# State research methodology v1 — proposed for owner approval

Status: PROPOSED, NOT APPROVED. Prepared September 5, 2026. This proposal replaces the wage-only localization approach for future business-state research. It does not certify or revise the current reference financial scenarios. The owner must approve this version before the state research rebuild starts. Approval must identify this file and its content hash; editing an approved methodology requires another version and approval.

**Decision requested.** Approve a system that researches and models each business-state pair independently, presents the results and their derivation, and rejects an indexable article when material local evidence or editorial review is missing. Different results must arise from different evidence and operating conditions. Random differences, cosmetic rewriting and one universal state multiplier are forbidden. An equipment SKU or other genuinely national input may have the same value in multiple states when the same applicable source supports it.

**First review: is the evidence about the right business and place?**

Freeze the business format before comparing states: NAICS, revenue mechanism, premises type, capacity, operating hours, staffing duties, owner participation and excluded activities. Maintain comparable formats across states. If a state requires a different feasible format, publish it as a separate scenario and explain the difference; do not silently compare a kiosk with a full restaurant. The existing Austin full-service case is not a calibration of the state counter-service case.

Every state record must specify its geographic target. A state-wide reference is a documented combination of local cases, not one capital-city listing relabeled as a state average. Select a major market, a secondary market and a smaller-market or nonmetro stratum when those distinctions exist. Record counties/metros, the business-establishment distribution used to select them, their weights, excluded areas and selection bias. Use compatible industry establishment weights where available, not population as a proxy for customers. Calculate each local case first, then aggregate comparable outputs. Do not combine median rent from one place, maximum price from another and minimum wage from a third into an apparently representative case.

When only one locality has adequate evidence, publish a locality-specific case inside the state page. Do not describe its numbers as a statewide average. A weighted reference scenario is also not an empirical average startup cost or a forecast for an individual address.

Each material input carries: input ID, business and geography, value and units, status (observed, reported, derived, assumed or missing), observation period, retrieval date, direct source, exact table/row or captured offer, reuse constraints, definition, transformation, uncertainty and reviewer decision. An HTTP 200 response alone does not verify an input. A derived input also identifies its parent observations and formula. Material conflicts remain visible and prevent promotion until resolved or bounded.

**Second review: do costs and revenue each have their own evidence and reconcile?**

| Input family | Required basis | Forbidden shortcut |
| --- | --- | --- |
| Premises | Comparable current commercial asking rents, area, use/condition, base versus gross/NNN, recurring charges and deposit terms; normalize to annual USD per sq ft and then the chosen premises | Residential rents; one national rent; one listing called a state average |
| Fit-out | Scope and retained infrastructure, work quantities, geographically applicable contractor/professional cost evidence, connections and installation | Same unexplained lump sum across every state |
| Equipment | Comparable manufacturer/vendor configurations, freight, installation and applicable acquisition taxes; distinguish new, used and retained equipment | Mixing incompatible configurations to manufacture a range |
| Paid labor | Applicable occupation/industry geography, actual paid coverage, local wage floors and overtime rules; separately derived employer taxes, benefit and insurance assumptions | A universal 18% employer load labeled as a state payroll calculation |
| Utilities and insurance | Applicable tariff or observed benchmark plus consumption/coverage assumptions and a second plausibility check | Scaling every operating cost with a single price index |
| Permits and government fees | Issuing authority, actual activity, address jurisdiction, threshold, effective date, initial/recurring classification and conditional items | Treating an agency directory, proposed fee or one city's fee as a statewide requirement |
| Selling price | Current comparable price observations and an explicit product/service mix, discounts and realized net revenue basis | Calling a menu price an achieved average customer check |
| Sales volume | Demand evidence and a transparent derivation, independent sector cross-check, capacity and paid-service-hour limits | Population growth automatically becoming customer growth; capacity treated as achieved demand |
| Collections and launch | Evidence or an explicit sensitivity for invoice terms, ramp, retention and seasonality; cash timing separate from booked revenue | The same unexplained ramp for every business and place |

Default sampling rules for a researched state reference are at least eight usable premises observations when leased premises materially affect the case, and at least six comparable customer-price observations, distributed across the selected local strata. A quote must meet the format and definition; unusable evidence is not counted. These are minimum collection targets, not representative-sample guarantees. Report count, geographic distribution and bias. A smaller sample may support a narrower case but cannot silently pass as the intended state reference. A premises-free format documents why rent sampling is not applicable.

A material input is one capable, under a plausible change, of moving opening funding, monthly profit or break-even by at least 5%, or of reversing viability; mandatory permits and operating prerequisites are material regardless of price. Each material cost needs one applicable direct observation or defensible derivation and an independent cross-check. Reposts of the same upstream figure count once. If sensitivity is not yet known, treat all main cost categories as material until the model establishes otherwise. Do not suppress a high-impact missing input to obtain a better completeness score.

Revenue follows the business mechanism. Restaurant/coffee: paid transactions per day × operating days × realized average ticket. Cleaning: average active accounts × monthly net fee, reconciled to visits, labor hours, travel, acquisition and retention. Salon: completed appointments × realized service mix, constrained by paid stylist time, cancellations and rebooking. Detailing: completed package jobs × realized net package price, constrained by labor hours, vehicle mix and site capacity. No model may assign the same assumed volume and price to all states and present the result as researched state revenue. Reuse an input only when its scope and evidence actually support reuse.

