Built primarily from a completed passive-recon deep-dive; this pass re-verified five load-bearing figures directly against live Carbonfuture pages and ran one fresh agent-discovery probe the original recon didn't cover. Every claim below cites its source and, where applicable, a re-runnable command.
Carbonfuture's own due-diligence page names nine assessment categories verbatim: Corporate/Reputational, Project/Operational, Financials, Carbon Accounting, Feedstock/Sustainability, Additionality, Permanence, Regulatory Compliance, Social/Environmental Impacts, each backed by "more than 100 data points and over 30 proof points," applied to "more than 100" projects over four years. Permanence alone is assessed via "accredited laboratory tests (such as H/Corg and heavy-metal analyses)." This is a real, professional-services-grade evaluation pipeline, the kind of rigor that makes a 1.24Mt, 10-year Microsoft offtake bankable in the first place.
Carbonfuture's own 2025 year-end announcement states "1.5 Million Tonnes Facilitated" for the year. The Microsoft × Exomad Green agreement (a 10-year, 1.24-million-tonne biochar offtake, tracked by Carbonfuture MRV+) is described elsewhere as the industry's first megatonne biochar carbon removal deal. The Series A2 close was led by SIX (operator of the Swiss Exchange), with Idemitsu Americas Holdings participating. That's notable because SIX is simultaneously a durable-removal buyer and now Carbonfuture's lead investor, a capital-markets operator vertically integrating into the trust-infrastructure layer it also buys from.
Carbonfuture's own financial-architecture article states plainly: "70% of suppliers expect to raise capital within six months and 43% within three months," and names "the near-absence of $1–5 million investors" who could bridge grant/philanthropic money to institutional project finance. This gap is real and Carbonfuture is right to name it, but it operates at the plant/project level (single-digit-million-dollar CAPEX, multi-year construction), a different unit of scale entirely from EWP's sub-dollar, single-task packets. Packet-sized funding does not fill a $1–5M equity gap. What it plausibly could fund: smaller-ticket ecological labor adjacent to a missing-middle plant once financed and operating, feedstock collection, invasive-species clearing, site prep, ongoing maintenance, which is squarely EWP's actual unit of work.
Carbonfuture/Puro/Isometric prove how much carbon, how durably, at what confidence interval for a population of tonnes: calibrated weighbridges, accredited-lab Corg/H-ratio and PAH assays, reactor telemetry at ≥2% accuracy, statistical batch sampling with a 5% materiality threshold, independent VVB sign-off before a credit issues. EWP's proof packet (photo, content-hash + EXIF cross-checked; geofenced GPS; a one-time actor signature, EOA ecrecover or fail-closed ERC-1271/6492 for contract wallets) proves that a bounded task happened, by whom, where: binary pass/fail, no population sampling, no independent human auditor, same-transaction on-chain settlement. Neither system substitutes for the other; the credible bridge named here is EWP as a last-mile field-capture input feeding a Carbonfuture-style aggregation dossier, never a parallel biochar-credit-issuance path.
We name this ourselves, before any Carbonfuture-side reader would have to find it: our proof-hash script implements a real, reproducible sha256-of-canonical-JSON convention over a packet's evidence bundle (photo/GPS/signature refs), deliberately modeled on Regen Network's content-hash-IRI pattern. Read directly against our live settlement route this session, the hash gets computed and returned/logged, but the on-chain write is still a template string (commented "in production"), not an executed transaction, and there is no public read endpoint serving a recomputable bundle for a given workId. This is the exact "full provenance behind each certificate" bar Isometric's own registry already clears publicly. Naming our own gap plainly is part of the honesty this brief is trying to model.
The original recon didn't check this dimension, so this pass ran it fresh. robots.txt (the file that tells search engines what to index) is real (200, 51 bytes, points at a real sitemap) and sitemap.xml is real (200, 28,769 bytes). But llms.txt (the file AI assistants read first), llms-full.txt, ai.txt, and openapi.json all return a TRUE HTTP 404 with a byte-identical (906 B) Webflow-branded error page, matching a nonsense-path control exactly. That's the honest outcome (a real 404, not a fake "found"). Separately, every /.well-known/* path (agent-card.json, ai-plugin.json, security.txt, x402) returns a distinct 404 body (88 B, "Invalid .well-known request"), a documented Webflow-platform behavior (Webflow reserves that namespace for its own Apple-Pay domain association), not something Carbonfuture itself configured. Net: zero agent-legible surface today for any AI representing Carbonfuture or looking to integrate with it, but at least an honest zero, not a misleading one. carbonfuture.com mirrors the same true-404 pattern.
At capture time, Carbonfuture's site-wide news banner reads: "Boeing and Carbonfuture Sign Multi-Year Agreement for at least 40,000 Tonnes of Durable Carbon Removal." Useful texture, not a headline finding: Carbonfuture already operates across a wide range of deal sizes, from the 1.24Mt Microsoft megadeal down to tens-of-thousands-of-tonnes multi-year agreements. The "missing middle" plant-finance gap this brief discusses sits below even that smaller deal, at the pre-offtake, pre-bankable stage, reinforcing that packet-sized labor funding is an adjacent-to-the-plant play, not a way to reach deal-level scale directly.
