python
• IPEDS
• State training provider lists
• Census Gazetteer

• Generative AI
• AI & Machine Learning
• Data & Analytics