Research that names its own kill switch.

A live board rather than an article: thirteen instrumented needles across company telemetry, ecosystem water level, and market regime. The stated doctrine is that one needle is only a note and three pointing the same way is a signal, with a close below the $305 gate turning the whole board red.

Places the cycle on the right shoulder of the top, at stage 2 of 5 in a writedown cascade, and argues the unwind is three moves rather than one crash: a rally that traps buyers, then writedown confirmation, then default adjudication. Navigates by a trigger dashboard and kill switches instead of by catalyst dates.

A do-not-participate verdict at high conviction. Grey-market prints cleared below the fixed $135 offer, the deal came at roughly 95x trailing price-to-sales with losses accelerating, and 20 percent of shares unlock two days after the first earnings report.

Splits the whole-company listing into a knowable, priceable body -- Starship, Starlink, cash flow, compute vertical integration -- and a narrative-supported option in orbital AI datacenters. Then argues the option's binding gate, frontier silicon that runs stably at 300 degrees Celsius, is not something SpaceX controls.

Argues today's prices are not an organic compute-shortage supercycle but a coerced arms race: Nvidia's circular financing funds rivals to the four hyperscaler landlords, so each is forced to keep buying while enterprise GPU utilization sits near 5 percent and reservation prices still rise. Carries a full bull case and stated falsification conditions.

Builds an industry LLM revenue run-rate from the bottom up and sets it against 2026 hyperscaler capex guidance, deriving a sustainable capex ceiling that leaves a gap of roughly five times. Weights three end games -- soft landing, capex bust, AGI fork -- and names the single trigger to watch: the first hyperscaler to cut capex guidance.

Screens 26 names across 800V HVDC power, memory, optical interconnect, and generation on two axes at once: a qualitative rank for position in the Rubin relationship graph, and forward PE and PEG out to FY2030. Every name gets CONFIRM, WATCH, or REJECT, with trough-to-peak PEG flagged explicitly as a cheapness trap.

Mass AI adoption requires cheap tokens, but every infrastructure layer is currently taking 72 to 78 percent gross margins, so repricing is presented as arithmetic rather than forecast, with 1880s copper and electrification as the precedent. Version 2 recalibrates against three May datapoints and names which parts of the original thesis failed.

Tests seven market narratives across TSLA, SPCX, SpaceX, and xAI and finds six falsified and one partly falsified, arguing the market prices an integrator-optimizer track record at zero-to-one creator multiples. A sum-of-parts anchor sits well below the IPO target, and the four option narratives are each valued near zero.

Deliberately adjudicates two separate propositions instead of one. "Not worth initiating a long here" gets medium-high confidence given the multiple, while "the mismatch is actively shortable" gets only low-medium, because the load-bearing question -- whether humanoid robots are shallow engineering that China commoditizes -- is genuinely unresolved.

Reads the May 2026 sixteen-CEO US delegation to Beijing by its absences: the roster sorts cleanly into five lanes, and the conspicuous names left at home are the actual policy document. Argues the visit was the cover page and the guest list was the argument. Geopolitical Briefs, Issue 02.