# Hard grade: "Two complementary experiments" thesis on Identity.md (IMD) and SIMD

- Thesis: tweet by @chukwue46258952 ("Noble"), https://x.com/chukwue46258952/status/2107626665274679321. The status ID decodes to 2026-10-07 00:18:49 UTC.
- Graded: 2026-10-07. All live figures below were read on that date.
- Author reach is out of scope and is not assessed here.

## Verdict

**Grade: C− (about 52/100).**

This is a well-framed essay. It states a falsifiable question, gives numeric kill-criteria, and sets "near-zero conversion" as its default expectation. That is better epistemic hygiene than most posts about this ecosystem. Its description of the IMD supply side is mostly accurate and checkable.

Its description of SIMD, the half the thesis is really about, is not supported by any primary source I could reach, and parts of it conflict with the sources that exist:

- The thesis tags @SuperIMD_eth, but the SIMD token site links to a different X account. Search-indexed summaries of @SuperIMD_eth say it covers **50%** of job costs on Ethereum. The SIMD site says **100%**, on Robinhood Chain.
- "Reimburses the full job price" does not match the only aggregate figure I found: 121.5 IMD over 386 refunded tasks is about 0.315 IMD per task, not 0.5.
- "Pays only for the initial exposure" (first job only) appears in no source. The SIMD site says fees cover "100% of Identity.md job costs", with no first-job limit.
- No vault address is published, so "fees → vault → reimbursement" and "multiple batches settled" cannot be checked.

Of the four thresholds, one is close to meaningless because it is already met by baseline activity. The failure rule cannot be evaluated on the timescale it sets.

### Scorecard

| Dimension | Score | Why |
|---|---|---|
| Factual accuracy, IMD side | 8/10 | Price, seats, explorer and acceptance rates check out. "Fixed" price is slightly overstated: the docs say "Current price". |
| Factual accuracy, SIMD side | 3/10 | Account and chain are mixed up, 50% vs 100% coverage is unresolved, "full price" and "first job only" are unsupported, and the vault is unverifiable. |
| Evidence and attribution | 2/10 | No links, no addresses, no numbers for the claims about the vault or batches. |
| Falsifiability and design | 6/10 | Good framing and a numeric kill-switch, but the thresholds are measured by wallet, have no control group, can be gamed with sybil wallets, and T4 is trivially met. |
| Internal consistency | 5/10 | "Removes first-trial cost" conflicts with a refund model in which the user pays first. The month-long failure window conflicts with the 45- and 90-day metrics. |
| Calibration and honesty | 8/10 | Defaults to the skeptical outcome and claims no results. |

## 1. Claim-by-claim check

Legend: **FACT-OK** = confirmed by a primary source. **PARTIAL** = true with caveats. **UNVERIFIED** = no primary source found. **CONTRADICTED** = conflicts with a source.

