# Methods and evidence rules

## Scope and geography

The policy unit is the **incorporated City of Portland**, including city areas outside Multnomah County. County data include people and firms outside city limits. The Portland–Vancouver–Hillsboro MSA crosses Oregon and Washington and contains multiple cities and counties. Each extract retains its native geography; no county or metro total is relabeled as Portland city. OSB information is city program activity, and its clients are a selected subset, not a city business census.

Prosper Portland's 2025 indicators provide a reported Portland-city small-business job series through 2024, sourced to a custom Oregon QCEW tabulation. Figure 1.06 gives two exact endpoints but does not publish the underlying year-by-year table. This package transcribes only those exact endpoints. The report describes size as 1–20 employees on the figure page but 1–19 elsewhere, and does not resolve whether “firm” means QCEW establishment-size class. This city series is retained **separately** from SUSB enterprise-size shares; its report source is hash-pinned and the ambiguity is carried in the caption and ledger.

The evidence cutoff is September 30, 2026. The year in each file is a **measurement** year, not the retrieval year. The main employer-size cross section is 2022, county nonemployers 2023, county jobs trend 2019–2025, OSB's first year approximately May 2025–May 2026, and BDS dynamics through 2023. No false common-year total is constructed. Comparison peers are eight preselected MSAs; this is a peer table, not a global or national ranking. Industry mix, business age, labor markets, sector concentration and regional boundaries can confound all simple peer comparisons.

## Definitions

SUSB sorts a firm's establishments by **enterprise employment**, including employees at establishments outside the region. Thus its `firm` count means employer enterprise with activity in the geography, and its metro or county employment is employment **at establishments in that geography**. A single enterprise can span jurisdictions. The `<500` and `<20` rows are overlapping subtotals. The nonoverlapping metro bins are `<5`, `5–9`, `10–19`, `20–99`, `100–499`, and `500+`. The county workbook uses `<20`, `20–99`, `100–499`, `500+`; county `<500` is the first three. The `<20` threshold best approximates local small teams, while `<500` follows the conventional federal threshold. Neither identifies local ownership, franchise status or independence. Source-native bands are retained in CSVs.

SUSB employment is for the mid-March pay period; payroll covers the year and receipts are gross revenue. `annual_payroll_per_job` divides annual payroll by mid-March jobs; it is **not** average worker earnings, hourly pay, benefits or causal size effect. SUSB values have Census disclosure-avoidance/noise flags, retained in extracts. Suppressed cells are kept empty rather than imputed. NES counts tax-reporting businesses with no paid employees; mailing or tax addresses may differ from business operating sites. NES receipts are gross before expenses. Nonprofit coverage differs across sources. Registration rolls are not an operating-business denominator.

QCEW tracks unemployment-insurance-covered establishments and annual-average employees, including many nonprofits but not most proprietors. Extracts select Multnomah County FIPS `41051`, private ownership code `5`, annual `A`, size code `0`. Disclosure-suppressed jobs and wages are blank. QCEW local job trends cannot isolate city effects or business size. Source NAICS vintages and revisions need checking before finer industry comparisons; broad two-digit sectors are used here.

BDS 2023-revision metro tables show establishment entry and exit, job flows and parent-firm age. Establishment entry includes a new location of an existing firm; it is not synonymous with firm birth. The firm-age-zero category identifies new employer firms, but categories include a large **left-censored** group whose establishment ancestry predates the source panel. BDS is an employer-establishment universe and cannot supply a Portland city cohort survival rate here. Its entry/exit rates use an average-establishments denominator in adjacent years, not the current number of firms. The pinned metro and age CSVs are separately extracted.

## Calculations and verification

`build.py` loads the revised 2022 Census MSA SUSB text file and 2022 county SUSB workbook, pins their SHA-256 hashes, checks nonoverlapping size-bin firm, establishment, job, payroll and receipt totals against published all-size rows, and emits machine-readable tables. Metro `<500` shares are `ENTRSIZE 08 / ENTRSIZE 01`; `<20` shares use `05 / 01`. County `<500` is bins `2+3+4`; the labels and all-size totals are checked. The county `<20` row is bin `2`. Charts draw only from these extracts or explicit page-4 OSB percentages and page-2/4 district counts.

Derived shares are displayed to one decimal percent; source data retain original integer units. `data/checks.json` records the pinned hashes and key reconciliations. Figures 12–15 summarize the OSB report and are **program-client** distributions. They must never be read as city industry, demand or effectiveness distributions. OSB page 2 lists district counts 103, 182, 154 and 142, totaling 581, whereas page 4 reports 759 unique businesses. The report does not reconcile the difference. The same report uses 1,600 touchpoints and 1,700-plus interactions without a published crosswalk. Both issues remain unresolved.

A source status of `screened` means located and triaged, not relied upon as conclusive evidence. `archived-reviewed` means a pinned raw file was extracted or the report was read; `reviewed` means directly inspected; `queued` means next-stage extraction. The crawl log records HTTP errors, access failures, hashes, content type and retrieval; errors are not silently treated as source absences. Private records and owner interviews are proposed only in the fieldwork kit.

## Claim and confidence discipline

`claims.tsv` gives the source ID, exact row or page, evidence level and limitation for each consequential present finding. In the dossier, **high** means directly reproducible published data with definitions checked; **moderate** means a documented report or a broad proxy with meaningful bias; **low/unresolved** means hypotheses, selected experiences, incompatible denominators or information not yet collected. A program outcome requires an appropriate comparator or a design that assesses counterfactual change. Contacts, satisfaction, loan approvals and gross job counts alone do not provide one. Selection bias, nonresponse, displacement, landlord capture and administrative cost are considered for each proposal.