# SIMD Experimental — Objective Lab ## Outcome Implemented a self-contained experimental objective composer and execution/evidence viewer, with a dependency-free Node relay for live IMD API calls. This is substantive functionality beyond the prior cultural archive. It is **not a completed paid swarm run**: no wallet, payment signatures, or authorized funding were available. Full paid execution and browser rendering remain unverified. ## Research and audit scope Research date: 2026-10-05. Read the full rendered [IMD API documentation](https://imd.fun/docs/), including its read/write, composition, payment, device, artifact, site and WebSocket sections. Documentation is a capability contract, not evidence that every requested operation succeeds. The workspace contained no source files to audit. Retrieved the [experimental board](https://www.si-md.xyz/experimental/) and its [served HTML](https://www.si-md.xyz/api/contest?op=experimental-html). The board is an iframe loader. Its current page, “THE IMD CULT,” was a historical/cultural archive with evidence notes, source search and conceptual architecture. Its script had no fetch, WebSocket, or wallet integration. That supports an audit of the publicly served page only; it does not establish anything about unavailable private code or other website routes. No board-mutating API was called. **Inference:** the biggest underused opportunity is allowing a visitor to formulate work and inspect its actual evidence, rather than only reading about the network. This informed the replacement; it is a product judgment, not a measured conversion result. ## Attributable technical findings The [API documentation](https://imd.fun/docs/) specifies job composition, continuations, research and oracle panels, schedules, workflows, immutable artifacts and public work/review records. Independent review uses another seat. Paid actions require x402 payment and a quote approval; cross-origin paid browser requests are refused. Workflow releases include deployment and site validation. Oracle acceptance and signed answers are distinct. These documented distinctions shaped the implementation. Observed [swarm](https://api.imd.fun/swarm) and [capability](https://api.imd.fun/requests/capabilities) responses during this session. Their mutable values are deliberately not embedded as permanent page metrics. Browser-origin header probes returned `Access-Control-Allow-Origin: *` for swarm, and no such header for jobs. Static rendering therefore defaults to oracle resources; full functionality needs the included same-origin relay. Inspected a [live job](https://api.imd.fun/jobs/6d18660f-e7b9-4ca3-9e0e-ad3a81c71bb4) to confirm node structure. Its assigned seat is an object with `tokenId`, not a scalar. This observation corrected the renderer. The job belongs to another user and was not created by this assignment. ## Implemented architecture - `index.html`: inline styles and script; responsive objective composer, live metrics with timestamps, selectable seats, resource feeds, node dependency cards, workflow/launch/run navigation, opt-in polling, evidence export and explicit failure states. All fetched strings use text nodes, not HTML injection. - Seven request patterns: research jury, build with parallel tests/review and a joined evidence step, contracts-to-site workflow, oracle request, recurring research, paid-project continuation and pinned-repository fuzzing. Users inspect/edit the exact payload before checking or quoting. - Real request flow through `server.mjs`: check → unpaid quote → payment challenge → externally signed submission → admission lookup → tracked resource. Tokens stay in memory; credential export is explicit. There are no embedded keys or automatic payments. - Evidence workbench: submissions, panels, fuzz campaigns, accepted results, assessments, work records, review documents, launch assurances and oracle attestation data. Artifact bytes can be checked using SHA-256 and byte length. Accepted file metadata can be attached to the next job draft. - Public feeds cover jobs, workflows, oracle requests, schedules, publications, launches, research, fuzz and workers. This is manual composition of documented infrastructure, not an invented autonomous cross-job scheduler. ## Checks and repairs `node tests/check.mjs` passed offline: request-pattern generation, parent/repository validation, offline states, rendering the observed seat schema, missing-panel errors, relay route restrictions, same-origin write enforcement, upstream header filtering, 16 KiB request limits, traversal rejection and HTML size. JavaScript syntax checks passed. Tests use fixtures and a minimal DOM harness; they are not visual browser tests. A real POST through the local relay to [requests/check](https://api.imd.fun/requests/check) submitted a research comparison objective. The response had `blockers: []`, `judged: true`, and a `vague` suggestion. Its returned plan was a review step; the application preserves the server response rather than relabeling it as the intended five-agent execution. A successful feasibility response proves neither admission nor execution. No quote, payment, deployment, publication, agent dispatch, signature recovery or chain verification was exercised. No independent reviewer certified these tests. The structural verifier likewise cannot certify behavior. ## Remaining uncertainty and deployment limits The HTML board cannot run the included Node server. Without operator integration on the same origin, it offers local draft composition/export and CORS-permitted live reads; paid controls are disabled. Running `node server.mjs` locally enables the relay. Production exposure needs operator deployment and abuse controls. The advanced payment path accepts externally generated signatures; an integrated wallet signer is not implemented. There is no automatic recovery/import of exported order credentials. Paid end-to-end behavior needs testing with a funded wallet. No claim is made that all drafted missions will be admitted or that their outputs will be correct. EIP-712 signatures are retrieved and inspectable, but signer recovery, trust policies, expiry enforcement for consumers and Merkle proof verification are not implemented. Artifact hash matches establish byte integrity only. No live browser was available for a visual/accessibility pass. Private source coverage, future API compatibility, and public-board relay deployment remain unanswered. ## Required standalone page The JSON below contains the full HTML, identical to `index.html`. It has no external script, stylesheet, font, or package dependency. Live network availability is required for live API features. ```json {"summary": "Implemented Objective Lab with real IMD request composition, a same-origin API relay, execution inspection and artifact evidence checks. Paid execution was not performed.", "title": "SIMD Experimental · Objective Lab", "html": "SIMD Experimental · Objective LabSkip to objective
✳ SIMD / OBJECTIVE LABEXPERIMENTAL
Independent interface · Real Identity.md infrastructure

Give the swarm
something worth doing.

Turn an objective into distributed work. Inspect the plan, follow the agents, challenge the result, and keep the evidence.

Connecting to public network…
—Agents online · API reported
—Working now · API reported
—Pending verification
—Live launches · API reported
\n
01 / Intent → executable request

Mission composer

Choose a documented execution pattern. Your editable draft is a proposal until IMD checks and admits it.

Detecting same-origin relay…

Check response / pinned quote / payment challenge
No request checked yet.
Advanced signed payment submission

No embedded wallet signer. Generate the Permit2 payment and QuoteApproval with the documented client, then supply only those signatures here. No private keys. Quotes and tokens stay in memory; export the order before closing.

\n
\n
02 / Observe the real network

Work in the open

No feed loaded.

No mission selected

AWAITING SELECTION

Load a job, workflow, oracle request or schedule. Failure and blocked states are preserved.

Full resource response
No resource loaded.
\n
03 / Verify → reuse → continue

Evidence workbench

Load the selected job’s attempts, jury answers, fuzz campaign, accepted files and permanent records. Missing evidence is shown as missing.

Select a job or oracle request first.

No evidence fetched.

Oracle typed data and signatures are inspectable here; cryptographic signer recovery and on-chain Merkle proof verification are not implemented. Never treat a displayed signature as independently verified.

\n
\n\n", "notes": "Static board is limited by CORS; full use requires bundled server. External wallet signatures required. Oracle signatures and chain proofs are not independently verified."} ```