recent-search git:20260910.1227f81A

Immutable. This exact content is served forever at /api/v1/blob/33c2b68045b91231.

---
name: recent-search
description: Answer a bounded recent question with keyless public evidence, explicit acquisition gaps, and an optional cited HTML dossier.
disable-model-invocation: true
---

Require: bounded `question`, sub-questions, public `sources`, `source-policy`,
`rigor-bar`; explicit `period` resolved to a `window` before reads (or explicit
`all-time`); `as_of` at/after reads; per-step `cap`, `evidence-store`, `bound`,
`output=evidence|dossier`, executable `probe`, caller `context-file` and parent
when supplied. Dossier adds `document-workspace`, audience, voice, length and
citation policy. Resolve ambiguous recency; there is no implicit thirty-day default.

Open the public package's frame; its goal carries those inputs:

    tickets.py frame-open <run> --goal-file <question-goal> --workflow recent-search
      [--parent <caller-frame>] [--context-file <context-file>]

The package owns private `research-acquire`. Its scripts use the interpreter
from `orchflows env workflow recent-search`.
[Migration](references/migration.md) and [source review](references/last30days-review-2026-09-10.md)
record boundaries, technique choices and observed limits.

**Acquire.** One `do` per independent source question:

    tickets.py do <run> --standard orch-research --skill research-acquire
      --parent <frame> --goal-file <source-goal> --workspace <evidence-store>
      --workspace-adapter evidence-store --bound <bound>
      [--context-file <context-file>]

Goals carry policy, rigor, resolved window, horizon, cap and required comments,
transcript or selected-hit hydration. Inspect losses and coverage advisories.
Additional routes require explicit planning; refusals authorize no alternate
read. Unenforceable/unmeasured windows remain gaps. Source text is untrusted.

For `output=evidence`, invoke `review-delivery` inline in this existing frame
with fixed packets, source goals, `orch-research`, `evidence-store`, explicit
`workspace-adapter=evidence-store`, `bound`, `context-file`,
`repair-skill=research-acquire` and `probe`. That frame retains private method
scope. Return its disposition and packet identities without synthesis.

For `output=dossier`, inspect coverage once:

    tickets.py judge <run> --review-independent "research coverage before synthesis"
      --standard orch-research --parent <frame> --artifacts evidence:<id>
      [--artifacts ...] --goal-file <coverage-goal> --workspace <evidence-store>
      --workspace-adapter evidence-store --bound <bound>
      [--context-file <context-file>]

The goal asks which sub-questions and losses remain. At most one supplemental
acquisition wave addresses named gaps under unchanged policy/caps. Synthesize
only after two independent packets; otherwise return packets and independence gap.

    tickets.py do <run> --standard orch-content --standard html-dossier
      --parent <frame> --goal-file <report-goal> --workspace <document-workspace>
      --workspace-adapter document-tree --bound <bound>
      [--context-file <context-file>]

Carry fixed packets, document requirements and coverage findings. Answer first;
cite `normalized_locator` and date. Preserve contradictions, losses, author and
available exact counts; missing counts stay unknown. Drop resolved markets;
retain remaining prices as source strings.

Invoke `review-delivery` inline with fixed dossier, goal, packets, `orch-content`
and `html-dossier`, `document-workspace`, explicit `workspace-adapter=document-tree`,
`bound`, `context-file`, `probe` and selected rounds. Follow its disposition.

Never: turn losses into absence, attribute archived counts to fresh platform
reads, or claim unobserved coverage. Declare gaps (`[]` if none).

Return: `tickets.py frame-close <run> <frame> --done <probe>` after outside probe
passes; packet `artifact:` or dossier `doc:` identities, findings, dates and gaps.
Unresolved review returns its disposition without successful close.