The Scenario Dossier: every assumption in the model, held against the market record.

How the research was done

Four independent research tracks, run in parallel, each covering one family of assumptions:

1. Starlink economics — constellation costs, audited financials, pricing history, capacity physics, churn precedents;
2. Rural ISP market dynamics — private-equity rollup behavior, price wars vs harvests, small-ISP failure rates, loyalty and retention data;
3. Utility & telecom economics — public-power rate records, peer-PUD histories, bandwidth usage/price curves, equipment lives, delay costs;
4. A forensic Starlink cost pass — the "strip the flattering accounting" deep-dive, whose ten most load-bearing claims were then adversarially re-verified one by one (all held; four gained precision corrections; two widely-circulated figures failed and are banned from use — see §3.2).

Rules of evidence: every claim carries a named source; conflicting estimates are shown side-by-side with each estimator named; vendor-published numbers are labeled as vendor-published; primary documents (SEC filings, FCC dockets, NTIA schedules, PUD fee schedules) outrank trade press, which outranks aggregators. Confidence tags — HIGH MEDIUM LOW — appear where a number's provenance matters.

The four grades

The map: every assumption, against the record

"Alone" = the card's solo effect from the 88% base (published lever values; "live" = computed in-browser, labeled as such in the game). Every row links to its full research section.

The headline finding

Two grades deserve special attention. "Generous to the fear" marks assumptions where our model charges more danger than any documented case (the NCI price war, the zeroed loyalty credit, the wear-out accounting) — meaning the published odds are floors, not ceilings. And the single most consequential extreme — the literal 15-year rate freeze, the one card that genuinely sinks the line — is controlled by nobody but the District itself.

Starlink: the machine, audited

SpaceX merged with xAI in February 2026 and went public on June 12, 2026 (Nasdaq: SPCX, ~$1.75T valuation, ~$75B raised) — so Starlink's finances moved from analyst guesswork to SEC-filed segment data. This dossier uses the S-1 as its spine and analyst estimates only as cross-checks.

The Starlink segment is genuinely profitable, and this document says so plainly. At SpaceX's internal launch cost the segment throws off cash — Quilty dates the first free-cash-flow-positive year to 2024, and Starlink generated ~$3B of FCF in 2025. Two honest caveats. First, that profit is contingent on flying its own rockets: analyst Tim Farrar argues “if Starlink was paying full price for its launches, it would be making a loss, even though it's an amazing business” — the vertical-integration moat, not a subsidy from a profitable rocket arm (the launch segment itself ran a ~$657M loss). Second, the consolidated company posted a ~$4.9B net loss in 2025 (capex $20.7B vs $6.8B operating cash flow) — but that burn is Starship and AI datacenters, funded by equity. In cash terms Starlink is the engine funding those bets, not the reverse. SpaceX has needed no primary capital since January 2023 (later raises were secondary/tender for employee liquidity; it even put $2B into xAI).

SpaceX S-1 & Q1 2026 filing via Via Satellite HIGH; IPO — TechCrunch, CNBC HIGH; FCF — Quilty 2026 forecast MEDIUM; capital raises — CNBC HIGH; mix — S-1 via Mostly Metrics MEDIUM.

Starlink: the true cost stack

The internal accounting flatters the segment — Starlink flies on its own rockets at an internal price of $0 while the rocket division booked a $657M operating loss, and launch costs are capitalized and depreciated. Here is the stack rebuilt at SpaceX's true costs.

The replacement treadmill

Unit costs, by named estimator

Everything else is small next to the treadmill

• Ground segment: 100+ US gateways (SpaceX FCC filing); cumulative ground capex ~$1.0–1.5B over eight years (Quilty, relayed LOW-MED) — hundreds of millions a year, not billions. Best priced proxy: Starlink's own Community Gateway at $1.25M + $75K/Gbps/mo HIGH.
• Terminals as acquisition cost: dishes cost ~$1,300 vs a $499 price in 2021 (Shotwell HIGH); subsidy declared over in Sept 2023; Gen-3 build cost now ~$300–400 vs $349 MSRP MEDIUM — and in June 2026 hardware briefly became a $10/mo rental for new residential customers — then reversed within weeks (the dish is buyable again at ~$349), a reminder that even Starlink's hardware pricing moves both ways HIGH.
• Opex minimized structurally: near-zero transit via open peering (559 networks, 69 IXPs); ticket-only support; self-insured satellites; FCC regulatory fee <$1M/yr.