| # | Thesis claim | Status | Evidence |
|---|---|---|---|
| 1 | Capped set of agent seats linked to NFTs; holders run the agents | FACT-OK | Bankless: "Each of the 2,000 identity.md NFTs offers one seat in the IMD swarm"; about 380 of 2,000 were enrolled as of about 2026-09-25. Bankless also says "accepted efforts get logged onchain as reputation." ([Bankless](https://www.bankless.com/read/inside-imd-ethereum-s-new-ai-swarm-experiment)). Explorer agents page today: "All 750 Online 707 Working 9" ([explorer.imd.fun/agents](https://explorer.imd.fun/agents)). |
| 2 | Jobs submitted, executed, judged, recorded on a public explorer | FACT-OK | [explorer.imd.fun](https://explorer.imd.fun/) lists jobs with states. The API exposes per-attempt `accepted`/`verdict` (`GET /jobs/:id/submissions`, [docs](https://imd.fun/docs/)). One caveat: the explorer is an off-chain web app. The docs describe agents as "scored onchain", but job records themselves are served from `api.imd.fun`, so "recorded on-chain" is only partly established. |
| 3 | Anyone can pay a fixed price, currently 0.5 IMD | PARTIAL | Docs: "Current price: 0.5 IMD each". "Current" means the price is adjustable, not fixed. Payment is by wallet signature (x402 / EIP-712) ([docs](https://imd.fun/docs/)). |
| 4 | High acceptance rates on large samples | FACT-OK (with scope) | Bankless: "About 86% of the swarm's ~50,700 attempts have been accepted, and only ~1% were outright rejected." Top agents on the explorer show 96–99% rates. These are **attempt-level acceptances by the swarm's own judges**, not satisfaction measured from paying users. "Reliable source of work" is therefore an inference from self-judged output. |
| 5 | SIMD does not run agents; it observes the network and funds first contact | PARTIAL | The SIMD site: "A protocol tracking IMD agents in real time, using protocol fees to cover 100% of Identity.md job costs" ([superimdc.xyz](https://www.superimdc.xyz/)). Nothing there says "first contact" or "first job". |
| 6 | Trading fees → public vault → reimburses full job price | UNVERIFIED / CONTRADICTED | The SIMD site states the intent ("Protocol fees are collected for one use: covering Identity.md job costs in full") but publishes **no vault address**. Search-index summaries of @SuperIMD_eth describe "50% of Identity.md job costs on Ethereum". The SIMD site says 100% on Robinhood Chain (chain ID 4663, token `0x577bc74a99aabbd43a354668dc7b7ed4b5e303ae`). The SIMD site's X link is **x.com/superimdx**, not @SuperIMD_eth. |
| 7 | Multiple batches settled; vault repeatedly holding substantial balances | UNVERIFIED | The only figures found are search-engine summaries of X posts: "121.5 $IMD has been paid out across 386 refunded tasks" and "vault currently holds 1,341 $IMD". I could not open the posts themselves (X needs a login). 121.5 / 386 = **0.315 IMD per task**, which is neither the full 0.5 IMD price nor 50% of it (0.25). Either the price changed, refunds were partial or mixed, or the figures are wrong. In every case "reimbursed at full price" is not shown. 1,341 IMD is about 2,682 jobs at 0.5 IMD, which is reasonably "substantial" for this network, if it is real. |
| 8 | "Most subsidies keep paying indefinitely. This one pays only for the initial exposure." | CONTRADICTED by available wording | The SIMD site promises to cover "100% of Identity.md job costs", with no limit per wallet or per first job. No eligibility rule was found anywhere. |
| 9 | "The mechanism is live and observable" | PARTIAL | The IMD side is observable. On the SIMD side, the token and the site exist, but the vault and its payouts are not linked anywhere a reader can audit. The DEX and explorer pages for SIMD (dexscreener, robinhoodchain.blockscout.com) returned Cloudflare challenges from this environment, so I did not check SIMD trading volume or fees. |

## 2. Independent measurement from the IMD API

I paged `GET https://api.imd.fun/jobs?exclude=oracle` and read `paidBy` from `GET /jobs/:id` for every job (method in the [docs](https://imd.fun/docs/)). This is my own crawl, run on 2026-10-07.

- **1,618 non-oracle jobs** from 2026-08-23 to 2026-10-07 (completed 1,545, blocked 43, cancelled 25, executing 5). The detail call failed for 18 of them, so those 18 are left out of the payer figures.
- **1,393 jobs have a `paidBy` wallet. 207 have `paidBy = null`** (unpaid or internal). The first paid job is dated 2026-09-23.
- **180 distinct paying wallets.** 97 of them paid exactly once. 83 paid two or more times, and 37 paid five or more times. 50 wallets paid on two or more distinct days.
- **Concentration:** the top 3 wallets paid for 660 of the 1,393 paid jobs (47%).
- **Top payer `0x9fadab91f6fa03dbd7f4f8a08a338704baacf63f`** paid for 289 jobs, all since 2026-10-04. It paid for **all 254 jobs whose objective carries a `SIMD` tag**: `[SIMD-COLLISION:*]`, `[SIMD-CONTEST:*]`, `[SIMD-THESIS]` (including this grading job), and "oracle mission for SIMD Discovery Arena". On Etherscan the wallet was funded about two days ago and holds no IMD on Ethereum mainnet ([etherscan](https://etherscan.io/address/0x9fadab91f6fa03dbd7f4f8a08a338704baacf63f)). I could not establish who controls it.

What this means for the thesis:

- **Inference:** most visible "SIMD" activity on IMD is a single wallet submitting SIMD-branded campaigns: hash-collision contests, contests, and thesis grading. That looks like a sponsor funding work centrally, not a vault rebating many first-time users. It does not disprove a separate rebate flow, because a rebate would show up as an IMD transfer from the vault to the user, not in `paidBy`. It does mean the thesis's picture ("first contact" for new users) is not what dominates the observable data.
- **Fact:** `paidBy` cannot tell a reimbursed payment from a self-paid one. Every job in the API looks self-paid. Thresholds T1, T2 and T4 can only be measured if the vault's payout transactions are published and joined to `paidBy`. That data is not public today.
- **Fact:** **180 paying wallets already exist, and 50 return on multiple days.** T4 ("≥40 never-reimbursed wallets appear as paying users within 90 days") is very likely met by baseline activity whatever SIMD does. As a test of SIMD it carries almost no information.

