A respectful, evidence-first read of WM's public web surface and public company record: what a green-bond auditor or an AI agent can and cannot verify today, and the one layer that would change it. Baseline credited first; then four findings, each reproducible from a URL and an HTTP status.
WM runs a genuinely strong estate, and a reasonable web-security baseline:
max-age=63072000; includeSubDomains; preload), a Content-Security-Policy with frame-ancestors https://*.wm.com (clickjacking protection), a fully-open robots.txt (welcomes crawlers), a well-formed 9,076-URL sitemap, and a real 404 handler (the soft-404 control returns a true 404).The gap is not the operations or the ambition. It is that the enormous environmental story has no per-site object anyone outside WM can verify, and, secondarily, that the machine surface (agent discovery, structured data) trails the security baseline.
No jargon: here is what each fix below is worth to WM in plain terms.
That per-site proof is exactly the credibility green-bond and ESG capital pays a premium for: it turns WM's enormous environmental story into something an investor or regulator can independently check.
Today the standard agent handshake hits an access-denied wall; a two-line fix lets raters and assistants read WM instead of bouncing off, for the cost of a config change.
WM's flagship progress is invisible to the JavaScript-free readers ESG raters use, so making it structured means the good numbers actually count instead of being missed behind a loading screen.
Every active landfill carries recurring environmental-monitoring obligations (methane and ambient-air surveys, groundwater and leachate screens including PFAS, stormwater-outfall checks); every closed/capped landfill carries restoration and revegetation obligations; every field crew carries safety obligations (confined space, silica, site-safety walks). Those are exactly the monitoring, restore, and build work-packet families EcoWealth already runs. Today the survey records, restoration sign-offs, and safety sheets live in site files and a PDF sustainability report; a green-bond investor, a regulator, or a community group has no independent way to confirm a specific task was completed at a specific place and time.
ambient_air_survey 164 · pfas_screen 159 · watershed_outfall_screen 251 · stormwater_inspection 227 · soil_carbon_baseline 155 · biodiversity_survey 218 · coastal_habitat_survey 233 · tree_planting 221 · confined_space_entry_survey 37 · construction_site_safety_walk 57 · silica_dust_exposure_screen 35. Re-verify: curl -s https://vealth.net/labor/statsThe entire /.well-known/ tree, agent-card.json, ai-plugin.json, x402, security.txt, even the directory itself, returns HTTP 403 (WAF-blocked). /llms.txt and /ai.txt are true 404s (the soft-404 control confirms). So an A2A or MCP client (the standard formats AI agents use to discover and call a service) that follows the standard and probes /.well-known/agent-card.json doesn't get a clean "not here": it gets an access-denied wall. That is a worse signal than a 404: it reads as hostile to agents. And it directly contradicts robots.txt, which is fully open (User-agent: *, no Disallow) and invites crawlers in.
/.well-known/ path prefix (so agent cards and a future security.txt can be read at all), then publish the files. Right now the standard agent-discovery handshake fails closed against WM, even though the crawl policy says "come in."for p in /.well-known/agent-card.json /.well-known/x402 /.well-known/security.txt /llms.txt; do curl -s -o /dev/null -w "%{http_code} $p\n" https://www.wm.com$p; done → 403 / 403 / 403 / 404.A normal, non-JS fetch of www.wm.com/us/en returns a client-rendered Next.js shell: <title>Loading interface…</title>, __NEXT_DATA__ present, 0 application/ld+json blocks and 0 og: tags. The flagship sustainability narrative, the $3B program, the 2030 target, the RNG network, lives on a separate domain (sustainability.wm.com, which carries exactly one ld+json) and in a downloadable PDF; www.wm.com/us/en/sustainability is a 404. An AI agent asked to compare WM's verified recovery and emissions progress against its stated targets, starting from wm.com, has no structured object to read.
Organization + Dataset object on the main surface, and, better, per-site outcome receipts, gets read once and read right, instead of missed entirely behind a loading placeholder or paraphrased from a PDF.curl -s https://www.wm.com/us/en | grep -c 'application/ld+json' → 0; curl -s https://www.wm.com/us/en | grep -oiE '<title>[^<]*' → Loading interface…There is no readable security.txt: both the root /security.txt and the standard /.well-known/security.txt return 403, and because the whole /.well-known/ tree 403s, a researcher can't even reach the standard location where one would live. For a critical-infrastructure operator of WM's scale, a published, machine-findable disclosure channel is table stakes.
security.txt (once the path block is lifted) is the courtesy that keeps a found vulnerability arriving by email instead of by headline, and it pairs naturally with publishing the agent card at the same location, closing findings 2 and 4 in one change.curl -s -o /dev/null -w '%{http_code}' https://www.wm.com/.well-known/security.txt → 403; /security.txt → 403.A single site's quarterly methane/air survey and its closed-landfill restoration obligation each become an EWP proof-verified work packet: bounded scope with safety framing, claimed by a field technician or a restoration crew, closed with photo + GPS + signature proof scored by a deterministic proof-checker, and settled to a permanent receipt on Base. Each receipt appends to that site's, and that crew's, verified completion history: the independently-checkable upgrade to a sustainability-report page, on the same rail that can attach a real ecological retirement receipt to the tons recovered or the methane captured.
Live proof it's not slides: EWP contract 0x76c17C…A14B on Base mainnet (source-verified, first settled packet workId 14); the priced tool menu at vealth.net/.well-known/x402; and the end-to-end building-agent golden path.
| Step | What ships | Who owns it |
|---|---|---|
| 0 · Prove it (today) | One agent pays $0.02 to sdg-assessment; one funded monitoring or restoration packet on one site settles on-chain. | EcoWealth rail, live now |
| 1 · Machine-legible surface | Lift the /.well-known/ 403; publish llms.txt + agent card + JSON-LD on the main domain + security.txt. | WM web/security team (drafts provided) |
| 2 · Proof pilot | 10 monitoring + restore + safety packets on one landfill, each closed with proof + settled; verified completion history per site + crew. | Joint, bounded scope |
| 3 · Standing rail | Per-site survey + restoration receipts as the default output; retirement receipts attached to the tons recovered and the methane captured. | Operator decision |
Two figures, both pocket change against a ~$25B-a-year company:
sdg-assessment endpoint and gets a bounded, standards-scored assessment back, keyless. Dry-run client: vealth.net/recipes.html.