What this episode opens for us — money angles, ways in, and what our own agents can build within days. One card is the reality check.
X's Global Affairs post (Aug 28, self-reported — X investigating X) claims a ~200,000-account Chinese bot farm, 200 accounts pushing "data centers raise your electricity bill" narratives. Hosts admit "no easy defense" and beg for "absolute truth." That truth exists with receipts: Quincy, WA — poverty 29.4% to 6.2%, data centers 57% of tax base, $120M high school, levy rates fell (CNN). Polling: Gallup has 71% opposing a nearby data center; the tape's 75% is the separate, also-real Heatmap poll. Nobody owns this receipts layer yet.
▸ Run an overnight agent sweep and ship a Data Center Ground Truth capsule pack — 12 citation capsules, JSON-LD, llms.txt-listed: Quincy case file, poll-vs-poll explainer, X-claim provenance note labeled self-reported. Draft outreach for John to data-center developers and economic-development offices.
$ Agent-priced capsules plus sponsorship from hyperscaler community-relations teams — the industry Peter says "needs to be telling this story."
Alex Wissner-Gross says on tape he cannot tell income-statement from balance-sheet circularity in Nvidia-financed demand and "would love more clarity." The record: Nvidia's $96.2B quarter (+106% YoY) is verified, with up to $100B pledged to OpenAI's buildout, reported $250B backstop talks, and $105B Ohio backing. Tape slips: guidance is +12.3% QoQ to $108.0B (tape said +10%), gross margin 75.0% (tape said 80-85%), and Dalio isn't "out" — Bridgewater added 22% more S&P 500. A named buyer just asked for a product.
▸ Deploy agents to build a receipts-first map of AI capital flows — who finances whose demand, sourced to SEC filings and joint releases — shipped as an FKG frontier brief plus a ten-capsule set; file a Bet Board card for a subscription version.
$ Investors on tape literally requested this visibility; verified circularity data is subscription-grade FKG inventory.
Alibaba's Wan3.0 generates 30-second single-pass video from documents, spreadsheets, slides, and webpages at $0.05-0.20 per second (verified). The tape's "$20-60K for a 90-minute film" is wrong: at real prices that film costs $270-$1,080 — 55-74x cheaper than said on air. Sora's consumer app is gone (Apr 26, 2026) and its API sunsets Sept 24, 2026. The gardens already run a working film pipeline; the cost floor beneath it just collapsed to lunch money, one day after Alibaba's $10.2B raise.
▸ Price the next RDKG film's shot list at Wan3.0-class rates through the existing visual pipeline, publish the corrected cost-math capsule (tape said $20-60K; record says $270-$1,080), and draft an MKG explainer on near-free verified media production.
$ Near-zero production cost makes garden films a loss-proof audience and citation engine.
The jobs segment aired the Principal survey (n=1,000, Jun-Jul 2026) but garbled it. Corrected: 4% expect AI cuts, 31% expect staffing and wage increases, 10% don't use AI, 1.4% of firms blame cuts on AI, and last quarter 52% hired / 38% maintained (tape said 30%) / 12% cut. "100% of job growth from small business" is really young firms (~20% of gross job creation, NBER). The Goldman story inverted on air: the record shows partners warned of "cognitive atrophy" while junior AI-exposed hiring fell 19% (Stanford).
▸ Package a Labor Signals, Verified capsule pack — corrected Principal figures, NBER young-firms finding, Stanford's -19%, Gates's real MIT Tech Review quote, Brynjolfsson's real WaPo line — and syndicate to BKG's small-business audience; agents draft and source it end-to-end today.
$ SMEs and their advisors pay for verified labor data; positions FKG against vibes-driven punditry.
A verified health cluster landed in one week. Rasonque (daraxonrasib) FDA-approved Aug 26 — first targeted therapy for previously treated metastatic pancreatic cancer: median survival 13.2 vs 6.7 months, response ~33% vs ~12%, a daily pill approved about five weeks after NDA acceptance. Corrections ride along: RAS drives ~19-20% of cancers pan-cancer (tape's 30% is lung-specific); enamel regrowth is peer-reviewed but ex vivo (Nature Communications); the "half a trillion (WHO)" merges two sources — WHO counts 3.5B affected, while $710B/yr comes from a separate 2024 study.
