House rules — how the table stays honest

• The table starts full, not empty. The five face-up cards are the conservatisms already inside our 88% base case: NCI's year-two exit, rates lagging inflation, bandwidth credited at a third of documented, the modest demand credits, and full-life accounting with a refresh charged. Pull them and the odds go up — that's the point.
• Ordinary stress and EXTREME are different cards. Starlink growing and NCI adding its own equipment are priced into the base; the red-edged EXTREME cards (a +3%/yr forever surge, a literal 15-year rate freeze, bandwidth to zero) go beyond anything in the observed record and say so on their face — so nobody accommodates a fear with a card that was never realistic.
• One card per suit. Playing a new Starlink card replaces the old one — assumptions can't silently double-count.
• The engine is the real model. Each move re-runs 40,000 futures of the v2.2 calculator (same PERT draws as the published 200,000-run model); each card's printed solo effect is the published lever value. Prefer sliders to cards? The full calculator is here.
• Every card is researched, not just asserted. The scenario dossier qualifies each one against the market record — Starlink's satellite-replacement economics, WISP failure rates, peer-PUD rate histories, the FCC's own equipment lives — with sources, verdicts, and the counter-evidence.
• The ✦ assistant is bounded on purpose. If it's enabled, you can describe a scenario in plain words and it will set up the table for you — but it only chooses which of our researched cards to play (or composes a custom one clamped to the range our research already covers). It never invents the odds: the same model computes every number, every choice links to its evidence, and it declines rather than speculate when a scenario is beyond our data. AI can make mistakes — it's a shortcut to the cards, not an authority.