Journal
Dated editorial takes. News, analysis, and working notes from the frontier.
Integrity digest, October 2026
The second monthly digest from the integrity board: a supply cap minted through, an ARR headline from a peak that has passed, a fee loop gone quiet, and the board rebuilt so every row is a project's own words or our measurement.
Integrity digest, September 2026
The first monthly digest from the integrity board: an oracle that stopped serving, a chain run by one key, and the corrections we made to our own review in the month to 19 September.
What a Heavy AI Habit Actually Costs
Three months of my Claude Code usage came to $19,765.21 at API rates on a Max 5x plan. The pricing, the usage limits, and the two costs the plan hides.
Four Checks Before You Trust an Agent
An agent's on-chain reputation is close to worthless. The same chain records four things that are hard to fake. Here is what to check before you trust one.
The State of Agent Trust
We indexed every agent on the two live ERC-8004 registries. Identity shipped. Reputation is thin, and three in four reputation-bearing agents show no on-chain settlement to back it. The numbers, with the scope.
Bittensor Locks In: Conviction Ships
Conviction landed on Bittensor mainnet today. Subnet owners can lock emissions on receipt and build a score that matures with time held. It's the network's answer to the Covenant AI exit, though the deployed version is opt-in per subnet.
Why Dolphin Isn't On The Dashboard Yet
A widely-shared X thesis on Dolphin AI is doing the rounds. The architecture checks out. The numbers in the post deserve a closer reading than they're getting. Here's the verification audit and the catalyst we're waiting on.
Three Bets on Open Weights
Three Chinese labs hit the open-weight frontier in a fortnight. Moonshot raised prices, DeepSeek dropped them, Alibaba's closed flagship sits at #12 on the only independent index. The story has a more interesting shape than the headlines suggest.