I graded the thesis D+. The core idea (repeat, self-paid demand matters more than activity) is sound, but two of its main factual claims are false on-chain, and two of its four thresholds can't be measured under how the refund rail is built. The report is in artifacts/report.md, with a short artifacts/README.md covering the question and limits. Nothing was committed.

What the thesis gets wrong (from the vault's verified contract source, its live settings, and full transfer histories):

• "Covers the full 0.5 IMD on every job" is false. The vault refunds 50% by default. A one-off boost covered the full price for the first 100 refunds, and that ran out on 2026-10-05 06:50 UTC. Of 582 refunds, 481 were 0.25 IMD.
• "Clean, full-price batches" is false. Half-price refunds dominate, and batches range from 1 to 50 payments.
• "Fees flow into the vault" is only partly true. Fees are claimed by one developer wallet and passed on by hand. The automatic in-trade mechanism isn't switched on. Of 2,500 IMD in claimed fees, only 1,825 reached the vault. The developer can also withdraw the whole vault in one call.
• The 90%+ acceptance rate is the wrong evidence. It measures agent sub-tasks judged by IMD's own checker across the whole network. It says nothing about buyers or about retention.

What the thesis missed:

• The protocol pays itself. 287 of 588 job payments since launch (49%) came from SIMD's own server wallet, which opens its contest and thesis-grading jobs. That wallet also received 49% of all refunds. This grading job is one of those payments.
• The subsidy mostly goes to people who were already paying. Before SIMD, 151 wallets made 792 unsubsidized payments in about 9.7 days. Since launch, about half of the other payments come from those same wallets. I found no sign yet that net user-paid spend has gone up (about 37 IMD/day now versus about 41 before), but that comparison is weak given the short windows.
• Two thresholds can't work as written. Nearly every payment gets an automatic refund (296 of 301 from non-protocol wallets), so a "fully self-paid job" (#1) or a "non-subsidized wallet" (#4) barely exists. Only one such wallet has appeared since launch.
• The other rules are inconsistent. Threshold #3 contradicts itself: if fees cover 75%, the other 25% is a vault draw-down. The "fail two for a month" rule can't be applied to thresholds measured over 45 and 90 days.

Limits:

• The data is from about 55 hours after SIMD launched, so no 21-, 45- or 90-day cohort result can exist yet.
• I couldn't fetch the original tweet (HTTP 402).
• I only scanned the one treasury address the vault is configured to use.
• I didn't check whether the refunded wallets are independent of each other.
• The SIMD dashboard contradicts itself (it says refunds are "not in force" while 582 have been paid), so I treated on-chain data as authoritative.
• This is a single-agent review; nobody else has checked it.