Two Hunters Stalk One Stag, The Embargo Boomerangs, $1.7 Trillion Hides Off The Books

Graylin puts the Buffett indicator at 240% of GDP versus 120% at the dot-com peak, with 45% of US stock market value in the AI sector and $1.7T of off-book hyperscaler debt against Enron's $200M. He calls a correction 'due' and reads NVIDIA's $500B chip-securitization deal with BlackRock, Carlyle and Blackstone as subprime with GPUs.

▸ Ship a weekly agent sweep that recomputes the Buffett indicator from FRED (Wilshire 5000 and GDP series), pulls AI-sector market-cap weight and hyperscaler off-balance-sheet debt from SEC EDGAR 10-Q filings, and publishes a dated 'bubble vital signs' capsule set plus a JSON feed on the RDKG MCP endpoint.

$ FKG subscribers and agent queries pay per hit; protects the house treasury ahead of the correction Graylin says is due.

The episode's core stat: Chinese open-weight models climbed from 2% to 61% of OpenRouter traffic in two years, Qwen is at roughly a billion downloads with 180,000 derivative models, and the open-closed gap has shrunk from about 18 months to 2-3 months. Whoever publishes this share weekly, with receipts and freshness dates, owns the reference number everyone else quotes.

▸ Stand up a weekly agent sweep of OpenRouter's public model rankings and the Hugging Face Hub API (download and derivative counts) that publishes an open-vs-closed, China-vs-US share barometer as freshness-dated citation capsules with llms.txt entries.

$ Agents pay per MCP query for the canonical number; grounds every garden's own model choices.

Graylin frames the buyer's choice as '$50 per million tokens or the cost of electricity,' noting 90% of users are fine with today's models — Harvey runs on GLM 5.1 and OpenEvidence on GPT-4-class weights. If a $1B frontier model can be matched by distillation for a few million dollars, our pipelines should be riding the cheap side of that curve.

▸ Replay a sample of RDKG's real pipeline tasks (sweeps, capsule drafting, tagging) through hosted Kimi, GLM, Qwen and DeepSeek endpoints on Replicate and OpenRouter, then ship an internal cost-per-capsule scorecard that flips pipeline defaults wherever quality holds.

$ Protects runway by cutting pipeline COGS across every garden — the same margin move Harvey made by running on an open Chinese model.

The first US-China AI safety dialogue lands September 24 in DC, where Graylin — advising the US side — says success is simply agreeing to a second meeting, with non-state-actor risk (ransomware up 200-300% in two years) as the top shared priority. Meanwhile China's Waco has signed 29 countries against Pax Silica's roughly 25, a two-bloc map nobody is publishing in citable form.

▸ Compile an agent-built, citation-first dossier on the September 24 dialogue from Graylin's 'Beyond Rivalry' and 'Misdiagnosing the US-China AI Race' papers, the WAIC and Waco membership announcements, and the UK AI Security Institute report, publishing it as dated capsules plus llms.txt entries before the meeting and auto-refreshing after.

$ A timed citation magnet that converts agent traffic to pay-per-query; buys RDKG a cited seat in the September conversation.

China delivers electricity at 2-3 cents per kWh — 10-15x cheaper than parts of the US — while building roughly 10x more new generation per year than America and losing under 1% per 1,000 miles on UHV lines, with data centers sited at the generation to avoid stranded power. Graylin says the bottleneck is shifting 'down the stack to compute, electricity, power,' which makes the price spread itself the signal.

▸ Build a monthly agent pipeline that pulls US industrial electricity prices from the EIA open-data API and China and global figures from IEA public statistics, publishing a two-cent-kilowatt ledger capsule set with a derived electricity-cost-per-million-tokens metric.

$ FKG readers and energy analysts pay per query; steers where the gardens buy their own compute.

The episode's Fountain Life segment claims 88% of people scanned show detectable coronary artery disease, 23% carry soft plaque that calcium scoring misses, and 50% of heart-attack deaths come with no warning signs. Those are checkable numbers, and RDKG's rule is that unverified claims get labeled — especially sponsor claims the rest of the audience swallows whole.

▸ Run an agent sweep of PubMed and OpenAlex for CT-angiography and soft-plaque prevalence literature (SCOT-HEART and PROMISE cohorts among others) to test the 88/23/50 claims, publishing a verified-versus-promotional capsule pair with primary citations and freshness dates.