Cross-check modeled stabilized revenue against the closest compatible published industry revenue/receipts measure, with industry, establishment size, employer status and period limitations. Revenue per establishment is a mean benchmark, not median achievable revenue for a new business. Census CBP and BLS QCEW supply business structure, employment and payroll rather than establishment revenue. Economic Census, applicable annual Census data or IRS industry data may support the financial cross-check. Currentize monetary benchmarks with an identified relevant index; do not inflate establishment counts. BEA regional price parities are context and plausibility checks, not universal business-cost multipliers. [CBP](https://www.census.gov/programs-surveys/cbp.html), [QCEW](https://www.bls.gov/cew/overview.htm), [Economic Census](https://www.census.gov/programs-surveys/economic-census.html), [BEA RPP](https://www.bea.gov/data/prices-inflation/regional-price-parities-state-and-metro-area).

Check the published geography of every source. For example, the current QCEW overview notes that from second-quarter 2025 its MSA presentation changed to totals only; use applicable county/state industry data or another defensible source rather than assuming a metro-by-industry table is available. Record official publication lags instead of calling old observation periods current.

Build opening payments line by line. Keep setup payments, refundable deposits, stock, training, pre-opening occupancy, contingency and operating cash reserve distinct. Model monthly operating costs, EBITDA, depreciation, EBIT, maintenance capital expenditure, receivables/payables/inventory changes and project cash. Paid owner work occupies a defined labor role exactly once. Owner compensation, passive-owner profit and distributions remain separate. Collected sales tax is not revenue; debt and income taxes must be separately modeled or explicitly excluded.

Calculate low/base/high operating cases from defensible price, demand, capacity, staffing, cost and ramp combinations. Startup lean/base/premium scopes are a separate dimension; do not confuse them with operating downside/upside. Show year-one and stabilized results. Payback uses cumulative cash and remains unavailable if not reached; do not replace that result with an arbitrary cap. Ranges are scenario ranges unless the evidence supports a statistical interpretation. Never label forecast confidence with an uncalibrated percentage.

**Third review: does the published page deliver and substantiate a useful answer?**

Every proposed indexable state article must pass all of the following, in addition to technical checks:

1. The heading, summary and main numbers match the researched format, geography, period and status.
2. Material opening costs, recurring costs and revenue have applicable evidence and derivations. No unbounded material assumption is presented as a researched result.
3. At least one consequential local finding explains what changes the decision relative to the common business guide. Different wording or numbers alone is not sufficient.
4. Sources support the associated claim, with meaningful conflicts, source lags and uncertainty visible.
5. Financial identities, capacity, payroll/owner treatment, cash timing and scenario coherence pass checks.
6. A real responsible publisher/contact and a qualified named reviewer or identifiable responsible team are recorded. The record states who checked evidence and financial interpretation, and identifies the exact content revision. Automated checks are labeled as automated.
7. The page gives its useful result, evidence boundary and strongest sensitivity before any commercial offer. Its financial disclaimer does not replace the research or the accountability.
8. A reader can find the answer, understand its limitations, inspect supporting evidence and use the calculation. Reader misunderstandings become revision tasks.
9. The approved methodology, source/input fingerprints and exact reviewed article version match. A later material change invalidates the previous approval.

A page can be ready for a limited, non-indexable benchmark demonstration while failing readiness as a local startup study. This distinction must be explicit in the internal release record and truthful in public copy. No word count, heading count, number of sources, varied wording or technical test total can substitute for the content decision.

**Results-first presentation.** The first screen or immediately following block contains: what the completed research found; the opening-cost, operating-cost and revenue result within its actual scope; the most consequential finding; a concise observed/derived/assumed breakdown; and a link to the relevant evidence and calculator. Deeper sections explain the completed investigation, rejected comparables, conflicting evidence and what changes the result. Reader-specific address or operating decisions can remain as next steps, but generic instructions to do the research cannot substitute for the article's research output.

Suggested financial disclaimer: “Published research and calculations support business planning and education. They are not personalized financial, investment, tax or legal advice, and they do not guarantee costs, revenue, profit or financing. Estimates depend on the stated format, location, source periods and assumptions. Check the requirements and commitments that apply to your circumstances.” The disclaimer must not imply completed research or expert review that did not occur.

**Source freshness and ongoing use.** Use the latest applicable official release and show its observation period. Recheck market price observations within 90 days and rents within 180 days; use stricter windows when volatility or a source notice warrants them. Check legal requirements and fee effective dates against the intended opening period at every material update. A source older than its working freshness window must be revalidated, replaced or explicitly excluded from a current-price claim. An official lag is disclosed and does not justify fabricating a fresh observation.

An update starts with evidence and affected inputs, then recalculates all dependent profiles, cards, comparisons and downloads, then obtains the necessary review. A wording change does not refresh a source date. A missing or stale material source blocks new indexable publication until reviewed. Preserve original inputs, snapshots and earlier versions in the internal evidence history.

**Execution after approval.** Create a separate business-state research contract that references the existing city-package contract without silently modifying it. Store business format, geography/strata, observations, derivations, assumptions, monthly scenarios, claim-to-source links, review decisions and dependencies per business-state pair. All state financial inputs must resolve through that record; the renderer must not fall back silently to shared commercial constants.

Begin with one business in three deliberately different state contexts to validate collection feasibility and the model, then review the method's outcomes before expanding the same approved workflow to all 50 states and the other businesses. The three-context pilot is a validation step, not a safe Google page-count threshold. No bulk research is certified merely because a generator emitted a set of pages.

The root AGENTS.md, publication policy and release checks must reference the approved methodology version. Approval is a deliberate owner action, never an automatic consequence of a passing build. If the method cannot be satisfied, retain the gap and stop promotion of that record rather than weakening the method without approval.