▸ Ship a six-capsule HKG specimen set — Rasonque, the RAS correction, enamel gel, Toregem's funded Phase 2, oral-systemic health links — each with primary receipts and freshness dates; draft outreach for John to longevity and dental-innovation contacts this week.
$ Citation-first health capsules are HKG's core inventory; clinics pay for verified patient-facing signal.
Tape: "10 drones made 19 million gallons in three hours." Rainmaker's own Alaska report says 2 drones, 19 flares, 7 missions, model-estimated 18.8M gallons (range 13.6-29.1M) — with no third-party check; GAO puts seeding gains at ~5-15% per storm. Salim called the proof the breakthrough, but the proof is self-graded. The atmosphere's 3.4 quadrillion gallons checks out (USGS). Weather modification is becoming a market with no independent verifier — plus downstream water-rights tension Salim flagged — exactly the vacuum the gardens exist to fill.
▸ Draft the Weather-Mod Claims Ledger: a TKG/AgKG capsule set separating measured from modeled rainfall claims, every vendor number labeled self-reported, plus a spec-only Bet Board brief for a third-party verification service aimed at insurers, water districts, and downstream farmers.
$ Verification-as-a-service for a brand-new industry; insurers and water districts need an honest referee.
Flock self-reports 6,000+ communities, 49 states, 20B plate scans monthly; Darth Vader really testified at San Diego City Council (Aug 21). But the tape's "people literally dying from misread plates" has no documented case. The verifiable record is damning differently: a Kentucky officer tracked an ex 2,000+ times via Flock (arrested Aug 2026), 50+ stalking cases, wrongful gunpoint stops, a scream detector scrapped after backlash. Even the Franklin quote aired inexactly. The civil-liberties debate is running on folklore, and receipts win debates like this.
▸ Compile a surveillance-claims capsule pack — documented harms only, Flock's scale labeled self-reported, the exact 1755 Franklin text with context — and publish to the gated specimen cabinet with llms.txt exposure so trusted agents cite us, not folklore.
$ Journalists, city councils, and litigators cite whoever holds clean receipts; citation gravity compounds.
OpenAI's Jalapeño chip is real but the tape inflated it: the "54x throughput" is OpenAI's own best-case single-user point against an older GB200 (full range 8.6-104x); honest gains are 1.5-1.9x tokens per kW at 700W vs 1400W, all self-reported. "An idea a few months ago" is wrong — design began mid-2024, over two years. "90% of compute is inference" runs high; credible 2026 estimates say 60-70%. The direction is what matters to us: inference prices keep collapsing, making garden agent fleets cheaper every quarter.
▸ Re-run the gardens' sweep-cost model at falling inference prices and expand nightly agent sweeps accordingly; publish a Jalapeno-honest-numbers capsule; sanity-check x402 agent-pricing floors so machine-lane prices track the cost curve down instead of getting undercut.
$ Cheaper inference widens our agent-labor margin; honest chip-claim capsules feed FKG.
The tape mocked Deloitte's lunar number as a lowball and floated "$2 quadrillion per year." Record: Deloitte's range is $343B-$566B cumulative through 2050 — the aired $566B is the HIGH case — and the quadrillion figure is the show's own untraceable math; the "Starmind" roadmap is actually ~1 GW/yr of orbital compute by late 2027 toward ~100 GW/yr by 2030. Salim adds that consultancies ship stale data (a big-four agentic report used 2.5-year-old inputs). Calibrated, receipt-dated foresight is exactly the gap between those poles.
▸ Write the RDKG lunar-economy brief with every claim labeled VERIFIED, GUESS, or SELF-REPORTED, receipts and freshness dates attached; include BKG's angle — infrastructure, construction, power, and life support named the biggest early lunar markets — and file it to the Bet Board.
$ Foresight buyers drown in stale consultancy PDFs; dated, labeled briefs are the upgrade.