Launch-subsidy accounting — Motley Fool MEDIUM; connectivity capex $4.2B — S-1 via Tunguz HIGH; terminal rental — RV Mobile Internet · Fierce HIGH; peering — PeeringDB HIGH; bear case (presented, not endorsed) — Johnson/Farrar via Stocktwits News MEDIUM.

Starlink: the pricing record — cheap where empty, expensive where captive

Pricing follows strategy, not cost — the case file

• Memphis, June 30 2026: half-price service + free hardware for the metro, amid the xAI data-center backlash — goodwill pricing HIGH.
• Pentagon, 2026: a reported fivefold increase (~$5K → ~$25K/terminal) once drone programs depended on it — Reuters-derived; SpaceX and the Pentagon dispute the report HIGH that it was reported.
• Aviation, July 2026: Global Unlimited $10K → $20K/mo, hardware $145K → $200K, existing customers migrated — no rationale given HIGH.
• Rural incumbents' bases: $15-off × 4 months + free dish promos, aimed precisely at rural ISP customers (named by Shentel's CEO as the driver of its churn) HIGH.
• The captive end: the Washington Post (June 11, 2026) documented a Nebraska family's 44% increase and a $59-promo customer told his bill would more than double — "archipelagos of regional monopolies" (Meinrath, quote verified) HIGH.

CNBC (2023 split); surcharge tracker; May 2026 increase; Bloomberg (Memphis); aviation doubling; Washington Post.

Capacity physics & churn precedents — grading the Starlink cards

Capacity is a hard, local ceiling

• Per-satellite capacity: V1.5 ~17–23 Gbps → V2 Mini ~60–96 Gbps → V3 claimed 1 Tbps MEDIUM.
• US median download: ~90+ Mbps → 54 Mbps at the Q3 2022 congestion trough → ~105 (Q1 2025) → ~128 (Q1 2026) as V2 Minis arrived — congestion is real, and so is the recovery (Ookla data) HIGH.
• Cells hard-cap and show "Waitlist/Sold Out"; Washington and Oregon cells waitlisted in Nov 2024 HIGH.
• Independent analysis (X-Lab working paper, Meinrath et al. — an on-record Starlink critic, disclosed): service degrades below 100/20 beyond ~6.7 subscribers/sq mi per beam; only ~17% of US Starlink speed tests met 100/20 in mid-2025 MED-HIGH, assumption-dependent. Okanogan averages ~3 households/sq mi county-wide — below the threshold in aggregate, though valley clusters can strain individual beams. County cell status is checkable only address-by-address; we claim nothing we can't verify.

The churn precedent table

Gen-3, Starship — and the steelman

The EXTREME card requires next-generation capacity to arrive on schedule and Starlink to hold predatory rural pricing forever. The record on SpaceX promises:

Pattern: directionally delivers, chronically late, at a quarter to half of headline. The delay cost is already visible in prices, not PDFs — the PNW surcharges exist precisely because V3 capacity hasn't arrived.

SpaceX latency paper; Ookla speed history; Starship grounding; Reuters on Starship spend; TMF Associates; Opensignal.

NCI & the private-equity record

NCI Datacom's owner — Core Fiber Partners, a Heritage Holding platform — is a fiber-first rollup: six ISP acquisitions 2022–24 (NCI, MiFiber, Wind Wireless, Hyak, Thrive, Hilltop), stated preference for "fiber or hybrid-fiber broadband" in rural towns of 1,000–20,000. Reselling capacity on someone else's wireless network is not its strategic center HIGH for the portfolio facts.

