# Hard grade: IMD + SIMD conversion thesis

**Verdict: 44/100 (4.4/10). A useful hypothesis and a promising measurement agenda, but an inadequately evidenced public thesis. The economic flywheel is not demonstrated.**

Assignment: `[SIMD-THESIS]:mux92j4g-wour9`. Research snapshot: 6 October 2026 UTC. This grade assesses the supplied text, not the token's investment merits. The submission ends mid-sentence at “showing subsidized users converting a”; no credit is awarded for an imagined continuation.

The strongest idea is the separation between accepted execution, subsidized activity, and independent repeat demand. The weakest move is calling aggregate balances, payout counts, and agent acceptance rates “cohort observations.” They contain neither a defined user cohort nor its subsequent behavior. A proposed causal pathway is not causal evidence. Adding four numbers makes the claim more testable, but does not establish that those thresholds are meaningful or have been met.

## Grade breakdown

Scores use an explicit reviewer rubric, not a claimed industry standard. Missing evidence loses points; uncertainty is not treated as proof of fraud or failure.

| Criterion | Score | Reason |
|---|---:|---|
| Mechanism specificity | 10/15 | Names a vault and reimbursement path; omits fee collection, conversion, custody, eligibility and reconciliation. |
| Technical honesty and attributable evidence | 5/20 | Conditional central claim is honest; categorical “observable on-chain” language outruns the supplied evidence. |
| Cohort measurement | 4/15 | Early activity counts are not cohort outcomes; no unique users, denominator, dates or maturation window. |
| Downside analysis | 8/10 | Clear zero-conversion failure case; incomplete costs and adverse incentives. |
| Operational thresholds | 8/15 | Numerical targets and windows help; definitions, rationale and reserve floor are missing. |
| Causal reasoning | 4/15 | Plausible learning mechanism; no counterfactual or separation from selection and rewards. |
| Depth and originality | 5/10 | Three-layer framing is useful; bootstrap-to-paid conversion is familiar, and the evidence section is truncated. |
| **Total** | **44/100** | **Credit for the question and framework, not for unproven success.** |

## Evidence ledger: what survives checking

**Verified primary-source response, limited to what it says.** IMD documents a current 0.5 IMD price for a paid request. A direct GET of its [live capabilities API](https://api.imd.fun/requests/capabilities) returned `job.open` payment amount `500000000000000000`, decimals `18`, Ethereum chain `eip155:1`, IMD asset `0xd34a99bc0f67ae1bbd63c660e6d0b0dd03e263b7`, and payment recipient `0x4e0fa57bde726079356537e2f34d671e9f41adbc`. That recipient differs from the named SIMD vault. This verifies the listed request price and recipient, not a reimbursement. [IMD API documentation](https://imd.fun/docs/#paid-requests).

**Verified acceptance counters, not useful-output outcomes.** [Agent #297's API](https://explorer.imd.fun/api/agents/297) returned 1,110 accepted attempts out of 1,186: 93.59%. The [seat-records API](https://api.imd.fun/seats/records) returned 723 seats totaling 878,870 accepted out of 913,791 attempts: 96.18%. Those totals include 8,379 rejected, 10,762 failed and 15,780 pending attempts. This supports the narrow statement that high acceptance counters exist on large samples. It does not identify reimbursed users, establish satisfaction, or estimate paid conversion. The public [agent page](https://explorer.imd.fun/agents/297) provides an attributable presentation of work history.

**Primary operator API, not independently verified chain state.** SIMD's [state endpoint](https://www.si-md.xyz/api/state), with vault `retrievedAt` of `2026-10-06T22:29:01.485Z`, identifies the thesis's vault on chain 1. It reports 1,656.912754 IMD balance, 36.5 IMD distributed, 144 jobs paid, 46 recipients, and zero `fullPrice` payouts. Yet its policy fields say `share: "100%"` and `payoutAmount: "0.5 IMD"`. All 40 returned payout rows are 0.25 IMD and `matchesShare: false`. A listed example is [transaction 0x3ebe…a806](https://etherscan.io/tx/0x3ebe4c8033e3a0abe4d9c670a506b03e79ae4b75dcdf010b9c949c35cba4a806). This is a transaction reference supplied by SIMD, not a receipt independently decoded here. The endpoint also labels its Ethereum RPC source “NOT CONFIGURED.” That field does not establish how the separate vault reader obtains its data.

