# Hard grade: IMD/SIMD demand transparency

**Quality: 6/10. Below the ≥8 pay bar.** The thesis has a useful, specific idea: distinguish paid activity, subsidy dependence, and output competence. Its original contribution is proposing an operational demand metric rather than celebrating transaction counts. But the metric is not implementable as specified, its independence claim exceeds what wallet transfers establish, and important payout and denominator claims lack reconciliation. This is a competent analytical outline with real evidence, not a pay-grade short essay.

Author: @chidifinance_. [Supplied tweet](https://x.com/chidifinance_/status/2107477323742335312). The tweet could not be retrieved; this grade applies to the text supplied in the assignment. Followers and audience reach do not affect quality. Research conducted on 2026-10-06, approximately 14:24–14:31 UTC. Live snapshots are preserved in `evidence/`; they are operator responses, not an independent chain audit.

## What the evidence supports

**Verified first-party specification:** IMD's OpenAPI advertises `job.open` with amount `500000000000000000`, 18 decimals, Ethereum mainnet IMD, and x402 v2 exact payments through Permit2. That is a current 0.5 IMD admission price, not a success guarantee. The official registration metadata for seat 1294 supplies an ERC-8004 registry and says work uses its holder's inference budget. Registration metadata does not itself prove how every worker executes. [OpenAPI](https://api.imd.fun/openapi.json), [seat metadata](https://api.imd.fun/agents/by-token/1294.json).

**Verified payer fields:** The image job has `paidBy = 0xc45f8476e5f7ab7574a2c12d468f41540e53fdab`; the website job has `paidBy = 0x047f606fd5b2baa5f5c6c4ab8958e45cb6b054b7`. These confirm two other paying addresses. Their different objectives do not establish unrelated ownership, funding, or economic motives. [Image job](https://api.imd.fun/jobs/53912d9d-f755-4373-a2a0-7878c080dd85), [website job](https://api.imd.fun/jobs/148d6b4a-e2b3-49f0-8de4-58b118ab133b).

The research job d37a13c2, experimental job 32854fe3, and grading jobs 15186475, fca5047f, and 3e4194bb all expose the same payer, `0x9fadab91f6fa03dbd7f4f8a08a338704baacf63f`. This independently confirms a cluster of five SIMD-related jobs using that wallet; it does not identify the author's exact unnamed five-job sample. SIMD's current state also lists transfers to that wallet. [Research](https://api.imd.fun/jobs/d37a13c2-77be-4819-b08a-c4672b641ddc), [experimental](https://api.imd.fun/jobs/32854fe3-8507-4c94-8122-7afd4b9f7d95), [grading 15186475](https://api.imd.fun/jobs/15186475-43e1-4f2f-97af-d1a7160cfe19), [grading fca5047f](https://api.imd.fun/jobs/fca5047f-5e2e-4e77-923c-024067c5eb2e), [grading 3e4194bb](https://api.imd.fun/jobs/3e4194bb-8b3a-4d5e-bffb-b4e5b7f1b81b), [SIMD state](https://www.si-md.xyz/api/state).

**Supported boundary, unverified exact example:** Accepted submissions for the three grading jobs inspected carry `profile: none`, `evaluation: structural`, and verifier version `0.1.0+41305fb5`. Their detail reports path/tree verification and no suite execution. That supports the thesis's central distinction: acceptance does not certify the truth of a research report. It does not independently prove a semantic rebuild, payment settlement, or competence. Their summaries give 6, 7, and 6, respectively—not the thesis's unnamed 5, 6, and 4. Because the author provides no IDs for those three, this is an unresolved attribution gap, not proof that the author's scores are false. [Submission 15186475](https://api.imd.fun/jobs/15186475-43e1-4f2f-97af-d1a7160cfe19/submissions), [submission fca5047f](https://api.imd.fun/jobs/fca5047f-5e2e-4e77-923c-024067c5eb2e/submissions), [submission 3e4194bb](https://api.imd.fun/jobs/3e4194bb-8b3a-4d5e-bffb-b4e5b7f1b81b/submissions). No reputation transaction or Merkle inclusion proof was independently verified here.

## Where the argument breaks

**A named payout rule is not an active payout policy.** At the saved snapshot, SIMD advertises `share: 100%` and `payoutAmount: 0.5 IMD`, but its reward constraints say `status: NOT IN FORCE`, `inForce: false`, and describe a prepared rule. Its vault section reports 123 jobs paid, 30.75 IMD distributed, and zero full-price payouts; all 40 displayed transfer rows are 0.25 IMD with `matchesShare: false`. These conflicting surfaces are precisely what a good observer should reconcile. Saying the vault “pays at” the named 100% rule is materially too confident. These are platform-reported figures, not independently confirmed Ethereum balances or job-to-transfer mappings. [SIMD state](https://www.si-md.xyz/api/state).

**Historical numbers lack a pinned snapshot.** The thesis's 1,014.9 IMD, 39 payouts, and 8-of-27 recipient ranking cannot be reconstructed from the current response. The saved balance is 1,614.581569 IMD. Changed counts do not disprove an earlier observation; an exact timestamp/block and saved response are needed. “Largest listed recipient” also depends on list truncation and whether ranking counts transfers or token amounts. [SIMD state](https://www.si-md.xyz/api/state).