What PE platforms actually do with side-lines

• Preseem's tracker logged 40+ WISP acquisitions in 28 months; GI Partners bought Rise Broadband explicitly to pivot it to fiber HIGH.
• Fast segment exits are documented: Starry exited a whole metro within a year of trouble; Trueline — a PE fiber rollup — shut down five months after launch, terminating ~300 staff by email HIGH.
• PE-broadband literature describes "cherry-picking of lucrative markets, cost cutting and price hiking… forgoing local upgrades" MEDIUM — advocacy sources, consistent and first-hand.
• Counter-evidence: median PE holds run 5–7 years; Heritage markets "long-term capital"; no wind-down appears in the 40-deal tracker. A whole-company year-two exit would be fast. The base card models a segment exit — the reseller line only, with ~half its customers migrating to other local RSPs and its transport purchases continuing.

Price war vs harvest

We searched for a documented case of a departing or PE-owned rural ISP running a multi-year price war before exit. We found none. The documented pattern is harvest: Frontier admitted it "underinvested for years" while raising prices on its declining copper base; CenturyLink stopped marketing its slow tiers rather than cutting price; Starlink raised rural prices 44%+ once customers were captive. Documented price cuts run the other way — incumbents dropping $15–40/mo when a second network enters (the Longmont pattern) — which is the argument for keeping a public network in the market. One honest scope-note, because it's easy to over-claim: that $15–40 price-discipline evidence (Berkman Klein, CPUC/UCSB, Longmont) is fiber-specific — the CPUC study affirmatively finds fixed-wireless and satellite entry do not move incumbent prices the way a competing fiber network does. So we do not lean on it to value the wireless network. The price discipline that actually applies to Okanogan's wireless market is the direct, measured one: markets with 3+ providers pay about $5/mo less (Consumer Reports, 22,000 bills, technology-neutral), and the District's eight retail providers compete over wholesale sold at cost under RCW 54.16.330 MED — fiber studies scoped out; the local mechanism is the load-bearing one.

Heritage portfolio; Preseem M&A tracker; Trueline; PitchBook hold data; Frontier; FBA/Longmont; CPUC/UCSB — fiber (not FWA/satellite) disciplines price; Consumer Reports (provider-count effect).

Reseller shocks: the failure base rate

The "vanishes overnight" mechanic is current reality, not a tail fantasy. One refinement the record adds: when the network survives — exactly the Okanogan situation, where the PUD's wires outlive any retail face — stranded customers re-home to other local providers (the Common Networks → Monkeybrains pattern; subscriber bases are the asset PE buys). The model keeps the loss permanent anyway, which is harsher than precedent.

It already happened here. The county's own RDOF story is a case in point: at the November 2020 auction the eleven eligible Okanogan block-groups were won by SpaceX (10, later rejected — never authorized) and CenturyTel of Washington / Lumen (one, around Twisp–Winthrop). Lumen's award was authorized and funded from about May 2022 — a real federally-backed competitor obligation in the county — and then Lumen defaulted and withdrew statewide in June 2025, leaving Okanogan with no active RDOF-funded provider. An absentee federal award named, funded, and gone in three years; the local network stayed MEDIUM — FCC Auction 904 records (DA-22-523, DA-25-670).

Benton RDOF data; KTTC: RadioLink; FBA WISP population; FCC Auction 904 (RDOF) results and authorization/default releases DA-22-523, DA-25-670.

The rate record: how utilities actually price

The rate suit is where the District plays against itself — so we researched how real utilities behave.

