What changed
FACT: EPA finalized the Lead and Copper Rule Improvements (LCRI, Federal Register 2024-23549), compelling drinking water systems to inventory and replace lead and certain galvanized service lines. FACT: Context.dev-style schema extraction now returns structured data from arbitrary public websites via one API call, removing the per-county custom-scraper cost that made records research an engineering-firm archival job. FACT: EPA is funding technical-assistance intermediaries for rural/small systems (Grants.gov 362798), meaning subsidized buyers exist for exactly this segment.
Why now
The LCRI is final and binding: inventories must be maintained and updated, replacement plans filed, and lines classified as 'unknown' must be resolved on the way to full replacement, with obligations running through the 2030s. Small systems have the largest unknown-line backlogs and the least staff. The capability side (schema extraction + LLM classification of permit text) just collapsed the cost of the exact task — mining undocumented county permit/assessor/tap-card records — that consultants bill for by the hour. Estimated ~12-month edge before state TA programs and engineering firms standardize equivalent tooling (hypothesis).
Converging signals
(1) Regulation: LCRI final rule creates a defined, compelled class — every community water system with lead/galvanized/unknown lines — with named submissions (inventory on the state's template, replacement plan) [federalregister.gov]. (2) Capability: single-API structured extraction from arbitrary public sites makes bulk mining of county records feasible without scraping infrastructure [context.dev]. (3) Money: EPA-funded TA pipeline to small/rural systems proves public money is flowing to help this exact segment comply [grants.gov/362798]. Rule + filer class + newly cheap evidence pipeline is a genuine three-signal convergence.
Customer pain
Small systems (often one part-time operator) must produce and annually update a line-by-line materials inventory with evidence, and resolve thousands of 'unknown' classifications. Today that means hand-searching county permit books, tap cards, and assessor records — work they can't staff and currently buy from engineering firms at archival-research rates. HYPOTHESIS (no complaint threads in input): the pain is inferred from the mandate structure and staffing reality of small systems, not from captured complaint signals.
Who pays
Primary: small community water systems (roughly 500–10,000 connections), paying with SRF/BIL lead-service-line funds and compliance budgets. Secondary and likely the better wedge: EPA-funded TA intermediaries (RCAP affiliates, state rural water associations) and the small engineering/consulting firms serving dozens of systems — a white-label records-mining engine makes their fixed-fee TA dollars go further. The beneficiary (the utility) and the buyer (often the intermediary) may differ; both are named and reachable.
Solved today
Engineering consultants do manual archival research billed hourly or as a per-system project (commonly five figures); some states offer free statistical predictive tools and templates; larger systems buy platforms like 120Water or Trinnex leadCAST; many small systems simply mark lines 'unknown' and defer — which the LCRI no longer permits indefinitely.
Why current solutions are bad
Consultant archival research is priced for the job's former difficulty, not its current one. Incumbent platforms are program-management suites priced and designed for mid/large utilities, not a 2,000-connection system that needs one deliverable: classified lines with citable evidence on the state's template. Free state tools predict materials statistically but don't produce the per-line documentary evidence that survives a state audit.
Proposed product
A per-system service (tool-assisted, not self-serve at first): utility provides its billing address list; the engine runs schema extraction over the county's permit/assessor/tap-card portals, LLM-classifies material evidence per address, links each classification to its source document, and emits (a) the state-template inventory rows and (b) the LCRI replacement-plan skeleton with prioritization. Charge a flat per-system fee far below the consultant quote, plus an annual update subscription — the LCRI's annual update duty makes this recurring.
MVP version
The kill test IS the MVP: one cooperating utility, 200 addresses, one county portal. Measure the fraction of lines classified with citable evidence versus manual review. ≥40% auto-classification with evidence links beats the consultant baseline; under 40% kills or narrows the play. Deliverable doubles as the sales artifact for every neighboring system in the same county (records source is shared).
30-day build
Recruit 1–2 pilot utilities via a state rural water association or RCAP affiliate; run the 200-address kill test on their county; validate against the state's inventory template requirements; price against the utility's actual consultant quote.
60-day build
If ≥40%: productionize the county-portal extraction recipes, add the evidence-link citation format states accept, deliver both pilots as paid engagements ($3–8k each, hypothesis on price tolerance), and package results into a one-page proof for the TA intermediary channel.
90-day revenue plan
Sell county-by-county: every utility in a covered county is a warm prospect because the records source is already mapped. Target 5–10 paid systems directly plus one white-label agreement with a TA provider or small consultancy; realistic first-revenue window is 90–150 days given pilot lead time — within the founder's funded ramp.
Distribution path
State rural water associations (training events, newsletters), RCAP/EPA TA intermediaries, state drinking-water program contact lists (public), and direct outreach to systems with large published 'unknown' counts (state inventory summaries are public records — a self-generating lead list that fits the founder's public-records strength).
Pricing hypothesis
Flat per-system inventory-resolution fee (e.g., $2,500–$10,000 scaled by connection count) vs consultant quotes typically multiples higher; $500–$1,500/yr annual-update subscription; white-label per-system rate for TA providers and consultancies.
Technical difficulty
Moderate. Extraction API + address matching + LLM classification is squarely in the founder's automation/AI-workflow strengths. The real difficulty is variance: county portals range from clean databases to scanned PDFs to nothing online. Expect a meaningful fraction of counties to be unminable — the model must survive selling only where records are good, which the county-by-county go-to-market accommodates.
Legal / regulatory risk
Low-moderate. Public records are lawfully scrapable; no licensure to prepare inventory data a utility files itself (the utility remains the submitter of record). HYPOTHESIS: regulatory risk that the current EPA revisits LCRI provisions — litigation and reconsideration noise exists — but the replacement mandate direction has so far been retained; a delay would stretch, not erase, demand. Do not represent classifications as engineering certifications; some states may require a PE stamp on the replacement plan itself — check per state and partner with a PE where required (flagged in kill arguments, not assumed fatal).
Platform dependency
Context.dev is one vendor, but the capability (schema extraction) is commoditizing across several providers and could be rebuilt in-house; no platform owner can deplatform a tool that reads public records. Low dependency.
Founder fit
Very high. This is structurally the FMCSA ELDT play: a federal rule compels a defined class to produce a submission, and the founder builds the evidence/filing layer and charges per system. It additionally stacks his public-records skill, industrial-operations credibility (small utility operators respect operational fluency, not degrees), and demonstrated-value sales style — the 200-address pilot IS the demo.
Breakout potential
49,000+ small systems, obligations through the 2030s, annual updates as recurring revenue, and the county-records-mining engine generalizes to adjacent verification products (septic, backflow, well permits) once proven. Replicable state-by-state like his thesis prefers.
Final recommendation
PURSUE, gated hard on the kill test. The mandate is real, final, and durable; the filer class is defined; public money is flowing to the segment; and the shape matches the founder's proven ELDT pattern almost exactly. The two honest unknowns — county record minability and whether the free-TA path crowds out paid tools — are both resolvable in 30 days for a few thousand dollars. Run the 200-address pilot before writing any product code beyond the pipeline.
Next action
Contact one state rural water association or RCAP affiliate this week to recruit a cooperating small utility, then run the 200-address county-records kill test and measure the citable-evidence classification rate against the 40% threshold.