Independent · unsolicited · we took a chance at optimizing a painful workflow
EcoWealth runs a proof-verified, tamper-evident, method-templated work-packet primitive, live on Base mainnet. This brief reframes that primitive as an Evidence Workpaper Protocol (EWP): it turns notice responses, deduction substantiation, clean-energy-credit proof, preparer due diligence, and amended-return evidence into structured, agent-legible, audit-grade workpapers with a preserved proof chain, without touching any real taxpayer data.
Everything here was learned by browsing irs.gov the way any taxpayer or preparer does: public pages, response headers, redirect chains, robots.txt, and standard agent-discovery probes. The point is an opportunity, respectfully framed, on top of a genuinely strong baseline.
✎ Not a teardown, and not an IRS project. Nobody asked us to; we just took a chance at making a painful workflow better, on our own initiative. This is an independent concept prepared by EcoWealth Corporation. It is not affiliated with, not endorsed by, and not in partnership with the IRS or any government agency. Every packet shown is fabricated, anonymized demo data.The value exchange
llms.txt, an A2A (Agent-to-Agent) agent card, and MCP (the interface an AI assistant calls to use a service correctly) tool schemas so any assistant can help a taxpayer assemble evidence and always route real questions back to an official resource.noindex, never deployed to any government system.Start with the deep audit, then the three-document walk
The ruthless second pass: a wider passive sweep (seven findings with inline proof + a full evidence log), plus a public-record read of how the agency funds, procures, and builds software: the IRA modernization money (now heavily rescinded) and its Strategic Operating Plan (now in a strategic pause / transition), the Paperless / Document Upload Tool progress, the EDOS and Enterprise Case Management (ECM) programs, and a quantified efficiency proof showing exactly which mail round-trips a structured workpaper removes.
Read the deep audit →The honest baseline (it's strong), four evidenced findings with inline curl/DOM proof, an evidence log, and exactly where the Evidence Workpaper Protocol plugs in.
Read the brief →A clickable, brand-matched walk of one demo packet (a CP2000-style notice) from draft to resolved, with the evidence checklist ticking and an audit-grade resolution receipt.
Watch it run →The actual file shapes that make the workpaper lane legible and callable by any AI, with a concierge persona that never overstates and always routes to a human.
See the files →Four things we noticed, respectfully, expanded to seven in the deep audit
Agent-discovery is greenfield. /llms.txt (the file AI assistants read first), /.well-known/agent-card.json (the file that lets an AI assistant use a site correctly), /.well-known/ai-plugin.json and security.txt all return 404; robots.txt (the file that tells search engines what to index) names no AI crawler. Nothing is built for agents yet: a first-mover position, not catch-up. 404 × 4
Evidence is document-and-mail-shaped. The notice pages tell a taxpayer to "reply by the date listed" and "if a response form is included, complete and sign it": there's no structured, machine-legible, verifiable workpaper format. This is exactly where EWP plugs in. DOM proof
Deep-link citations age. A widely-bookmarked notice URL 301-redirects to a renamed slug (cp2000-notice → cp2000-series-notice); the online-account URL likewise. Content survives, but an AI's citation drifts. redirect
Header hardening headroom on a solid base. HSTS, nosniff, and X-Frame SAMEORIGIN are present; Content-Security-Policy, Referrer-Policy, and Permissions-Policy are not. Incremental, not a vulnerability. headers
What a dedicated evidence-workpaper provisioner would run · prototype roadmap
EcoWealth's real offering is a dedicated, always-on provisioner: here, a prototype pointed at one thing: turning mailed evidence into a verifiable object. This is the first ninety days it would run, each step a felt efficiency, all on fabricated data and EcoWealth infrastructure. It is a demonstration and a proposal, not an IRS engagement, adoption, or claim of use.
A prototype llms.txt + agent card so an assistant can help a taxpayer assemble the right evidence and always route real questions to an authoritative resource at irs.gov. Fewer misrouted taxpayers, from day one.
The CP2000 demo, end to end: structured requirements, attached evidence, and a proof chain that closes on one precisely-named bounce, never an open-ended "send documentation." On fabricated data.
A vendor-neutral evidence-packet + resolution-receipt layer that sits above existing case systems (ECM / Pega / the Document Upload Tool), not a replacement, adding the append-only audit trail and replayable receipt the upload flow lacks.
The hand-off
We're not claiming the IRS uses this, and we're not asking for any system access. We're showing a rail that could sit underneath administrative casework: structured evidence, an explicit next step, and a proof chain a resolved item can be audited against, demonstrated end-to-end on fabricated data. Start with the brief, then watch a notice resolve as a workpaper.
The same primitive, as open agentic commerce
Everything above is a demonstration on fabricated data, not an IRS system, program, or claim of adoption. The work-packet primitive beneath it, though, is a live public endpoint. Here is how any institution's agent (not the IRS) could exercise it today.
Fund it a few cents. Give any AI agent (Claude, GPT-class, Llama, or your own) a Base wallet holding a few cents of USDC. No account with us, no API key, no sales call.
Point it at the endpoint. The work-packet program answers HTTP 402 with a standard x402 (a web-native way for AI agents to pay per use) payment schema; the agent signs a USDC authorization and pays $0.10. A runnable client with a dry-run default is published at vealth.net/recipes.html.
Get a verifiable packet. Back comes a bounded, proof-verified work packet, and every step the agent took (payment, settlement) is an on-chain receipt anyone can verify, forever. The EWP contract is live on Base mainnet; workId 14 settled the first packet end-to-end.
Why this isn't scary
The agent cannot do anything invisible here. Every action embeds its own receipt: permanent, public, verifiable. That is more accountability than most software offers, not less: the same tamper-evident-proof idea behind the workpaper above.
The bet
This is a demonstration, not an IRS program, adoption, or endorsement. The shift underneath is real, though: agent-discovery is still greenfield almost everywhere (as the findings above note), and AI competence is scaling. Any institution's agent can exercise this open, on-chain rail today for $0.10; being early costs an afternoon; being late just means the proof-verified track records stay unbuilt while the standard is still greenfield.
One concrete thing to try: have your own agent (not the IRS) make one $0.10 call against the live endpoint and watch a verifiable receipt appear. The demo above, exercised for real, on public rails, on fabricated data only.