**Inference from that discrepancy:** a stated 100% rule is not evidence of actual 100% reimbursement. A 0.25 IMD payment is 50% of a 0.5 IMD request if it is the sole reimbursement for that request. There may be top-ups, batches, policy changes or incomplete history. Without the job-to-payment join, neither historical full coverage nor its absence can be established. The thesis must reconcile these possibilities instead of asserting that coverage is already proved.

**Secondary discovery only.** A [Sotwe mirror of @SuperIMD_eth](https://www.sotwe.com/SuperIMD_eth) displayed older search-index figures of 30 jobs, 14.75 IMD distributed and 527.69 IMD balance. Opening the page displayed different posts claiming 386 refunded tasks, 121.5 IMD paid, and a 1,341 IMD balance. These are attributed operator claims through a third-party mirror, not corroborated ledger totals. They describe rewards for thesis submissions as well. Relative timestamps and changing indexed content prevent treating these figures as a dated cohort series. The discrepancies may reflect different products, windows or update timing; they cannot responsibly be collapsed into one dataset.

**Unverified:** SIMD trade fees reaching this vault; repeated historical 100% batches; every payout representing one paid job; multiple economically independent builders receiving reimbursements; any organic conversion or repeat-retention rate. No decoded chain receipt, fee contract trace, or linked user cohort was obtained in this review. The thesis supplies no such links itself.

## Mechanism: the address is a starting point, not proof

The claimed route is trading → fee collection → funding the IMD vault → refunding a customer's IMD request payment. The live request recipient and purported refund vault are distinct, so this is a two-leg payment-and-reimbursement story. The public [SIMD client](https://www.si-md.xyz/app.js?v=66) also describes a Hire flow paid by a server wallet. Direct customer reimbursement and server-paid access must be measured separately: a visitor to a sponsored interface need not be the wallet shown as the IMD payer.

To make “grounded in public on-chain behavior” defensible, publish: chain and token identities; trading pool and fee collector; collection rate and beneficiary split; transactions converting fees to IMD if necessary; vault deposits classified by origin; outbound token logs; and the job/payment/reimbursement identifiers connecting both legs. Explain whether “vault” means a controlled wallet or a contract and who can change payout rules. Those details remain unanswered here.

A transfer proves movement of an asset once its receipt is checked. It does not, by itself, prove fee provenance, a customer's job, useful work, or full reimbursement. One transaction can contain several recipient transfers; one job can receive several transfers. Therefore “every paid job is a visible outbound transfer” is an unsupported one-to-one mapping.

## The conversion test needs definitions before thresholds

None of A–D is shown to pass. They are proposed decision rules, not observations, calibrated benchmarks or protocol commitments.

| Test | Fatal ambiguity | Minimum repair |
|---|---|---|
| A: ≥25% convert within 30 days | Jobs are not users; “organic” can still be reward-funded. | Denominator: all first-time subsidized customers in a fixed enrollment window. Numerator: those buying at least one subsequent qualifying job within 30 days at full net cost, with no reimbursement, task reward or known sponsor funding. Report uncertain funding classifications separately. |
| B: ≥40% buy a second paid job within 60 days | Sixty days from first subsidy or conversion? | Anchor at conversion; count a second distinct qualifying purchase. Keep immature cohorts censored, not failed. Deduplicate continuations and scheduled runs under a published rule. |
| C: fees cover ≥80% for 14 days | Undefined safety floor, zero-demand days, top-ups and asset units. | Define fee-derived inflow net of swaps/gas, separate donations and reserve transfers, specify UTC days and minimum reimbursement volume, and fix the floor before observing results. Publish daily reconciliation and liabilities. |
| D: ≥50 outside-cohort paying wallets | Wallets are not independent customers; no deadline. | Add a window, related-wallet exclusions, funding-origin checks and a disclosed false-negative risk. Use “unique wallets” unless independence is evidenced. |

At 80% coverage the reserve still funds a 20% daily deficit. Fourteen days above that ratio establish short-term coverage, not equilibrium. A safety floor delays exhaustion; it does not eliminate a recurring deficit. Also include gas, conversion losses, operational expenses and committed unpaid reimbursements before calling the model economically sustainable.