## 3. Problems with the experiment design

1. **No counterfactual.** Conversion among reimbursed wallets is never compared with conversion among non-reimbursed first-time payers over the same period. Without that, a 30% return rate cannot be credited to the subsidy. In my crawl 83 of 180 payers (46%) already repeat, with no subsidy separated out.
2. **A rebate is not the removal of cost.** If reimbursement follows payment (the source wording is "refunded tasks"), the user must already hold IMD, bridge or buy it, and sign a payment. The cold-start friction the thesis targets (acquiring the token, trusting the flow) is still there. Only the net cost is refunded afterwards. The thesis's claim "SIMD removes that first-trial cost" overstates this.
3. **Wallet-level metrics are sybil-prone.** If eligibility is "first job per wallet", one user with N wallets collects N rebates, and later "self-paid" jobs from a main wallet look like conversions. No sybil control is described.
4. **The failure rule cannot be evaluated.** "Any two thresholds fail for a full month" cannot be applied to T1 (a 45-day window) or T4 (a 90-day window) after one month. The rule needs a fixed evaluation date.
5. **T3 is underspecified.** "Without continuous draw-down of reserves" has no number. It is also unclear whether fees are counted in SIMD/ETH on Robinhood Chain or in IMD, and no conversion or bridging path is described. The price of SIMD drives the coverage ratio, so this measures the token's trading activity more than demand for IMD work.
6. **Quality evidence is self-referential.** The acceptance rates come from the swarm's own reviewers and judges. The thesis's mechanism ("consistent, high-quality output" leading to repeat payment) needs evidence from the payer side, such as repeat-payer rates or output delivered and used, not attempt acceptance.
7. **Survivorship in T2.** The median days-to-convert is computed only among converters. With a 45-day window it is bounded by construction, so T2 adds little beyond T1.

## 4. Separating facts, inferences, uncertainty and open questions

**Facts (primary-sourced or measured):**
- IMD has 2,000 NFT seats. 750 agents are registered and 707 are online today.
- Job price per the docs: 0.5 IMD ("Current price").
- Bankless reports about 86% acceptance over about 50,700 attempts (as of about 2026-09-25).
- SIMD is an ERC-20 on Robinhood Chain (4663) at `0x577b…03ae`. Its site claims fees cover 100% of IMD job costs and links to x.com/superimdx.
- My crawl: 180 paying wallets, 1,393 paid jobs, top-3 wallet share 47%, and all 254 SIMD-tagged jobs paid by one wallet since 2026-10-04.

**Inferences (mine, labelled):**
- "SIMD" as seen on the explorer is mostly a centrally funded campaign, not a first-contact rebate.
- The thesis merges at least two branded efforts: @SuperIMD_eth (said to be 50%, Ethereum) and SIMD/@superimdx (100%, Robinhood Chain). They may share an operator, but I found no evidence either way.
- T4 will very likely pass whatever the subsidy does.

**Uncertainty:**
- The 121.5 IMD / 386 tasks and 1,341 IMD vault figures come from search-engine summaries of X posts that I could not open. Treat them as unverified.
- The controller of `0x9fad…f63f` is unknown.
- Whether any first-job-only rule exists is unknown.

**Unanswered questions (what would upgrade this thesis):**
1. What is the vault address, and on which chain? Where is the list of payout transactions?
2. What are the eligibility rules: first job only, a cap per wallet, sybil controls?
3. Is coverage 50% or 100%, and which account (@SuperIMD_eth or @superimdx) is authoritative?
4. How do SIMD fees on Robinhood Chain become IMD on Ethereum, and who executes that step?
5. Will someone publish a dataset joining payouts to `paidBy`, so that T1, T2 and T4 can be computed by anyone?

## 5. Bottom line

As framing, the thesis is good. As a description of SIMD, it is speculative and partly at odds with the sources. It presents a design ("pays only for initial exposure", "full price", "visible capital path") as if it were fact, when no public source states the design and no public data shows it. Its one strong move is to make skepticism the default, and the evidence available today supports that default. The visible SIMD activity on IMD is concentrated, and centrally paid work does not test conversion.

## Sources

- Bankless, "Inside IMD, Ethereum's New AI Swarm Experiment": https://www.bankless.com/read/inside-imd-ethereum-s-new-ai-swarm-experiment
- IMD docs (pricing, API): https://imd.fun/docs/
- IMD explorer (jobs, agents): https://explorer.imd.fun/ and https://explorer.imd.fun/agents
- IMD API, crawled 2026-10-07: https://api.imd.fun/jobs
- SIMD site: https://www.superimdc.xyz/
- @SuperIMD_eth profile (content not readable without login): https://x.com/superimd_eth
- Top payer wallet: https://etherscan.io/address/0x9fadab91f6fa03dbd7f4f8a08a338704baacf63f