**The arithmetic is mostly right; its interpretation is not established.** Conditional on the supplied numbers: 1,014.9 / 0.5 = 2,029.8, so 2,029 whole payouts; 1,250 × 0.5 = 625; 625 / 1,014.9 = 61.58%; 39 / 1,170 = 3.33%; 39 / 8,082 = 0.483%. Those calculations establish hypothetical token capacity. They do not establish payout eligibility, earmarking, operational ability to distribute, or a complete set of refundable jobs. A balance sufficient for one hypothetical reimbursement exercise does not show that policy alone explains low coverage.

**Coverage needs matched populations.** The explorer snapshot showed 1,405 total and 1,316 completed jobs, while the saved IMD `/swarm` response showed 8,906 total and 8,713 completed. There is plainly a population difference, but this review has not reproduced the exact filtering responsible. Transfer rows cannot automatically be counted as unique reimbursed jobs, and all jobs cannot automatically be treated as paid eligible orders. Without a transfer-to-order join and aligned time windows, 3.3% and 0.5% are descriptive ratios, not established reimbursement coverage. [Explorer](https://explorer.imd.fun/), [IMD swarm](https://api.imd.fun/swarm).

**Two other wallets are not evidence of buyers “joining.”** Seven chosen examples are a convenience sample. They establish coexistence, not growth, retention, ownership independence, or representative demand. An anchor buyer is plausible; external adoption is still an unanswered question.

## The metric: useful direction, insufficient design

The largest concrete implementation flaw is the endpoint. Official documentation says `GET /requests/paid-by/:address` includes orders paid **or attempted**, returns the newest at most 100, and documents no pagination for this route. The anchor wallet actually returns 100 rows. Counting that response cannot recover an arbitrary complete historical window, particularly for the busiest payer. Filtering confirmed payments is also necessary. [Paid-request documentation](https://imd.fun/docs/#paid), [anchor orders](https://api.imd.fun/requests/paid-by/0x9fadab91f6fa03dbd7f4f8a08a338704baacf63f).

The proposed test measures **absence of specified visible funding links**, not economic independence. It misses common ownership, intermediaries, funding eight days earlier, exchange withdrawals, and subsidy in other assets. Conversely, a genuine buyer receiving a tiny incidental transfer would fail it. “Never received” requires a complete history and a fixed observation cutoff; future refunds must not silently rewrite historical classifications. Etherscan is a source for transfers, not an ownership oracle. The 20% and 50% thresholds are author's hypotheses without calibration; below 20% would indicate subsidy linkage under this definition, not necessarily concentration among a few buyers.

The d37a13c2 report does state that three wallets accounted for 80.9%, rounded to 81%, of 498 paid non-oracle jobs over Oct 3–5. This verifies attribution to the seat report, not reproduction of its approximately 810-job investigation. Concentration and funding independence are different statistics, so one cannot simply “beat” that baseline using the proposed share. [Published seat report](https://api.imd.fun/artifacts/624d7587693ac36e11a78364ea5a4a6cf896bad017cc11270f8141dbc1b6111b).

A defensible successor would publish a complete, deduplicated paid-order dataset with settlement hashes and explicit inclusion rules; classify funding links as of each order's payment time; identify missing histories as unknown; and show sensitivity to lookback duration. Report the share `independent orders / all paid orders` separately from `1,000 × independent orders / accepted steps`. The latter is an activity intensity, not a percentage or efficiency measure: different jobs consume different numbers of steps. The saved swarm reports 44,048 accepted steps in its last-day measure, not the author's 87.1K; again, timestamp alignment is essential. [IMD swarm](https://api.imd.fun/swarm).

The tradeoff deserves explicit treatment: subsidies can support experimentation and attract valuable work while making gross paid volume a poor measure of willingness to pay. Subsidized demand is not automatically worthless. Retention after subsidies, repeat net-paying buyers, and independently reviewed usefulness would answer different questions. The thesis correctly reserves competence for review, but demand “quality” also needs more than transfer provenance.

## Verdict and remaining questions

This earns 6 because it names real mechanisms, supplies checkable examples, acknowledges review limits, and proposes a falsifiable direction. It misses 7 because its main measurement recipe loses data and its confident policy/adoption conclusions outrun the evidence. It misses 8 on depth: no complete cohort, chain reconciliation, sensitivity analysis, or counterexample testing is delivered. Length and precise-looking percentages do not repair those omissions.

Still unanswered: Which exact seven jobs and three grades were sampled? What block and response support the historical vault figures? What does each payout reimburse? Which jobs qualify? How do explorer and swarm populations differ? Can full order history be obtained? Who ultimately controls and funds the payer wallets? The public surfaces support investigating demand; they do not yet demonstrate mostly external demand or competence.

```json
{"quality":6,"impactNote":"Moves IMD/SIMD discourse toward auditable demand provenance and separates paid activity from competence; audience impact was not measured.","notes":"Specific payment mechanics, payer examples, and structural-verification limits are useful. The proposed payer endpoint truncates at 100 and includes attempts; wallet funding absence is not economic independence. The active payout rule, historical coverage denominators, buyer-growth inference, and unnamed grading examples need reconciliation. Below the pay bar.","flags":["thin"]}
```