Internal specification · how FishDog research is presented
Report Format v2
The presentation rules FishDog deliverables follow, and where each rule comes from. Every rule below is derived from panel research with senior research buyers — the studies on this site are rendered to this specification, and this page is the specification. Reports themselves carry none of this commentary.
Structure
The deliverable is a bundle, not a page
| Layer | Artifact | Job |
|---|
| 1 | Decision brief — 2 pages, print/PDF | What the buyer works from. Page 1: the decision with effect sizes and ranges, plus a scope box (N, field dates, sample frame, who paid) above the fold. |
| 2 | Interactive page | The audit layer behind the brief: filter, drill from any headline number to tagged verbatims. Never the system of record. |
| 3 | Data pack — CSV + codebook | Serves the rebuild test. Buyers verify a number against something they trust before acting; if they can't rebuild it, they won't cite it. |
| 4 | Board copy — 1 page, generated | The consensus-shaped summary readers otherwise hand-build for the level above them. Always labeled as the summary layer over the split. |
All four layers carry the same version stamp, and their numbers match to the digit. Changes get a changelog line; numbers never move silently.
Page order
What a report opens with
- Scope strip. N, field dates, who the respondents are, method, version, exports — visible before anything else.
- Decision statement. One sentence: what to do, with the effect size. The decision, not the findings.
- The split. Positions with counts, conviction per camp, who holds which view, and what would flip each camp. Disagreement leads; any consensus statistic is derived from it, never the opener.
- Position detail with receipts. Every stance drills to the respondent's own words, tagged with role, segment, and date.
- Contested index. All questions ranked by disagreement. Unanimity is reported with a candor note, not a victory lap.
- Board copy. The liftable one-page summary, print-ready.
- Panel ledger with recruitment candor. Recruited-not-generated provenance, over-recruit and curation counts, skews and title-resolution limits stated plainly.
- Provenance box. Where models were used, how the analysis was checked, links to raw responses and the data pack.
- Appendix. Full narrative, confidence grades, limitations.
Hard rules
Non-negotiables, each tied to its evidence
- Never gated. No login, no expiring links, no appendix behind a form. Exports on everything. Gating was a unanimous adoption-killer in every panel fielded.
- Version-frozen. Visible version and date; silent number drift destroys trust ("I walk into Monday quoting ghosts"). Near-unanimous across panels.
- Visible N everywhere. On every chart and filter state, with small-n candor on the tile. Hidden denominators are kill-signal #1.
- Exact tallies, ugly actuals. No rounding distributions into tidiness; report gaps as gaps. Too-tidy numbers read as modeled fluff — a top machine-tell.
- Verbatims keep their burrs. 2–8 per document, tagged role/segment/date, placed next to the finding they evidence; one challenging quote where the data allows. Buyers detect authenticity by quote texture; sanitized quotes read as composites.
- Charts ~70%, narrative ~20%, quotes ~10%. One-line so-whats; labeled axes; zero-based bars; no 3D, no decorative gradients. Chart crimes are a unanimous kill-signal.
- Effect sizes with ranges. Never hero point-estimates without spread; falsification triggers stated plainly. "I trade the error bars, not the average opinion."
- No boilerplate voice. The "in today's rapidly evolving landscape" family is banned; uniform prose cadence is machine-tell #1. Linguistic variety is a product requirement.
- AI provenance disclosed. Where the model was used, how it was QA'd, link to raw responses. Requested unprompted by buyers; candor is cheap and differentiating.
- Reports carry no meta-commentary. The report never explains its own format or congratulates its own design. This page is where the format lives; the report just follows it.
Provenance of the rules
Where this specification comes from
Format v2 was derived from FishDog panel research on how senior research buyers consume, trust, and reuse external research: a three-panel study (38 curated respondents: insights buyers, investment professionals, C-suite operators, fielded 2026-07-10) and a two-panel VP-and-up replication at n=100 (50 senior finance decision-makers, 50 C-suite executives). Key inputs: divergence-first presentation beat consensus-first presentation in every panel; the two-page brief is the near-universal working format; a web page is accepted only as the audit layer behind a frozen artifact; buyers strip-mine reports and re-author them upward, so a generated board copy replaces work they do by hand today.
The full evidence, counts, and verbatims live in the studies on the study index.