The output is prose you have to trust.

Five questions a reader asks. Five dead ends.

A run produces a report; the searches, pages, browser steps and model calls behind it are not kept. Each question a careful reader asks of the report stops at the same wall: there is no record to answer it from.web-research/DEEP_RESEARCH.md · docs/RESEARCH-POC.md §12

The five, as failure modes

• not recorded“Which page said this, in those words?” The report carries fluent prose and a list of URLs; the page that was actually read, and the bytes the sentence rests on, are gone with the run.
• a different run every time“Run it again, exactly.” Nothing was recorded to re-execute from, so a second run is a second opinion, not a check of the first.
• half of real“What did it cost?” The number printed is an estimate at batch rates, about half the real spend.
• nowhere to put it“This claim is wrong.” There is no place in the run for the reader's disagreement to be recorded against the claim it judges.
• answered anyway“Did anything fail?” Failures are answered over, confidently.

Our own earlier tool was no different.

The web-research tool this project replaces — a browser-use clone driving Gemini through OpenRouter — ran the same eight questions the kernel is now measured on. It answered confidently when every browser agent had failed, and printed costs at half the real spend. Its eight runs are the baseline in the benchmark set.

16–122 calls · 157–739 s · $0.11–$0.94 estimated per question · 2 of 8 runs with no report. The dollar column is the tool's own estimate at batch rates; the scout run it priced at $0.107 cost about $0.21 in fact. The two error rows ended without a report and are still on the benchmark: the kernel's Q7 and Q8 are judged against nothing.bench/sets/gold-8/questions.json (baseline calls, seconds, estimated_usd, status) · bench/sets/gold-8/caps.toml (the batch-rate note) · docs/E1-2026-09-16.md Q7 (OpenRouter 402)

next: how it works — the whole system as composed, append-only, content-addressed ledgers.