The cycle in one paragraph
PRAPI Research Cycle 1 tests three pitch-framing hypotheses surfaced from the descriptive survey of 24 practitioners in Report 1. Each hypothesis is pre-registered with locked variant definitions, a 30% minimum-detectable-effect, a sample-size threshold of 200 per arm for significance claims, and a 10-item confounder disclosure list. The analysis pipeline is open-source and the aggregate dataset publishes alongside this page for independent replication.
What we're testing
- H1 — Deadline + concrete data point. Pitches that name a specific external dated event (regulatory deadline, conference, earnings window) AND a brand-internal numeric claim outperform pitches with both stripped to generic framing.
- H3 — Founder voice, first person. First-person founder voice ("I", "we") outperforms third-person agency framing ("[Brand]'s founder reports...").
- H5 — Operational insider data. Claims sourced from the responder's own systems (internal product metrics, named customer outcomes, support-ticket aggregates) outperform closest-equivalent publicly available industry statistics.
Full variant definitions and treatment-isolation rules: pre-registration.
Status
This cycle is accumulating. The auto-filled methodology footer below shows the live per-arm sample size from the PRAPI corpus. Numbers refresh each time you load the page. Once the cycle clears the 200-pitch-per-arm threshold AND the 14-day outcome freeze, this page flips to the final report.
What ships when the cycle closes
When the threshold clears, this page becomes the Part 2 findings report:
- Per-hypothesis verdict (validated / refuted / inconclusive) with the headline relative-lift, 95% CI, Welch p-value, and Cohen's d.
- Confounder breakdown — per-arm distribution across the 10 confounders, with stratified sub-analyses for any dimension showing material imbalance.
- Negative results publish too. If H1 / H3 / H5 is refuted at n ≥ 200/arm with |effect| below the 30% MDE, that finding ships with the same disclosure depth as a validation. This is a pre-registered commitment.
What does not ship
- No row-level pitch data. Aggregates only (see CSV download link in the methodology footer below).
- No journalist identification. Pitches are aggregated across outlets; no individual journalist names, addresses, or beats surface in the report.
- No contributor identification without explicit opt-in. Users whose pitches contributed appear in /research/cycle-1/contributors only if they opt in via account settings.
How to participate
If you're a PRAPI user, you can opt in to the experimental track at app.prapi.dev/pr-pitch/settings#research. Opt-in is reversible at any time; opting out preserves historical rows in the corpus (per the pre-registration) but stops future capture. Per-pitch exclusion is also supported — you can flag any specific pitch as not-for-research from the outcome capture queue.
If you're a journalist or researcher who wants to look at the data, the aggregate CSV is published alongside this page on every cycle.
References
- Pre-registration (Cycle 1)
- Methodology
- Descriptive findings (Part 1, Report 1)
- Open-source analysis pipeline:
github.com/Startvest-LLC/marketing-agent
under
src/lib/research/experiments/.