• The cadence: regulated utilities file general rate cases every 2–5 years as the norm; some stretch past ten HIGH.
• The recovery rate — the base card's anchor: national public-power rates rose +11.7% over 2020–24 (APPA/EIA via POWER Magazine) against ~21% cumulative CPI — a ~55–60% inflation recovery. (The CPI denominator is our computation from standard BLS series; the division is ours, the inputs are theirs.) MED-HIGH
• Peer WA PUDs, specifically: Chelan PUD's Feb 2020 plan was "its first permanent rate increase since 2000" — a ~20-year effective freeze — and it resolved it with automatic 3%/yr escalators explicitly covering wholesale fiber, extended at 3–4%/yr through 2026 HIGH. Pend Oreille PUD maintains a full wholesale fee schedule (RSP internet ports at $40/mo — the same figure as Okanogan's proposed tier) re-issued August 2024 HIGH, primary PDF.
• The affordability edge, re-anchored to the primary data: EIA-861 (2024) is the source under the APPA headline — public-power residential bills rose far less than IOUs over 2020–24. But the same table places Okanogan's own electric rate 2nd-highest of six WA hydro PUDs (~10.2¢/kWh), i.e. it has the least electric-margin room to cross-subsidize broadband — which is precisely why the wireless line has to stand on its own, and (per the audited surplus and the bottom-up P&L) does. Every WA PUD peer sells broadband wholesale-only under RCW 54.16.330 HIGH — EIA-861 Table 6.

POWER/APPA, re-anchored to EIA-861 Table 6 (2024 residential rates by utility/ownership); Chelan 2020 · 2024 extension; Pend Oreille CNS fee schedule; rate-case cadence. Honest counter-evidence: EIA notes electricity prices overall outpaced inflation post-2022 — that's the IOU-driven aggregate, and it supports the upside card, not the base.

Bandwidth: usage vs price, measured

Against that, wholesale prices erode: IP transit fell ~12%/yr compounded (2022–25) while "capacity demand has been doubling nearly every three years" (~+26%/yr) — TeleGeography. Volume has outgrown price erosion in every measured period; the model's 5%/yr net bandwidth-revenue line is mid-range of the implied flat-to-+8%.

OpenVault OVBI 2Q25 · 4Q23 HIGH; TeleGeography erosion · demand vs price HIGH. District figures: audited statements & FOI billing (The Data).

Local demand, loyalty & win-back

Is a local-retention credit earned?

• Consumer Reports (73,000 members, 54 ISPs): all seven top-rated ISPs are small/local/municipal; the giants scored worst HIGH.
• ACSI 2024: smaller fiber ISPs scored 77 — within 3 points of the national leader, ~9 above big cable HIGH.
• Preseem's small-ISP dataset ties NPS ≥ 61 to churn below 3%/yr (vs ~15%/yr national average) — self-reported vendor survey, labeled MEDIUM.
• Peer-reviewed: bundling reduces churn across services, strongest in turbulent demand (Prince & Greenstein) — and several Okanogan RSPs sell phone/managed-IT bundles HIGH for the finding.
• The skeptic's ammunition, kept visible: providers under ~5,000 subs can churn worse when support is under-resourced — loyalty is earned by service, not smallness; JD Power has FWA satisfaction rising; one 2026 survey put Starlink satisfaction at 89% MEDIUM.

Does demand come back after an upgrade?

• Direction — among the best-documented facts in rural broadband: fixed wireless went from ~zero to 12% of all US internet households in five years — the fastest-growing broadband technology in US history; Starlink built a top-10 US ISP on 85%-rural pent-up demand; electric-co-op fiber builds report rural penetration rising faster than urban HIGH.
• Switch-back sentiment is on the record: price-hiked rural Starlink customers told the Washington Post they'd "switch to fiber tomorrow" — and of customers who do leave Starlink, ~50% choose cable, only ~14% fiber-class services HIGH.
• Magnitude — vendor-grade, labeled: the only quantified WISP-upgrade win-back numbers are vendor case studies (Tarana/Nextlink: 1,000 subs in a month; Preseem: a 50% at-risk conversion) VENDOR. The game computes this card's effect live and says so on its face.

