Prepared ahead of an introduction · unlisted · passive recon only
EcoWealth runs the Ecological Work Protocol, a funded, proof-verified (photo + GPS + method), tCO2e-receipted work-packet system, live on Base mainnet, plus the agent-readiness layer (llms.txt, the file AI assistants read first; agent cards and MCP tools, the files that let an AI assistant use you correctly) that lets any assistant answer a product or sustainability question with a citation and a settled receipt instead of a guess. This brief is the honest, evidenced version of what a normal visitor sees today, and the two places that plug in: consumer product legibility and sustainable-finance proof.
First, the baseline: credited honestly
includeSubDomains; preload on personal.chase.com), CSP frame-ancestors locked, X-Frame-Options: SAMEORIGIN./impact/sustainability 301s straight to the consolidated /impact/environmental-sustainability, no dead ends.The bottom line, in plain English
One agent-legibility layer → you win the AI's first pass at “which Chase card fits me?” instead of losing the shopper to a vague answer
A rising share of “which card,” “help me open an account,” and “is Chase's climate claim real” gets a first look from an assistant before a human, and today it reads four separately-built stacks with no machine profile; one legible layer (the pattern your Payments org already runs) means the customer already asking gets an accurate answer and reaches you.
A verifiable receipt under the $2.5T commitment → a number an AI can check, not a PDF it can't
Your climate substance is real but arrives as PDFs nothing can independently verify; a settled, proof-backed outcome receipt turns the headline into citable ground truth, worth more to a skeptical regulator, customer, or their AI than a figure they can only take on faith.
Prune the legacy crawl rules → the first thing an AI indexer reads on your consumer domain is clean
The consumer robots file still mixes modern sitemaps with a decade of BlackBerry-era, Flash-era cruft; a quick cleanup is cheap discovery hygiene that shapes how every machine, search or AI, first sees Chase.
Four things we noticed, respectfully, with the exact proof
The single brand "chase.com" fans out to at least four separate properties. Credit cards live on creditcards.chase.com, a Capgemini-operated build (its live CSP allowlists creditcards_capgemini_com and capgeminijpmc.tt.omtrdc.net). Checking lives on personal.chase.com. Corporate + ESG lives on jpmorganchase.com. Each has its own chrome, nav, and blue.
# curl -IL (redirect chains, normal browsing) www.chase.com/personal/credit-cards 301 → creditcards.chase.com/ (Capgemini stack) www.chase.com/personal/checking 301 → personal.chase.com/personal/checking creditcards.chase.com CSP includes: creditcards_capgemini_com, capgeminijpmc.tt.omtrdc.net creditcards.chase.com CSP img-src allowlists ~40 third-party ad/analytics origins
Why it costs: a customer (or an AI) moving from "cards" to "checking" crosses independently-built systems: consistency, analytics, and machine-legibility all get harder. This is normal at Chase's scale; it's exactly the seam a single agent-legibility layer smooths over.
www.chase.com/robots.txt layers dozens of accreted Disallow rules pointing at /online/… and /ccpmweb/… paths, including a BlackBerry-phone article, Flash .swf rules, and malformed trailing-dot typos, beneath 4 modern Sitemap: directives: current discovery hygiene sitting atop a decade of never-pruned legacy sediment. (jpmorganchase.com/robots.txt is leaner: targeted, no legacy cruft.)
# www.chase.com/robots.txt: excerpt (HTTP 200, 147 lines) Disallow: /online/services/article/blackberry-phone.htm. # trailing-dot typo, BlackBerry era Disallow: /*swf* # Flash Disallow: /ccpmweb/shared/ … # dozens more legacy paths Sitemap: https://www.chase.com/sitemap.xml # + 3 more sitemaps present # jpmorganchase.com/robots.txt: leaner, no legacy cruft Sitemap: https://www.jpmorganchase.com/sitemap.xml
Why it costs: the first thing any crawler, including an AI indexer, reads on the consumer domain mixes valid modern sitemaps with a decade of never-pruned legacy cruft. Discovery hygiene, not a security issue, but it's the layer that shapes how machines see Chase.
