Skip to content

See how you have labelled the brain's search

Machine verified A machine ran this on our install and checked the result against the CLI. A test report is linked. Proven against quoxcli 0.6.1 on 2026-11-14.

How am I testing what the brain finds?

Each time you mark a search result as right or wrong, the brain gains a test it can be scored against. One question lists them all.

What you get: every label you have given: the question, whether a note was a right or wrong answer to it, and any question you said the brain should not answer at all because nothing in it should.

quox · talking to Quox

I've been labelling search results to test how well the brain finds things. How many labels have I given so far, and which question did I say the brain should not answer at all?

▤ AOCL │ L1 identity complete · allowed · 2ms

▤ AOCL │ L3 policy gate complete · proceed · 12ms

◐ QUOX │ command lookup ✔ QUOX │ command lookup │ matched: true, 0 tui, 1 cli · 1.9s

⛓ AEE │ envelope opened · ops.tool · quox_command_lookup ok

◐ QUOX │ cli ✔ QUOX │ cli │ 0 exit code, 7168 duration ms · 11.4s

⛓ AEE │ envelope opened · ops.tool · quox_cli ok

⛓ AEE │ envelope opened · stream_success

You've given 3 labels so far.

The question you said the brain should not answer at all is "what does the e-bike 
insurance cost". You marked it "should abstain" on 12 November at 16:14 UTC, which means
nothing in the brain should answer it.

The other two labels are both for "when is the workshop open at christmas". You marked
one note as relevant and a second note as irrelevant.

If you want to see how the brain scores against these labels,  quox brain eval  replays
them against live search.

25.2s · 2 tools · 0 approvals

A real turn on our own install (CLI 0.6.1, 12 Nov 2026), replayed at speed. Names changed. Type the same words into quox.
✔ Machine verified against the CLI on 12 Nov 2026 · test report mwdrja

Quox read your labels: three in all. The question the brain should not answer is “what does the e-bike insurance cost”, and the other two mark one note right and one note wrong for “when is the workshop open at christmas”. Quox only read the labels; it did not add one or run a score.

brain judgements
quox brain judgementsVerdict           Query                                     Note                                        Atshould_abstain    what does the e-bike insurance cost       -                                           2026-11-12T16:14:19.955Zirrelevant        when is the workshop open at christmas    43b52a353a4d401ebdfceb3d3a238003eb1c5f7c    2026-11-12T16:14:18.803Zrelevant          when is the workshop open at christmas    8f00901c9e59f44e69f367c256d60f2bbc6f58b1    2026-11-12T16:14:17.799Z
From the same scratch workspace (fictional Kestrel Cycles data), run after the turn above.

Each row is one label, newest first. should_abstain has no note, because it says no note should answer that question. You add a label with quox brain judge.

Quox and the terminal read the same three labels. Listing them changes nothing. Together they are the test set the brain is scored against: quox brain eval replays each labelled question against live search and reports how often the right note came back, and whether the brain held back where it should.