Who the retention credit actually serves — the ratepayer, from the Census

The stickiness and affordability defaults aren't set for an average American broadband market. They're set for this county — and the 2024 exhaustive audit replaced the site's earlier inferences with hard American Community Survey and USAC figures for Okanogan County (FIPS 53047):

• 2,694 county households lost the $30/mo ACP subsidy when the program ended in 2024 (about 15% of households), and USAC paid $1.68M into the county over the program's life — the single strongest verifiable equity fact in the file. For a household on that margin, a $10/mo satellite hardware rental or a demand surcharge is not a rounding error HIGH — USAC primary.
• Older, poorer, more disabled than the state: 20.7% in poverty, median household income $63,207 (vs WA ~$94K), 23.2% aged 65+, and a 16.7% disability rate (ACS S1810/B18101; roughly a third above the US ~13%, and 26.4% of residents with a disability live in poverty). This is exactly the population for whom a self-install, mail-back-the-dish, chat-only support product is a barrier, not a substitute HIGH — ACS.
• A quarter of the county already leans on satellite — as a stopgap, not a choice: of home-internet subscriptions, wired broadband only 52.1%, satellite 25.3% (8.9% satellite-only), cellular-only 11.6%, and 9.0% no internet at all (ACS B28002). High satellite presence confirms the substitution pressure the model already prices; it is not, and we don't read it as, proof of churn HIGH — ACS.
• The Colville Reservation is further behind still: 79.1% of reservation households have any broadband and 14.9% have no internet at all — roughly 8 points worse than the county — and on-reservation Native households face 25.5% poverty (ACS B28002/B17001 for the reservation geography) HIGH — ACS.

None of this is a new lever — it's the evidentiary floor under the ones already in the model: the win-back credit, the ~60% rate recovery, and the conservative stickiness slice all describe a market of price-sensitive rural households with few alternatives, which is what the record shows this county to be.

Consumer Reports; ACSI 2024; Preseem benchmarks; Prince & Greenstein; FWA growth; where Starlink leavers go; Tarana/Nextlink (vendor). Ratepayer profile: ACS — Okanogan County (poverty, income, age); ACS B18101/S1810 (disability); ACS B28002 (internet subscription); USAC ACP (2,694 county households).

Equipment lives & accounting — which test do regulators use?

Standard cost-of-service accounting recovers capital over the service life via depreciation with replacement cycles — no regime demands full payback before the gear dies. So the baseline card (15-year project life, one full electronics refresh charged) matches how FCC/NTIA and ratemaking actually treat wireless plant: long-lived structures plus replaceable radios. The strict card — payback before a drawn wear-out year (median ~11) — is harsher than the FCC's own regime, kept because skeptics ask for it.

FCC DA-03-2738 App. A; NTIA schedule; BEAD guidance. All HIGH.

The cost of waiting — has it reversed?

The old rule — electronics get cheaper if you wait — no longer holds:

• 2025 tariffs pushed network equipment up 5–20% with longer lead times (Dell'Oro; multiple trade analyses); BLS producer prices for communication gear turned upward against their historical flat-to-down trend MED-HIGH. Counter-evidence: tariff policy is volatile and could reverse; fiber electronics are least affected.
• The clock inside the three-year card: Cambium's PMP 450 end-of-support is December 31, 2027 — and the vendor carries a going-concern qualification, defaulted under its credit agreement, and was delisted from Nasdaq in March 2026 (now ~$0.19 OTC). Even that date isn't bankable HIGH — SEC filings.
• Customers don't wait with you: 18% of rural subscribers switched providers within the past two years, price the top reason; Starlink converts "last resort" into first choice; and only ~14% of Starlink leavers pick fiber-class options — the win-back door is narrow. No controlled study measures satellite win-back for delayed rural builds; we label that inference HIGH for the switching data.

Dell'Oro; Cambium EOL · SEC 8-K; 2026 rural subscriber study.

