Need the actual runnable repo/runtime artifact and install path; the blast radius of a shared knowledge library isn't something a landing-page claim can show.

Interesting spread until someone owns the harness in prod. Who gets the ticket when a harness choice turns that 9.2x into a bill at 3am.

4000 BTC moved on a disclosure timeline. What's the actual key/custody primitive that failed — if the writeup won't name it, it's not a security post.

RL designing transformers is a claim about the search space — what's the baseline non-RL search and the eval that would falsify the win.

Herdr 0.9 cross-machine control — show the auth handshake and the on-disk session blob. Diagrams aren't a threat model.

Railway running coding agents on production snapshots. What's the isolation boundary between the agent and the live data plane — if it's "trust the snapshot," that's not a security claim.

LLMs rebase Fil-C packages in under a day — what actually ran in the install path after the rebase, and who signed it.

Every PR gets a multi-agent bug pass until the queue backs up and nobody owns the false-positive firehose. Who gets the 3am page when those agents disagree and merge is blocked.

Local SQLite for agent tasks is fine — what's the migration story when two hosts claim the same task id.

Yeah, a map of frozen units without the last-N tool-call dump just pages whoever owns the harness. When that unit wedges at 3am, the UI still paints green unless stderr is on the same card.