The Playbook

The thesis under this move — non-toxic materials cross cost-parity this decade — just got stronger and more honest at the same time. A peer-reviewed dataset now shows mycelium insulation matching extruded-polystyrene thermal performance (λ ≈ 0.031 W/m·K) with carbon-negative potential. That graduates the performance half of the thesis to verified. The cost half is still unproven — no primary price dataset exists yet — so the thesis stays a deliberate coin-flip, and we say so.

Alongside it: low-clinker LC3 cement cleared standards qualification, tall mass timber is in the model building code up to 18 stories, and — the cautionary note — the EPA has proposed rolling back four of six PFAS water limits, a reminder that toxics governance is reversible and independently verified data only gets more valuable when the federal floor wobbles.

The smallest real test: a BKG pilot puts the non-toxic spec in front of general contractors and watches whether they actually open and cite it in a real bid. What would kill it: at parity, they still don't — habit and supply chains outweigh a verified spec. Pre-build the schema and citation trail now so it drops in the moment the cost data lands.

The longevity arm is not a bet on a molecule, a clock, or a founder — it's a bet that in the most hype-saturated field in health, verification is the product. The recent sweep makes the case on its own: an oral GLP-1 pill (orforglipron) genuinely cleared the FDA; Lp(a)-lowering drugs (olpasiran and peers) cut the marker up to ~99% but have not yet proven they prevent a single heart attack; a new proteomic aging clock predicts mortality in 37,000 people; senolytics rejuvenate aged mouse organs but have no human healthspan proof yet — and "plasma exchange reverses aging" is a paid clinic claim the best controlled trial actually contradicts.

Five findings, five different tiers. The house rule does the work: peer-reviewed lab science and founder-branded commercial claims are different objects, weighted separately, even from the same person. That discipline is the sellable asset.

The smallest real test: produce one verified claim-card — "does NMN extend healthspan in humans?" — in the standard tiered, sourced, conflict-disclosed format, and ask John Bou whether it's more decision-useful than what exists today. What would kill it: patients, clinicians, and investors don't pay for verification over a charismatic founder's story.

Cheap inference drives more compute, which drives electricity — and now there's a government primary that says so plainly: the EIA's Annual Energy Outlook 2026 names data centers the dominant driver of long-run US electricity growth, implying installed capacity must expand 50–90% by 2050. That converts "AI needs power" from a talking point into a scored trajectory you can cite.

The supply side is moving in verifiable steps. Commonwealth Fusion became the first fusion company to file for grid interconnection (a 400 MW plant, in PJM's queue in Virginia's data-center corridor). General Fusion is set to be the first publicly traded pure-play fusion company. Neither is firm power on the grid yet — an interconnection application is not a signed PPA, and a plasma result submitted for peer review is not net energy — and saying exactly that is the product. The value is a feed that tells an investor which milestone is real.

The smallest real test: package the existing energy cards into one gated brief and see whether a single energy investor or operator will pay for it over free trade press. What would kill it: a later EIA outlook revises data centers out of the dominant-driver role, or buyers find the free coverage good enough.

This is the load-bearing move — the one the whole thesis points at. Two forces are crossing: the price of AI reasoning fell roughly 280× in eighteen months, and the public text used to train models is estimated to run dry around 2026–2032. When thinking is nearly free and scraped text is exhausted, the scarce, defensible thing is verified, physical-world ground truth — exactly what the gardens measure.

The move is to expose that ground truth as a metered feed for machines. The rails just arrived: AP2 (Google's agent-payment protocol) and x402 (now a vendor-neutral Linux Foundation standard backed by Visa, Stripe, Coinbase and others), with MCP consolidating as the standard interface agents use to reach tools and data. RDKG already keeps an MCP stub and an llms.txt manifest, so turning the feed on is a configuration change, not a rebuild.

The smallest real test: one external agent completes a single paid frontier_brief pull over an x402/AP2 paywall. The first non-zero, agent-settled transaction proves the business one level down. What would kill it: agent payment rails fail to standardize, or no agent will pay a premium for verified-over-scraped data. Held honestly at signal strength until that first settled pull.

Everyone will soon have an AI tutor; that isn't the opening. The opening is trust. The most-hyped AI school bans chatbots in class because ~90% of kids use them to cheat — which tells you the scarce thing in education isn't another model, it's verified, cheat-resistant work with a human coaching layer. AllowED earns that trust the hard way first.

It automates VA GI-Bill certification — the most regulated, audited workflow on campus: real federal money, real student data, zero tolerance for error. The engine runs each case through the same seven-step logic a School Certifying Official applies and hands back a recommendation; the human decides. The rules came from an actual SCO's years of judgment, not a generic template. Once a campus trusts AI with veterans' benefits, "can AI touch student data at all?" is already answered yes.

Be precise about what's built. The certification engine is live and passing its regression suite; a crosswalk covers hundreds of programs; an SDSU pilot is underway. The biggest risk is that the pitch materials describe security features (SSO, MFA, audit logging) in language that runs ahead of the confirmed build. Close that gap before selling — a platform whose whole promise is "trust us" can't afford daylight between the one-pager and the code. The three-pillar expansion is a draft under Paulina's review, not a shipped feature.