If there were 40 eligible users—not 40 jobs—25% conversion would mean ten purchasers, and 40% repeat conversion just four repeat purchasers. Those counts are fragile. The product of the two thresholds is 10% of the original cohort under aligned definitions. The supplied job count cannot support this calculation empirically because the user denominator is unknown.

“Miss these for two consecutive months” is ambiguous: any test or all four, calendar months or entry cohorts? Sixty-day follow-up can make the newest cohorts ungradable for months. Pre-register failure rules and evaluate only mature cohorts. Publish counts and uncertainty intervals alongside rates; do not hide absent measurements behind a nominal pass/fail dashboard.

## Causality: a plausible story with live alternatives

Lower net prices may induce trials, and useful trials may teach customers that the output is worth buying. That is a hypothesis. The statement that skepticism ceases to bind after a quality signal is internalized is stronger than the evidence permits. Customer usefulness, reliability, turnaround time, security and integration effort can remain binding even after successful execution.

Alternative explanations include existing IMD enthusiasts selecting into subsidies, token incentives attracting reward seekers, builders purchasing work that benefits their own holdings, repeated wallets controlled by one party, contemporaneous product improvements, and speculative trading funding both the vault and promotional activity. High acceptance counters do not discriminate among these explanations.

For credible causal evidence, randomly assign eligible new customers to a fixed trial subsidy or a contemporaneous control, keep access and product quality comparable, and measure subsequent full-net-cost purchases after incentives end. Pre-register the primary endpoint and attribution window. Report intent-to-treat outcomes, exposure, attrition and confidence intervals. A phased rollout or matched comparison is weaker if randomization is impractical; explicitly disclose remaining selection bias. Observe actual delivered utility and repeat spending rather than acceptance alone.

The flywheel has a further missing arrow: even if subsidized trials cause paid IMD demand, why should that demand generate SIMD trading fees? Traders paying SIMD fees and customers purchasing IMD work can be different populations. Conversion could validate IMD's service without making SIMD fee revenue durable. Evidence is needed for the demand-to-SIMD-revenue link, not merely the fee-to-subsidy link.

## Downside and originality verdict

The near-zero-conversion downside is stated clearly and earns substantial credit. Extend it to positive-but-unprofitable conversion, falling speculative fee revenue, hidden reward funding, increased task prices, concentrated customers, and subsidies attracting low-value work. A refund balance cannot establish runway without a burn rate and liabilities. A nominally accepted task need not deliver externally valuable work.

The three-layer distinction is useful, but “durable value sits entirely in layer 3” overstates it: execution quality matters, and genuinely useful sponsored demand can have value even without immediate customer payment. Independent repeat purchases are the appropriate test of this particular self-sustaining customer-demand claim, not the only possible source of value.

**Final assessment:** keep the conversion question, the conditional downside and the layered demand model. Replace asserted on-chain proof with reconciled, attributable transactions; replace “cohort observations” with either actual cohorts or honestly labeled activity snapshots; and treat the thresholds as proposed experiments. Until then, this is a thoughtful research proposal dressed as a stronger evidentiary thesis than the record supports. Lack of verification is not proof that the mechanism does not operate.

## Research and verification limits

All reads were public and read-only. X returned HTTP 403. Etherscan address/token-transfer pages could not be fetched, and direct Python retrieval of the vault address and Blockscout transfer API returned HTTP 403. IMD and SIMD JSON endpoints succeeded via Python even when the web reader could not open them. No paid requests, wallet connections, signatures or protocol writes were made. The unrelated content at `identity.md` was excluded; this review uses the IMD surfaces linked above.

Saved primary API responses and the labeled SIMD extract are under [evidence/](evidence/); [manifest](evidence/manifest.json) records their source URLs. The extract is a snapshot of operator-reported fields, not an independently authenticated blockchain archive. Local checks establish readable files and reproducible arithmetic only. They do not certify source truth, receipt authenticity or the economic claim.