The Software Layer.

The problem

The single most expensive touchpoint in home health is the first visit. A nurse sits with a new patient and fills roughly a hundred OASIS fields; the documentation happens outside working hours, which is how nurses leave — and agencies pay for that churn in overtime, missed visits, and failed quality scores. №008's care spending problem lands on this screen first.

Who pays

The agency, from the reimbursement and quality budget line. Pricing is per-clinician-per-month; the wedge is measurable in avoided overtime and faster claim submission. ~11,353 agencies, 83.5% for-profit — a buyer base that already buys software and has no loyalty to the incumbent once pricing moves.

The build

Ambient capture at point of care (in the home, no hardware) that populates the visit form in the agency's existing EHR, with a clinician review path and QC before submission. The defensible version is workflow, not transcriptions: routing, coding integrity, audit trails on OASIS items that drive reimbursement.

Kill conditions

WellSky or Netsmart ships the same workflow natively and bundles it at zero marginal cost; or the EHR-vendor lock-in proves the form population is locked to one EHR and the install cycle exceeds 18 months.

The problem

The largest care workforce in the United States — 59 million people, $1.01 trillion of economic value — has no systems of record. Medications, appointments, doctor contact, sibling handoffs: all in group texts and one person's head. This is the labour gap of №008 from the demand side: professional care workers are leaving faster than they are hired, and families are absorbing the difference.

Who pays

Not the family. The graveyard says so. Three buyer candidates exist, in order of budget line: employers (caregiver benefits — the family caregiver is frequently their own employee), payers (Medicare Advantage / Medicaid MTM programs that now have caregiver-support lines and 2025–26 payment shifts), and insurers of long-term-care products. Sell to the entity that carries the actuarial exposure to the collapse.

The build

A coordination layer, not a companion: shared care record, medication and appointment state, visit notes from agency staff flowing into the family view, escalation path to 211/911, and a dashboard for the payer showing avoided ER events. The product is boring on purpose. Companion apps die; claims-adjacent coordination software compounds.

Kill conditions

Medicare Advantage standardizes caregiver-data requirements and the EHR vendors (Epic/Oracle) absorb the family layer as a feature; or payer adoption stays below a handful of plans after two contract cycles.

The problem

№008's point that immigration is the only lever just became a pricing problem. A new H-1B petition for a worker outside the US now carries a six-figure entry cost on top of legal fees; the cap is still 85,000; and each quarter a new rule, fee, or court ruling reshapes the cost surface. The question every employer now asks — "which visa, which country, which route, at what total cost?" — has no software answer. It is answered by lawyers, case by case, at case-by-case prices.

Who pays

The employer (talent/compliance budget) and the immigration law firm (cost of case). Two-sided: a routing API for HR systems (given role, candidate, offered wage → ranked viable categories including O-1, L-1, TN, E-3, EB-NIW, and hire-in-country cost, with the $100K fee math and lottery-odds adjustment) plus a case-ops console firms will actually run. The policy chaos is the moat: the change frequency favors a software-native operator over service-heavy incumbents (OnGlobal, Topia, Envoy, eimmigration, Boundless).

The build

A category-and-country decision engine with a documented cost model per route, maintained against agency announcements, plus document automation for the surviving routes. Lighthouse ($7M seed, ~$7M raised, AI-automated paralegal work at six-figure engagements) proves the services side; nobody owns the routing math itself. That math is the product.

Kill conditions

The fee is struck down and the lottery reverts to random (demand persists, pricing compresses a 3–5x); or a single services incumbent (Fragomen/Envoy-class) ships the routing model inside their portal and the pure-software position has no margin. Note the asymmetry: the bet survives both policy outcomes flat or up — that is the convexity that clears test 3.

The problem

№008's local-finance problem has a named dataset in Japan. Nine million empty homes, most of them detached houses in shrinking towns whose owners died without heirs or with disputed heirs. The 2024 inheritance-registration law is a forcing function: every inheritance becomes a registration event inside a three-year window, and the window's first penalties arrive around 2027 — a scheduled, dated wave of owners who must touch the system or pay.

Who pays

Municipalities (subsidy line: akiya-bank administration and demolition grants), estate professionals (ji-toshi, lawyers — per-estate fees), and redevelopment investors (pipeline diligence fees). Thin wallets, fragmented buyers, and a legal gatekeeper (fixed-office holders control registration). This is the convexity file: the thesis survives if the market is bigger or smaller than estimated, because the data asset is built either way.

The build

Title-chain reconstruction from registry data + inheritance events, an estate-status pipeline per dwelling (owner dead / heirs unknown / registration pending / penalty due / demolishable / convertable), and a municipal console for akiya-bank operations. The moat is the graph: dwelling → ownership → inheritance → penalty date. Nobody has it; the law is now forcing the raw material to surface.

Kill conditions

The national registry digitization lands (a government system absorbs the data layer), or estate attorneys refuse software in a market where relationships replace contracts. Monetization is the weak edge here: B2C is structurally weak in Japan and the B2G path is slow. Size the position accordingly.

The problem

№008's sector shortage has a pipeline in Japan, and the pipeline runs through software that does not exist. Each care facility that hires a foreign care worker must run a mini immigration operation: language support, contracts, residence-status renewals, retention, and now (since April 2025) a mandatory multilingual support obligation with local-community coexistence duties. The government is subsidizing the pipeline to triple it inside five years — the budget line exists; only the system layer is missing. Today the pipeline runs on brokers, paper, and the facility's own part-time clerk.

Who pays

The receiving facility, from the government's care-worker recruitment subsidy line (subsidy = budget, documented above), and the staffing brokers who will adopt the ops tool before they are forced to change. Pricing per foreign worker per tenure — a model that scales with the government's own target (44,367 → 135,000 workers in five years).

The build

A facility-side operations console: immigration document state per worker, renewal deadlines, multilingual support delivery log (evidence the 2025 obligation is met), retention tracking, and broker integration. The buyer list is public (registered "accepting organizations"); the regulatory event (Apr 2025) is the forcing function; the target (2029) is dated.

Kill conditions

Fragmentation wins — 500,000+ small facilities never centralize, churn eats the base, and the subsidy budget shifts to direct worker payments rather than employer operations. Or the government ships its own portal (it has the SSW support site; the depth gap is real but close).

The problem

The problem itself is real — №008 problem 5, the local-finance wedge, is the earliest and least-publicized hit. But the software translation fails the method at test 1 and test 2. The buyer cannot name today's budget line for "our town shrinking," the pain is on a P&L horizon beyond any municipal electoral cycle, and the sale is relationship-and-customization against a public tender process. This is where a founder who mistakes novelty for value goes to die, so it is written down where it can be found.

The one version that survives

Sell the index, not the OS: a dated, citable shrinking-town dataset (houshold decline, service closures, tax-base deltas per municipality) sold to insurers, developers, and the state-level agencies that actually write multi-year budgets. A data product with a public-facing citation loop; no implementation project, no procurement theater.

Kill conditions / the honest ones

Even the index version is a niche. Marking it as the trap is the file's content: the wedge is real, the software business is not. Files F and D (akiya) are the same wedge — D survives because Japan's law forced a dated, bounded dataset with a named professional buyer; F dies in the general case because nothing dated is forcing the buyer to buy.