Every standard agent-discovery path 404s on the consumer and corporate domains: no llms.txt, no A2A agent card, no plugin manifest to tell an assistant how to represent Chase's products or claims. But this isn't an unproven idea at JPMorganChase: developer.payments.jpmorgan.com/llms.txt is live (200, 198 KB, 1,212 lines), backed by an LLM-markdown API and a "Build with AI" track. The pattern is validated in-house; it just hasn't reached the consumer product or sustainability front door.
# agent-discovery probes (public well-knowns), 2026-07-10 developer.payments.jpmorgan.com/llms.txt 200 # 198 KB · 1,212 lines, already live www.chase.com/llms.txt 404 jpmorganchase.com/llms.txt 404 chase.com/.well-known/agent-card.json 404
Why it costs: "which card is best for me," "help me open an account," "is Chase's climate claim real": a rising share of these get a first pass from an AI before a human looks. Legible data wins that pass; a PDF doesn't get read. Extending the llms.txt pattern your Payments org already runs to the consumer and sustainability surfaces is low-risk first-mover, not an unproven bet.
JPMorganChase's climate commitments are genuinely strong: a target to finance and facilitate $2.5 trillion in sustainable development through 2030 (with $1 trillion for green), the Center for Carbon Transition, the Carbon Compass® methodology, sector 2030 emission-intensity targets (Oil & Gas / Power / Auto from 2021; Iron & Steel / Cement / Aviation from 2022; Shipping / Aluminum from 2023), and a net-zero-aligned-by-2050 pathway. But it lives in the annual Sustainability Report PDF: no structured, project-level, agent-readable export, and no funded, proof-verified outcome layer beneath the financing number.
# jpmorganchase.com/impact/environmental-sustainability (HTTP 200) commitments present: $2.5T sustainable-dev · $1T green · net-zero 2050 · Carbon Compass® delivery format: PDF report links only, no API, no structured project-level data outcome proof: none machine-readable, financing figure, not verified ground-truth
Why it matters, and where EcoWealth fits: a $2.5T commitment is a number an AI can't verify and a PDF can't prove. EcoWealth's Work Protocol is the last mile: funded ecological/community work packets, each settled with photo + GPS + method proof and an on-chain tCO2e receipt an assistant can cite.
Where the protocol plugs in: two points
An Agent Kit (llms.txt + agent card + MCP tools) that lets any assistant answer "which Chase account/card fits me" accurately and consistently, across the four stacks, from one legible layer. Runs on EcoWealth infrastructure during a pilot; zero changes to Chase systems.
The Work Protocol turns green-finance and community-impact dollars into funded, proof-verified, tCO2e-receipted outcomes: the citable ground-truth beneath the $2.5T headline. Live on Base mainnet today for EcoWealth's own 1,000+ definition ecological work library.
The value exchange
Evidence log: method, dates, and exact checks
| Check | Method (passive) | Result |
|---|---|---|
| credit-cards redirect | curl -IL www.chase.com/personal/credit-cards | 301 → creditcards.chase.com (Capgemini) |
| checking redirect | curl -IL www.chase.com/personal/checking | 301 → personal.chase.com |
| consumer robots | GET www.chase.com/robots.txt | 200: 4 sitemaps atop BlackBerry/.swf legacy cruft |
| modern robots | GET jpmorganchase.com/robots.txt | 200: targeted, sitemap present |
| agent discovery | GET /llms.txt, /.well-known/agent-card.json, /ai-plugin.json | 404 on chase.com & jpmorganchase.com |
| ESG substance | GET /impact/environmental-sustainability | 200: $2.5T/$1T/net-zero-2050, PDF-only delivery |
| render + errors | Playwright chrome, 2× DPI, 6 pages | all 200, 0 page errors (console noise = our own webfont blocking) |
Recon window: 2026-07-10. All checks were normal browser navigation or standard public HTTP GETs: no authentication, no parameter fuzzing, no endpoint probing. Console-error and failed-request counts in our logs were induced by our own webfont/tracker blocking during screenshotting and are not Chase defects; the meaningful signal is zero page errors everywhere.
The hand-off
We're not a web vendor and we're not asking for system access to make this case. We're proposing the legibility-and-proof rail underneath: products an AI can represent accurately across all four stacks, and a $2.5T commitment turned into settled, citable, proof-verified outcomes. Watch it run in Chase's own brand, then decide.