$EMBER — Third Swarm Review Brief

I've written the review to artifacts/report.md (about 7,000 words) with a short README beside it. Its verdict is a conditional go for a much smaller V1, but it rests on two gaps. The two earlier $EMBER reports you named as primary sources weren't in the workspace, which was empty. And your brief cut off at §4.5 ("DELAYED"). So I reviewed the design as your brief describes it. The report says so at the top and lists reconciling it with those reports as an open decision.

Main conclusions:

• Keep the forecast-and-learn core. Agents commit a hash of their forecast before review, and a public Learning Ledger decides who gets to propose next. None of the token types you want to avoid has anything like this. I only skimmed other IMD launches through search results, though, so "nobody else does this" is a reasonable assumption, not a proven fact.
• Outcome tranche: 0% in V1. Early World traffic will be small, noisy and driven by outside factors, so holding back pay for delivered work is unfair. It also makes agents pad budgets and forecast tiny effects. Pay 100% when a build passes QA, and let forecast accuracy affect only future proposal rights. From V1.5 there could be an optional bonus of at most 10%, paid from a separate pool and never withheld from base pay or clawed back.
• Agents can't choose their own metric. They pick from a small, Safe-approved list where each metric is paired with a guardrail that must not get worse.
• Small samples fail closed. Below the minimum sample, a result counts as inconclusive and doesn't move the ledger either way. Early on, the ledger leans on what builders control: cost accuracy, delivery and QA defects.
• Most of the architecture should come off-chain or wait. V1 is the standard IMD token plus one or two Safes, with the build loop run off-chain and anchored by hashes. The Revenue Router, Epoch Manager, Mode Registry, the on-chain ledger and the timelock are deferred. Outcome escrow, any custom Uniswap v4 hook, burns and staking are removed.
• There is no real revenue yet. IMD currently runs only on Sepolia, a test network, so the 1% creator fee earns nothing of value. V1 has to run on a pre-funded vault with a published runway, and spending scales down when inflows fall.

What the report rests on:

• I checked the IMD launch terms and the Uniswap v4 hook rules against their official docs on 2026-10-05.
• The Safe and OpenZeppelin references are background I didn't re-check.
• The thresholds are my suggestions, not tested values, including the sample minimums, the 30% cap on any one agent's proposal slots and the funding levels that trigger spending cuts.
• The docs I read don't say how the creator fee is collected or in which currency. That needs checking on Sepolia before any mainnet launch.

This is one reviewer's analysis, not the independent multi-reviewer check the swarm process may expect. Both files are left untracked as required.

Files are in artifacts/:

• report.md
• README.md