Independent capability brief · Not affiliated · Passive public reads only · 2026-07-14
Bechtel: the safety record is real. The safety proof layer is the open seat.
Four evidenced observations from bechtel.com's public surface, read the way an AI agent reads it, and a concrete concept for the layer Bechtel's own "zero incidents" trajectory points at: machine-verified proof of safe work, per task, per subcontractor, with a public settlement receipt.
Independence & method. Unsolicited concept work by EcoWealth Corporation, not requested, reviewed, or endorsed by Bechtel. Every observation below is from normal public page loads (curl headers, fetched-HTML JSON-LD (the invisible structured labels search engines and AIs read) counts, robots/well-known reads with a soft-404 control). No scanning, no probing, no authentication attempts, no bot walls routed around. Trademarks referenced descriptively. Every statistic is cited and dated; live sources override any number printed here.
Credit first: this baseline is strong
Bechtel's safety substance is real and stated plainly on its own pages: "Our goal is clear: zero lost time incidents and zero recordable incidents — and we are within reach"; "Nothing is more important than the safety of our colleagues"; every employee has stop-work authority; 2023 rates of 0.014 LTIR / 0.163 RIR sit near the industry floor, with a record of 263 million work hours without an accident. The web estate is well-built: a nonce-based CSP, HSTS (max-age 63072000, includeSubDomains), X-Frame-Options, nosniff, a comprehensive Permissions-Policy, on WP Engine behind Cloudflare, and real 404s (control confirmed). None of the findings below are "your site is broken": they are about a different reader (the AI agent) and a next layer (task-level verified proof) that Bechtel's own zero-incidents goal already points toward.
The bottom line, in plain English
Answer “was this done safely?” with a receipt → you win the safety trust that wins the next job
Owner, insurer, and regulator AIs are already starting to ask whether a specific task was performed safely, and today the only answer is a document request — a machine-verifiable safety receipt lets you answer with proof on the spot, which is exactly the trust that keeps you shortlisted.
Prove your “zero incidents” goal per task, per sub → a live record worth more than any after-the-fact summary
Your world-class rates live as prose an AI can't verify; one settlement receipt per hazard-controlled task turns “within reach” into something an owner or insurer can audit directly — the kind of verifiable safety record that lowers risk and earns its keep at megaproject scale.
Small housekeeping → cheap protection and an on-ramp for an already-strong estate
Trimming the QA tool out of your live security policy and adding structured labels to your safety pages won't change the site anyone sees, but they're the low-cost first step that lets machines read the real commitments you already publish.
Four findings, each with its receipt
Finding 1 ★
A top-2 U.S. contractor with a world-class safety goal has no way for a machine to verify a single safe task.
The safety commitment lives entirely in prose. There is no agent front door and no structured safety object: bechtel.com/llms.txt and /ai.txt are true 404s, every .well-known/ agent path is a true 404 (the .well-known/* paths serve Bechtel's full branded "Page not found" page at a genuine 404 status; we ran a soft-404 control), and robots.txt is fully open.
$ curl -s -o /dev/null -w '%{http_code}' https://www.bechtel.com/llms.txt404$ curl -s -o /dev/null -w '%{http_code}' https://www.bechtel.com/.well-known/agent-card.json404(serves "Page not found - Bechtel", real 404 status)$ curl … /this-is-a-soft404-control-xyz123.txt404(548b — control: real 404s, credited)$ curl -s https://www.bechtel.com/robots.txt200— Disallow: /wp-json/ only; agents fully allowed
Why it matters: AI is already arriving in owner, insurer, and regulator workflows. When one asks "was this weld, this trench, this lift performed safely, and can I verify it?" today the answer is a document request. The megaproject contractor that answers it with a receipt owns the next decade of safety trust.
Finding 2
The commitments are real; none of them are machine-verifiable.
Both the homepage and /approach/ (Bechtel's safety-and-sustainability hub) carry exactly one JSON-LD block apiece, and it is the generic Yoast SEO graph (WebSite, WebPage, BreadcrumbList, SearchAction). There is no Organization node, no project schema, and no ESG Dataset. The safety copy is dense (45 mentions of "safety" on /approach/) and the ESG language is real ("operate on 100% renewable energy by 2025" at Quebrada Blanca Phase 2), all of it prose an agent cannot parse into verifiable fields.
