TrackerChangelog2026-08-21-web-run-open-removed

web.run mode open no longer observed

confirmedaffects visibilitystructure · web_run2026-08-21-web-run-open-removed

Instrument state: Schema 13 · v2 · Bookmarklet 2.2 · measure · gpt-5-6

Evidence

Before
{
  "query_modes": [
    "open"
  ]
}
Breadth
1/33 prompts · local
Modes
instant
Measuring days
21 Aug 2026 → 22 Aug 2026 · 1 field diffs in events.jsonl
Raw data excerpt (K5-01__instant, 2026-08-21)
{
  "routing.query_modes": [
    "search",
    "slow"
  ]
}

Daily data of 21 Aug 2026 →

Context

web-run/open

Impact

For tools: open|turn0search23 is not a search query; it opens a result. Parsers that count every pipe line as a query overstate the fan-out. Separate field search.opened_ref_count.

For visibility: Opened pages are read, listed pages are merely seen. Segonzac: 74 % of opened pages get cited, 7 % of those only retrieved. The step from hit to open is the biggest filter before a citation.

Status history

21 Aug 2026observed
22 Aug 2026confirmed

Open

In standard mode we saw not a single open across three days (opened_ref_count 0 in all 33 runs). Does only thinking/deep research open pages, or is opening invisible in the stream?

What decides which hit gets opened — position, domain, excerpt?