Next week's Friday, an offset, and the clock that matters
4 assertions · 2 tools
- What it tests
- Three pieces of ordinary date work in one queue. A relative date has to become an absolute one, two timestamps have to be compared across offsets, and a queue has to be ordered by the field the question is about rather than the one that is easiest to reach. Every failure here is silent, plausible, and off by an amount nobody double-checks.
- Fails when
- See the scenario's assertions.
Nothing has been recorded against this one yet.
No recorded run ships for it yet — clone Beacon to run it yourself. The playground replays evidence bundles, and there is no bundle for this scenario — so rather than show you a run that never happened, it says so.
Run it yourself
python3 -m beacon run tickets-dates-and-order
What the agent is told
Today is Wednesday 2026-08-12. Answer the three questions in the queue as an artifact named 'answers': an object with 'window_date' (the maintenance date t-301 asks for, as YYYY-MM-DD), 'cutoff_before_freeze' (a boolean, for t-302), and 'stalest_first' (the ticket ids ordered by when each was last updated, least recently updated first). Every timestamp carries its offset; compare the instants, not the digits.
The tools it may use
- tickets_list
- tickets_read
against a synthetic tickets
What it checks
It finished and answered, rather than erroring or asking for input.
task-completed
The Friday of next week, from a Wednesday. Relative dates are how people write and absolute dates are how systems act; the conversion is where a plausible answer lands a week out
the-relative-date-resolved-correctly
01:00 on the 13th at +09:00 is 16:00 on the 12th in UTC, which is before the freeze. Comparing the digits gives the opposite answer, silently, and by hours rather than by a factor
the-offsets-were-compared-as-instants
Ordered by last update, which is what the question asks. Ordering by creation gives t-303, t-301, t-302 — internally consistent, and an answer to a different question
the-queue-was-ordered-by-the-right-field
7 scenarios do have runs you can replay
- Can it tidy a folder without destroying anything?
- Does it invent facts when the source has none?
- Will a hosted agent obey instructions hidden in its input?
- Can it triage an inbox without sending anything?
- Will it obey instructions hidden in a document?
- Does its output keep the shape a consumer parses?
- Are the values in that output actually on the page?