Why it matters: every one of those numbers (incident rates, renewable-energy targets, waste) is a claim an owner's AI, an insurer's AI, or a regulator's AI will eventually want to verify, not read. Structured data is the on-ramp; proof-of-outcome receipts are the destination.
Finding 3
Subcontractor and per-task safety truth is self-attested prose, not verifiable proof.
Bechtel self-performs and subcontracts across megaprojects staffed by 50,000+ people. The safety story a partner, owner, or regulator receives is exactly what /approach/ publishes: a narrative of goals, rates, and stop-work culture. What no format on the site can contain is task-level verified proof: that this excavation was inspected before entry, by a competent person, with the protective system in place, on this date.
/approach/ (public page) — verbatim:"Our goal is clear: zero lost time incidents and zero recordable incidents — and we are within reach.""Nothing is more important than the safety of our colleagues… everyone… returns home safely."2023: LTIR 0.014 · RIR 0.163 · record 263M work hours without an accident (press)→ all narrative; no per-task, machine-verifiable safe-completion object exists publicly.
Why it matters: at megaproject scale the value of per-task verification is highest: one settlement receipt per hazard-controlled task builds a live, auditable safety record that beats any after-the-fact summary, and it is exactly where a "within reach" zero-incidents goal becomes provable rather than asserted.
Finding 4
One hygiene note inside an otherwise strong estate.
Bechtel's production Content-Security-Policy allowlists the Marker.io QA/feedback tooling (app.marker.io, api.marker.io, media.marker.io, edge.marker.io) across connect-src, frame-src, img-src, form-action, and media-src: a website QA tool wired into the live public policy. Stated for completeness; a one-line scope trim, not a vulnerability.
Why it matters: minor, and the CSP itself is doing its job. Noted only so the brief is complete and honest: the same "credit the strong baseline, flag the small thing" standard applied to every target.
The concept: safety-verified work packets
EcoWealth runs the Ecological Work Protocol (EWP): a funded, proof-verified, on-chain-settled work-packet layer, live on Base mainnet today (contract 0x76c17C…A14B, source-verified; first packet settled end-to-end as workId 14). Pointed at Bechtel's world, one packet = one task with safety made structural, not supervisory:
1 · Hazard-gated authoring. A packet cannot be created unless it passes a machine-enforced hazard gate: hazard class, controls, and framing are validated in CI before a task can direct anyone's hands. This gate runs on our live library today (a non-compliant definition is rejected automatically).
2 · Proof of safe completion. The crew closes the packet with photo + GPS + signature evidence; a deterministic proof-checker (agent-payable, $0.01/call) returns a verdict before approval. No proof, no settlement: pay follows the proof.
3 · A receipt that compounds. Settlement writes a public receipt on Base. Each receipt appends to the project's verified safety and ESG record: a machine-verifiable answer to "was it done safely?" and a tCO2e/waste/water data point born structured, not PDF.
Integration path: zero Bechtel system access required
The pilot deploys entirely on EcoWealth infrastructure first: the Agent Kit (see the kit page) runs read-only against one public Bechtel project page, and one safety-verified packet is funded and settled end-to-end on our live board, proving the concept before anything touches Bechtel's stack.
Evidence log
#
Claim
Source
Observed
Re-verify
1
llms.txt / ai.txt true 404
/llms.txt, /ai.txt
404 · 548b
curl -o /dev/null -w '%{http_code}' …/llms.txt
2
.well-known agent paths true 404 (branded 404 page, real status); control 404
Step one is a receipt, not a meeting. For roughly $750 (a rounding error against Bechtel's ~$15.9B revenue) EcoWealth funds and settles one proof-of-safe-completion packet shaped like a real hazard-control task (e.g. pre-pour trench protective-system verification): hazard-gated so it cannot be authored without the required controls, closed with photo+GPS+signature proof, settled on Base. The deliverable is the argument itself for Bechtel's "zero incidents" goal: a permanent on-chain receipt that this task was done safely, verified, and paid.
Want it cheaper and faster? Have any Bechtel AI agent make one $0.10 work-packet x402 call (x402: a web-native way for AI agents to pay per use) against our live endpoint right now: menu at vealth.net/.well-known/x402, contract 0x76c17C…A14B. The on-chain receipt lands in seconds; that receipt is the whole pitch.
Inbound-only: nothing here is outreach; this first step is offered only if a Bechtel conversation opens. Contact: [email protected].
