radar · production
24h7d30d
window anchor: newest run
Tier 2 compare by version
Runs this window
1,047
of 1,911 all-time
Failure rate
23.8%
▲ 18.9 pts vs prior window
Completion rate
52.4%
▼ 19.6 pts vs prior window
Unclassified
3.2%
excluded from rates, never hidden

Completion trend

Owner question #1 · #3 — “a rolling completion rate, windowed so drops are visible” · outcome mix per day

completed 52.4% rejected 9.8% starved 8.1% handed_off 2.7% failed 23.8% unclassified 3.2%
90% 70% 50% 30% 1.3.2 · 73% reddit parser breaks — every span still green 55% Jun 25Jul 2Jul 8
1.4.0 (current) — 55% 1.3.2 — 73% ▼ 18 pts between releases
GET /api/metrics/completion-trend?window=7d&by=version — via the query seam (AgDR-0013), never raw SQL

Tool & channel breakdown

Owner question #1 · #2 — failure rate per channel, and consumed-vs-called (was the result actually used?)

ChannelCallsConsumedFailed (window)
reddit2462% 89.0%
hn41261% 1.2%
web38957% 1.6%
arxiv11864% 2.5%
Every row clicks through to its runs — nothing on this screen is a dead end.
VersionRunsCompletedFailed
1.3.249873.1% 3.9%
1.4.054955.0% 23.8%
Mix-shift caution: traffic mix moved between releases — judge the per-channel split within each version, not this global rate alone.
GET /api/metrics/breakdown?by=channel&window=7d&compare=version

Decision-path funnel

Owner question #2 — where candidates die between fetched and shipped

candidates fetched
11,240
passed relevance gate
5,058
gate discards 55% — largest single dropout
consumed by draft
3,890
run completed
1,362
GET /api/metrics/funnel?window=7d

Show me the failing runs

Owner question #4, the GH-71 bar — aggregate → filtered run list → single-run span table. “89% failed” is a claim; these rows are the evidence.

Overview → channel = reddit outcome=failed window=7d · 219 runs, newest first
RunStartedOutcomeStepsTokensFailure signature
run_9f27c1Jul 8, 14:03failed 63,412search_failed · 0 results parsed
SpanKindStatusAttributes
windsock.runsessionOKchannel=reddit · version=1.4.0
planllmOKtokens=812
search.reddittoolOKresults=0 · http=200 · 412ms
gate.relevancegateOKcandidates=0 · passed=0
search.reddit (retry)toolOKresults=0 · http=200 · 388ms
draftllmSKIPPEDno candidates to consume
outcomestampOKwindsock.outcome=failed · reason=search_failed
Every span is green — HTTP 200, no exceptions. The run still failed: the channel returned nothing to consume. This row is why the aggregate view exists.
run_9f26aaJul 8, 11:02failed 63,388search_failed · 0 results parsed
windsock.runsessionOKchannel=reddit · version=1.4.0
search.reddittoolOKresults=0 · http=200
outcomestampOKwindsock.outcome=failed · reason=search_failed
run_9f251dJul 8, 08:01failed 73,590search_failed · 0 results parsed
windsock.runsessionOKchannel=reddit · version=1.4.0
search.reddittoolOKresults=0 · http=200
outcomestampOKwindsock.outcome=failed · reason=search_failed
Click a run to expand its span table. Deliberately a plain table — analytics first, not a trace viewer.
GET /api/runs?channel=reddit&outcome=failed&window=7d · GET /api/traces/run_9f27c1

Design draft for RM-007 · numbers from the July 2026 Radar dataset (1,911 runs; reddit 89.0% vs hn 1.2% / web 1.6%) · views built backwards from user-discovery/ranked-questions.md · all panels read the AgDR-0013 query seam · mock data, not live.