Web accessibility

A dated record of what Dinoer’s discovery step found and missed on a real campaign, counted in the collected pages, so that a later run can be compared with it.

2026-08-14: reputation research, SearXNG discovery against one generalist search

Context: a real subject (a person and their sole-trader business, with their consent), three query targets sent to SearXNG, campagne.py. The light collection kept 28 pages; one page was sent to the browser, without success.

The comparison: a single question on the same subject to a generalist, signed-in search assistant surfaced three pages absent from all 28: a profile on a professional network, a project page, a stock-photo credit. Checked in collecte.jsonl, not in the automatic report.

Cause: most of the 28 pages came from business directories and company data sites; the three queries did not reach the kind of page the assistant found. The ranking and the cut of the report played no part.

A second finding in the same corpus

Two of the 28 pages, a profile on a code-hosting platform and a video channel, were fetched by the light collection, but their text was only a cookie banner and a sign-in prompt. The light collection does not run JavaScript, and the escalation to the browser did not trigger: it fires on an empty text, and both pages returned some.

This gap is separate from the first: better queries would not have fixed it, since both pages were already found and fetched. Both findings, and what they changed →

What was checked

The report of the same campaign was checked on its own: the full stream of events of the delegated model, not the helper that keeps only the final text, contained exactly three events (step_start, text, step_finish) and no tool_use. Nothing left the corpus while the report was written, on the same run that showed the corpus itself had gaps.