Verified: Tesla registered 79 robotaxi Model Ys in a single day (Aug 27), Texas fleet now 270 against Waymo Austin's 736; Nevada raised its robotaxi cap from 10 to 5,000 — the tape's "2,500 in Vegas announced" is Tesla's internal goal — and Cybercab's public launch hits Austin Sept 3, two days from now. Salim's downstream call: gas stations become autonomous-fleet depots; insurance, cleaning, and charging all reshape. Luminar is already liquidated. The physical buildout will be constructed, insured, and underwritten — garden territory.
▸ Schedule an agent sweep for the Sept 3 Austin Cybercab launch; draft BKG signal cards on depot retrofits and charging conversions plus an FKG note on the per-mile safety and underwriting data gap, all filed with receipts within the week.
$ Builders and insurers need verified fleet data before capital moves; we sell the receipts.
China's AI-companion rules are real and specific: CAC plus four agencies issued interim measures Apr 10, 2026 (effective Jul 15) — virtual romantic AI partners banned for minors, parental consent under 14, disclosure duties, two-hour break reminders. US states are moving; Alex's bet that some states keep it legal is a guess. Fertility corrected: South Korea is 0.76-0.8 and rising off 2023's 0.72 floor, not the falling 0.75 aired. Somnia's robot specs are its own "placeholders." Parents, schools, and health systems have no verified tracker for any of this.
▸ Build the first verified AI-companion regulation tracker as an EdKG/HKG capsule pack — actual regulatory text, effective dates, jurisdictions, corrected demographic numbers — agents draft and source it end-to-end this week, labeling every vendor spec self-reported.
$ First-mover citation authority on a rulebook every parent, school, and platform will search.
Verified: Apple's Mac Studio M5 Ultra ships with 512GB unified memory (the M6 is Apple's first 2nm chip), enough to run frontier-class open models on a desk. Salim's on-tape read: the winners are law firms and HIPAA-bound healthcare that can't use the cloud — four clustered Studios approximate a small private data center, shifting economics from per-token rental to an owned capital asset. Gardens' verified corpora served locally over MCP, with The Call as the voice front-end, fits those clinics exactly.
▸ Write a spec-only Garden-in-a-Box brief: verified corpus plus MCP endpoint on an on-prem Mac cluster for HIPAA-bound clinics, The Call as the interface; size the market and draft John's healthcare one-pager. No code — Garden Engine backlog only.
$ HIPAA-bound clinics are underserved, budget-holding buyers for private verified knowledge.
The energy segment needs correcting before anyone cites it: US natural gas runs $2.87-3.44/MMBtu (EIA), not the "$6" aired; Actinide's 15.38% enrichment is 5x commercial reactor fuel (3%), not "beyond government labs" (Centrus hit 19.75% in 2023; weapons labs 90%+ decades ago); China's solar surge direction is confirmed (Ember: +636 TWh global in 2025, China more than half) but the tape's exact 8x monthly figures aren't pinned. The real verified signal: every SMR promises 2030 while fuel supply lags — a live HALEU bottleneck.
▸ Compile ten energy ground-truth capsules — EIA gas prices, Ember solar data, the enrichment landscape, the HALEU supply gap, solar's verified ~8,000x potential — with primary receipts, and pair them with the data-center evidence pack so siting debates cite gardens, not vibes.
$ Energy numbers decide siting fights; verified figures become the citation of record.
Standing reality check. The blocking engineering P0 is unchanged: BKG auth fix plus paywall consolidation through to the first paying customer. Nothing on this board pulls an engineering hour off it. Every card above ships as what AI agents, AI video tools, and drafted documents can do autonomously within days — sweeps, capsule packs, corrected-number explainers, spec-only briefs, outreach drafts. Anything that needs code gets filed to the Bet Board as a spec and waits its turn behind the first dollar of revenue.
▸ Hold the line: run every card as agent work through existing production tools; file anything build-shaped as a spec-only Bet Board entry; re-verify weekly that zero engineering hours moved off BKG auth and paywall.
$ The first paying customer is worth more than every unbuilt idea above it.