Sources & methodology

Every number on this site — and where it comes from.

Two kinds of number live here: public data anyone can look up, and our model outputs anyone can reproduce. Both are traceable to a tracked source in our documentation.

  1. our model 77% of Salem renters can't afford the typical below-market 2BR — computed from the ACS 2024 renter-income distribution (B25118; 10,194 renter households) against the modeled 2BR: $2,906 market × the current 5% signing discount → $2,761/mo; income needed at the 30% line ≈ $110k/yr; share below it 76.8%, rounded 77%. Retrieved via the data.census.gov table API, 2026-09. ↩︎ the problem chapter
  2. public data Salem rental vacancy 1.9% (healthy ≈ 6%) — City of Salem Housing Roadmap Report Card, Oct 22 2025. salemma.gov · evidence_base §4a ↩︎ the problem chapter
  3. public data Salem renter median income ~$58k (vs ~$85k all households) — U.S. Census B25118, ACS 2024 5-year. Census Reporter · evidence_base §9a ↩︎ the problem chapter
  4. public data Median 2-bedroom rent $2,906/mo — Salem 2026 current-listings median (hand-pulled Zillow per-bedroom listings, corroborated by the Zillow ZORI City index and Zumper/RentCafe, 2026); the same basis the released Salem region bundle carries. ↩︎ the problem chapter
  5. public data 55% of Salem renters cost-burdened (rent > 30% of income) — ACS 2024 5-year, table B25070: 54.9% of burden-computed renter households; severe burden (50%+) 30.6%, above its 2017 level (27.4%). ↩︎ the problem chapter
  6. public data Median home value $542,800 — ACS 2019–23 5-year. DataUSA · evidence_base §4c ↩︎ the problem chapter
  7. public data Salem rents +34.8% 2020–2025 — RentCafe market trends, 2026. RentCafe · evidence_base §4b / §7 ↩︎ the problem chapter
  8. public data Salem real gross rent +27% vs real median household income +7%, 2016→2024 (nominal +66% vs +39%) — derived from ACS 5-yr: median gross rent $1,106→$1,836 (Census B25064), median household income $61,083→$85,153 (B19013), CPI-U adjusted. 2016 via Patch / 2016 ACS, 2024 Census-verified. Note: all-household real income. Renter-specific income (B25119) tells a different post-2020 story — see the squeeze chart's composition note on the story page. evidence_base §9c ↩︎ the problem chapter
  9. our model The current-model record — corpus libertybee_fwrd_salem2026: 23 funding levels ($2M–$20M) × 500 seeds = 11,500 runs, engine 0.6.0, clean floor $7M (every run from $7M through $20M survives). One of 44 released datasets — 89,800 runs across nine regions and three compensation bases — each reproducible from seed via the published reproduction gate. Full-project census over three versions: 148,309 persisted runs ≈ 2.97M simulated years (counted from every store, 2026-09). ↩︎ what we built
  10. our model Discount exchange rate (V2-era exhibit) — measured at the founder-subsidy pay basis and retired from the story page; the grid and its finding (no guaranteed funding level past 40% off) live on the version archive, and the depth-ladder datasets ship in the release. ↩︎ the version archive
  11. our model Tenant savings vs. market under the current deal — median $8,196 kept per household over its whole stay (median stay 4.0 yr), before credits; top quartile $36,715+; households still present at year 20 (n=571) median $274,391; curve medians $15.1k @ yr 5 / $65.3k @ yr 10 / $155k @ yr 15. Corpus libertybee_fwrd_salem2026 @ $7M — 47,624 households across 500 seeds; counterfactual = each lease's signed rent carried at its own run's market drift. Households housed per org: median 94 @ $7M → 291 @ $20M. ↩︎ what people keep
  12. public data Median market renter tenure ~2.2yr (public-housing benchmark ~4.9yr) — Redfin / Census ACS; Collinson et al., NBER w21071 (2015). NBER · evidence_base §10a ↩︎ what people keep
  13. public data Below-market/rent-protected tenants ~19% less likely to move (causal) — Diamond, McQuade & Qian, NBER w24181 / AER 2019. NBER · evidence_base §10b ↩︎ why people stay
  14. our model Share of realized retention attributable to regional scarcity ≈ 48% (the deal ≈ 52%) — re-measured on the V2 engine (0.6.0), by 20-year composed turnover with retention off (base, β=0/γ=0 → 21.8%), the deal only (γ=0 → 17.6%), and full (γ=0.5 → 13.6%): scarcity's slice = (deal-only − full) ÷ (base − full), pooled over 8 seeds at a viable funding level, each run in its own freshly-restored database. Strongly tenure-dependent: scarcity dominates early when the deal is shallow, the deal takes over as it deepens. reproduce: scarcity_remeasure.py (V2-engine measurement, not re-run on the fair-wage basis; cited for the mechanism.) ↩︎ why people stay
  15. public data In-kind housing transfers cost roughly 2.4× per dollar of tenant benefit — Olsen, Journal of Political Economy 1972. evidence_base §10e (local olsen1972.pdf) ↩︎ what people keep
  16. our model Tenant Credit System — of credits accrued across the current-model record @ $7M: 49.1% redeemed (≈ $6,683/household), 24.1% forfeited, 26.8% still held at year 20 (median run: $1.27M accrued / $628k redeemed / $309k forfeited). Value staying with tenants (rent savings + credits redeemed), median org over 20 years: $3.9M @ $7M → $11.1M @ $20M. Same corpus as [11]. ↩︎ the credit system
  17. public data The squeeze chart — rent: Zillow ZORI (all-rentals composite, smoothed), ZIP 01970 = Salem, June values 2017–2026 ($1,771 → $2,731); income: ACS 5-year B25119 renter median per release vintage, 2015–2024 ($39.5k → $57.9k; wide MOEs, overlapping windows), retrieved via the data.census.gov table API 2026-09; burden shares from B25070. The reach line = 30% of monthly renter median income. ↩︎ the squeeze chart
  18. our model The current deal, as configured — sign 5% below market; tenure reductions 3% / 3% / 5% at months 36 / 72 / 120, cumulative off the original signed rent; timely-payment credit 10%. Read directly from the parameter records shipped inside every released dataset. ↩︎ the deal ladder
  19. public data Wages — BLS Occupational Employment and Wage Statistics, May 2025 (bulk flat file, downloadt.bls.gov): each region's own metro (Boston, Seattle–Tacoma, San Francisco–Oakland); hire-in = metro median, career cap = 75th percentile; occupations 11-9141 property management, 11-3012 administration, 49-9071 maintenance; benefits 25% of base (declared rule). ↩︎ method · compensation
  20. our model The release — 44 frozen datasets, 89,800 runs: the 2026 Salem record, the 2025 vintage kept as the historical comparison, six North Shore towns, Tacoma, and San Francisco, across three compensation bases; every dataset with its SHA-256 checksum; engine 0.6.0 throughout; the living-series archive (26,668 runs) alongside. ↩︎ what we built

Full provenance, with verbatim quotes against link-rot, lives in evidence_base.md in the repository. This ledger is the site's window onto it.

Liberty Bee
↑