The inflation regime

The model draws inflation around a 2.5% center with a 12% chance of a high-inflation decade and a 9% cap. Against the anchors: CBO projects 2.0% average PCE inflation 2027–35; the Fed's longer-run SEP is 2.0%; options markets priced a ~32% chance of >3% five-year inflation at the 2021 peak; and 2–3 of the last nine decades averaged ≈5%+ (none above 7.25%). Verdict: centered half a point above the official anchors, with a tail probability below both the historical base rate and peak market-implied odds — mildly conservative, not alarmist.

CBO; Fed SEP; Minneapolis Fed; decade averages.

The service lifecycle & the adoption gap — why satellite is a different kind of product

This section documents an argument that is largely qualitative, and is honest about that: it strengthens the case on the demand/loyalty side but, as the model note below explains, it does not justify a new numeric lever, because no published Starlink churn or return figure exists to anchor one. Steelman first: Starlink is a genuine product, its self-install is near-plug-and-play for many, and it scores well on satisfaction (ACSI 2026: 76, top-3 non-fiber; ~73% of users "rarely/never" lose service). The argument is not that it's a bad product — it's that a satellite plan and a local ISP are different kinds of product across the fifteen-year life of a connection.

Support & repair — the relationship, quantified where it can be

Who can actually adopt it — the install & demographic gap

• Self-install is near-plug-and-play on a clear lot; but a strict obstruction rule (5%+ sky blockage → frequent disconnects) means forested Okanogan parcels often need a mast, roof mount, drill, and NEC-810 grounding. The paid-install market is the proof: Starlink sells a $199 pro install through OnTech in its own checkout; independents charge $200–$600 ($600–$1,500 forest/pole). HIGH (Starlink checkout)
• Okanogan's demographics concentrate the populations that struggle: 22–23% are 65+, 6.3% are 75+, 20.7% in poverty, 24.7% college-educated (vs ~35% US), 9% veterans. Nationally (Pew): ~⅓ of seniors never use the internet; smartphone ownership (the setup app's requirement) falls to 17–31% in the 75+ brackets; 15–30% of seniors have an ambulatory disability precluding roof/mast work. NTIA's five digital-equity groups (aging, low-income, disabled, veterans, rural) are a near-complete overlap with this county. HIGH (Census/Pew)
• The June 2026 shift to a $10/mo hardware rental (no purchase) plus the ACP subsidy ending (23M households, ~5M disconnected) stack the affordability barrier onto the lowest-income rural households. HIGH

Could a local shop just support Starlink instead? — the reseller economics

• The channels a small local business can actually enter pay one-time income only: an installer takes a $199–$600 fee; the referral program pays a one-time ~$120 credit. Neither earns a share of the monthly bill to fund ongoing support.
• Under Starlink's own reseller terms (primary document), opening or repairing the dish voids the warranty; all replacement routes through SpaceX. A local shop can mount and aim it — it cannot fix it. HIGH (primary terms)
• The one tier that pays recurring revenue — Authorized Reseller — is invite-only, currently closed to new applicants, and gated at a business plan of ~$5M/yr (≈2,500 active terminals), built for national enterprise integrators (SageNet, Speedcast, Trace Systems…). Structurally out of reach for a county shop. HIGH

Full research: three parallel July 2026 tracks (support/satisfaction; install/demographic; reseller economics). Primary sources include the SpaceX/MetTel reseller Terms (warranty-void & unilateral-change clauses), ACSI 2026, BBB, US Census/ACS & Pew, and Starlink's own checkout & support pages. Two honesty notes: the "~35% pay for a pro install" figure is single-source (directional, not anchored); and Okanogan's exact disability % is inferred from the rural benchmark (Census QuickFacts blocked automated capture). Where a figure is soft, the pages hedge it.

Card-by-card mechanics

Each card overwrites its suit's parameters in the live engine (the v2.2-parity calculator: 40,000 PERT-drawn futures per move). Published solo values come from sensitivity_levers_output.json; "live" cards are computed in-browser and labeled.

Parity is tested: a scripted harness plays every card solo and compares against the published lever file — all within ~1.6 points (the documented calculator-vs-Python tolerance).

