Use this when
find statistics, research sources, verify sources, competitive research, find data.
Process
Before searching, check whether the topic is a known research trap (a vague demographic ask, a suspiciously round number, an overly literal phrase, or a generic single noun); if so, ask for a sharper topic instead of running doomed searches. For named-entity topics, decompose into discrete searchable angles: the primary entity's own statements, counter-perspectives from critics or competitors, practitioner discussion, and any tangential entities worth checking.
For every statistic, record the exact value, source name, URL, and publication date, and verify the number actually appears on the source page before passing it along. Only use Tier 1 sources (government, standards bodies, major research institutions) or Tier 2-3 sources (recognized industry research firms, established trade press); reject content mills, unsourced roundups, and affiliate sites, and red-flag round numbers with no named methodology.
When multiple sources cite the same upstream study, treat them as one source, not several, for coverage purposes: name the upstream as the primary citation and mention secondary write-ups only when they add original analysis.
Score research quality across five weighted dimensions before handing it off: groundedness (30, every claim ties to real evidence), specificity (25, named entities and exact numbers beat vague phrasing), coverage (20, at least two independent sources per load-bearing claim, ideally proponent and critic), actionability (15, the reader can do something concrete with it), format compliance (10, inline citations, no invented titles, no em dashes). Research scoring below 70 needs remediation before it reaches Wren; below 50 is a do-over.
Flag any claim that cannot be verified instead of forwarding it as a weak citation, and hand off only claims that clear this bar.
Attribution
This playbook's research-quality framing, including the scoring approach and the pre-flight trap checks, is adapted from the last30days-skill project (Matt Van Horn, MIT License, https://github.com/mvanhorn/last30days-skill).