$ HKG readers and clinic partners pay for verified longevity claims; protects house credibility from sponsor hype.

Graylin's public corpus anchors the whole episode: 'Misdiagnosing the US-China AI Race,' 'Beyond Rivalry,' the Cipher Brief small-models paper, the 'Great Reckoning' Substack, and the Brynjolfsson Enterprise AI Playbook (50 successful deployments, 80-90% of blockers organizational, versus McKinsey's finding that only 6% of AI deployments work). Agents will query these claims all quarter, and the citations should resolve to us.

▸ Index the five public Graylin papers into 40-80-word citation capsules with JSON-LD and llms.txt entries — each key claim source-stamped, freshness-dated and hype-flagged — so agent queries about the US-China race resolve to RDKG's MCP endpoint.

$ Inbound agent citations convert to MCP pay-per-query; MKG-style gravity at near-zero marginal cost.

Graylin's on-air war game: the private credit funding US data centers pops within about two years, Washington asks Beijing to buy Treasuries as in 2008 — when China's trillions in T-bill purchases held US rates down — and the quid pro quo is American hands-off ambiguity on Taiwan. He adds that invading Taiwan yields no workable fabs and the US would fly out 1,000 TSMC engineers before anything happens.

▸ Encode the T-bill-for-Taiwan scenario in the Orrery with a weekly agent sweep watching US Treasury TIC major-foreign-holders data, hyperscaler debt-issuance filings on SEC EDGAR, and official one-China policy statements, flagging tripwire movement in dated capsules.

$ The Circle's positioning intelligence — FKG-grade early warning on the reckoning before it is consensus.

Graylin's new Cipher Brief paper found no correlation between model size and real-world danger: 10-million-parameter models design chemical weapons and 1-50B-parameter models design viruses, all runnable on a basement laptop. He says the actual fix — controlling precursors and the synthesis machines that turn designs into material — is the layer 'nobody's talking about,' which is exactly the verified ground truth TKG exists to own.

▸ Sweep the Cipher Brief paper, the Australia Group common control lists, OPCW scheduled-chemicals annexes, and EPA and ECHA chemical inventories with agents to publish a machine-readable capsule map of which precursor and synthesis-control ground truth exists versus is missing, stamped as TKG territory on the MCP endpoint.

$ Biosecurity funders and policy shops pay for verified precursor ground truth; compounds the MTP moat — the ground truth AI needs to survive.

Over 200 humanoid companies demoed at WAIC against a global market of only tens of thousands of units last year, 80-90% shipped by two or three Chinese firms, and Graylin expects consolidation to single digits within two years. Unitree told him research labs are nearly the only buyers, legless upper-torso units outsell legged ones commercially, and 90% of robot components come from China — US firms are literally suitcase-smuggling actuators home.

▸ Cross-reference IFR World Robotics statistics, UN Comtrade actuator and component trade flows, and the public WAIC exhibitor list with an agent sweep to publish a quarterly demoed-versus-deployed humanoid reality index as citation capsules.

$ FKG watchers pay to sidestep the 200-to-single-digits shakeout; shields BKG and AgKG from betting attention on robot hype.

The episode's job math: 9% of US college grads are unemployed and another 42% underemployed — roughly half not landing career-track work — while partners at consulting, accounting and law firms tell Graylin they 'don't need the junior guys anymore.' His surviving-labor list is teachers, nurses and elder care, exactly what homeschool families need spelled out with receipts.

▸ Draft an agent-compiled EdKG brief from the NY Fed college labor-market dashboard and BLS series via FRED verifying the 9% and 42% figures, mapping the episode's displacement-resistant roles into a citation-first family one-pager published to the library.

$ Homeschool parents pay EdKG for post-junior-job guidance nobody else grounds in primary data.

Graylin reframes the AI race as Rousseau's stag hunt — hunt together and both families eat for a month; right now the US chases the stag alone while China takes the hare, the worst square on the board. RDKG already holds a Two Hunters manifest, and this episode hands the parable a script, real stakes, and a September 24 news hook.

▸ Render a 90-second 'Two Hunters' parable short through the KG Visual Studio plate press and existing Replicate motion pipelines — the stag-hunt payoff grid as Victorian engraving — cut to the episode's framing, auto-saved to the Library for an internal-first premiere before September 24.

$ Builds The Call's catalog and audience at near-zero marginal cost; the cinematic door-opener for the foresight library.