Evidence gaps & failed verifications

• Starlink publishes no churn figure; NoaNet publishes no wholesale price history; no controlled study measures satellite-to-WISP win-back rates. Where those numbers would sit, we say "inference" and show the analogy used.
• WISP-upgrade take-rates are vendor-published (Tarana, Preseem); small-ISP churn benchmarks are self-reported surveys. Labeled at every use.
• Quilty's free-cash-flow methodology (how it prices internal launches) is paywalled — flagged, not assumed.
• Global gateway-station counts conflict across trackers (definitional); Okanogan's own Starlink cell status is checkable only address-by-address and is not claimed.
• No documented case shows Starlink raising prices specifically after a competitor's exit — the harvest claim rests on the promo-then-hike and surcharge record, which is extensive.
• The physical-plant count we quote (158 access points / 46 sector areas) comes from the District's own published service-area map but is not independently reproducible cell-for-cell; the Institute for Local Self-Reliance reported 143 access points in a 2013 build-plan snapshot. The two differ by build year and definition, not contradiction — but until it's sourced to the FCC tower registry or an as-built inventory, we caveat the number rather than assert it.

Corrections this research forced on our own pages

• Starlink's US residential pricing restructured January 2026 and rose in May: tiers now $55 / $80–90 / $130 plus the $10/mo hardware rental. Older "~$120/mo" references correspond to today's Max tier.
• The PNW demand surcharge peak of $1,000 later eased to ~$500 across most of the region (the ~$1,500 peak is Fairbanks, AK).
• SpaceX's IPO moved Starlink's finances from analyst estimates to SEC-filed data — this dossier and Why Not Just Starlink? were rebuilt on the S-1.
• The 158 AP / 46 sector infrastructure count now carries an explicit provenance caveat wherever it appears (coverage, stability, the business-case flyer), cross-referencing the third-party 143 figure rather than asserting a single precise number.
• The carrier-customer dollar figures (NoaNet ~$665K, NCI ~$248K, and the rest) are labeled as annualized from one month of the District's billing, not summed across the year — flagged everywhere they're shown so the basis is never mistaken for a full-year total.
• The debt framing was tightened. The District's ~$50M is its total revenue-bond debt (~$51.8M, 2024) — overwhelmingly (~96%) electric, but a single combined obligation with all District revenues pledged under one covenant. So the telecom refi (~$1.6M) is a small slice of that same debt, not a legally separate lien; we no longer describe it as "separate," only as small and self-supporting.
• The county's own RDOF history is now named (§2.7): Lumen/CenturyTel held an authorized, funded RDOF obligation around Twisp–Winthrop from 2022 until it defaulted statewide in June 2025. We surface it rather than leave the competitive record silent on it — the outside federal award came and went; the local network stayed.
• The price-discipline claim was scoped to fiber (§2.6). The Berkman Klein / CPUC / Longmont evidence that a competing network cuts incumbent prices is fiber-specific — CPUC finds fixed-wireless and satellite entry don't do it — so we no longer lean on it to value the wireless network, and instead rest that point on the measured local mechanism (provider-count effect + cost-based wholesale under RCW 54.16.330).
• The Wharton/Penn "muni fiber can't cover its costs" study is now engaged head-on (on Answering the critics): its cases were all greenfield fiber at $77K–$204K/home; Okanogan is already-built, wholesale, and — per the bottom-up P&L — covering its costs, which is the exact distinction that study turns on.

How this connects to the model — the three references

Everything about the payback model is public and split across three places, each answering a different question. Read together, they are the complete context for how the model was built and why every default sits where it does.

The division in one line: model documentation is the mechanics and the structural inputs; this dossier is the evidence behind the competitive inputs. Neither is the whole story of the model's defaults on its own — together, plus the code, they are.

Regeneration rule: when the model version changes, sensitivity_levers_output.json regenerates and every lever value on this page updates automatically — the research prose and verdicts are re-reviewed manually against it.