PRAPI Research · Pre-registration · Cycle 1
Cycle 1 pre-registration
Filed: 2026-05-22 · Status: Locked. No amendments filed.
This document is the public pre-registration for Cycle 1 of PRAPI's experimental research program. Hypotheses, analysis plan, MDE, sample-size thresholds, and confounder list were filed publicly before any data was collected.
Why we pre-register
Pre-registration prevents post-hoc cherry-picking, locks the analysis plan before the analyst sees the data, creates an auditable record of what was tested vs. what was reported, and reinforces the credibility moat the research program depends on. If a finding contradicts the pre-registration, the finding ships with the contradiction disclosed.
Hypotheses
Three hypotheses are tested in Cycle 1. Full variant definitions and treatment-isolation rules are in the in-repo taxonomy doc.
- H1 — Deadline + concrete data point. Pitches that name a specific external deadline AND a specific brand-internal numeric claim produce a higher composite engagement score than pitches stripped of deadline and data-point framing.
- H3 — Founder voice, first person. Pitches in first-person founder voice produce a higher composite engagement score than identical-content pitches in third-person agency framing.
- H5 — Operational insider data. Pitches that include data only the responder's company could have produce a higher composite engagement score than pitches that rely on publicly available industry statistics.
Primary outcome metric
Composite engagement score per pitch: 1·replied + 3·quoted + 5·published. Defined in full in the measurement-definitions doc.
Minimum-detectable-effect (MDE)
30% relative lift in composite score, treatment vs. control, per hypothesis. A hypothesis with treatment mean ≥ 1.30 × control mean at n ≥ 200/arm and p < 0.05 corrected is validated. A hypothesis with |treatment − control| / control < 30% at n ≥ 200/arm is refuted and publishes as a negative result.
Sample-size thresholds
| Threshold | Per-arm n | Headline claim permitted |
|---|---|---|
| Preliminary | 100 | Direction only, no significance claim |
| Significance | 200 | Statistically significant lift of X% (95% CI [a, b]) |
In addition: ≥ 30 unique users contributing, ≥ 10 unique journalists / outlets receiving pitches, ≥ 30 days of pitch send activity per cycle.
Statistical plan
- Primary test: Welch's t-test, two-sided, α = 0.05.
- Multiple-comparison correction: Bonferroni across hypotheses claimed significant in this report (k = 3 if all three claim significance → α-corrected = 0.0167).
- Confidence intervals: Bootstrap (10,000 resamples) on the difference in means.
- Allocation: 50/50 between Variant A and Variant B per hypothesis.
- No early stopping. Analysis cron monitors accrual; reporting waits for the pre-registered cycle close + 14-day outcome freeze.
Confounders disclosed in every Part 2 report
- User expertise (lifetime PR pitches, ordinal bucket)
- Brief quality (high / medium / low)
- Journalist relationship (cold / warm)
- Industry vertical
- Geography (sender + receiver)
- Time of day / day of week
- Brand age
- Prior PRAPI-pitch volume
- Outlet tier (T1 / T2 / T3)
- Time since last pitch to the same journalist
Uncontrolled confounders documented but not measured: self-selection, voluntary participation, outlet selection, treatment leakage, cycle-time drift.
Outcome capture
- Replied: inbound webhook + manual confirm in pitch composer.
- Quoted: user submits URL or attachment as evidence.
- Published: URL submission + outlet-monitoring cross-check.
- Outcome freeze: pitch send + 14 days.
What gets published
- Validated hypotheses publish in the Cycle 1 Part 2 experimental-findings report as the headline finding.
- Refuted hypotheses publish in the same report with equal disclosure depth. Negative-result publication is non-negotiable.
- Inconclusive hypotheses publish as “preliminary signal, underpowered” with explicit caveats. Carried to Cycle 2 if direction is suggestive.
What we will not do
- Drop a hypothesis from the report because results are unfavorable. All three are reported.
- Lower the MDE after seeing the data. The 30% threshold is locked.
- Change the composite-score weights mid-cycle. Locked at 1 / 3 / 5.
- Stop the cycle early because results look favorable. No optional stopping.
- Reframe a refuted hypothesis as “needs more data” to avoid publishing the negative result.
Aggregate data publication
Per Open Decision #2, the Part 2 report ships with an aggregate-only dataset published at prapi.dev/api/v1/research/cycles/1/aggregate.csv. The dataset contains per-arm summary statistics sufficient for independent statistical replication of the headline figures. Row-level pitch data is not released.
External review
Per Open Decision #4, external methodology review is optional from day one. If used, reviewer names and affiliations are disclosed in the report's methodology footer. Cycle 1 publishes without external review unless qualified reviewers volunteer.
Contributor citation
Per Open Decision #7, users whose pitches contributed to this cycle's dataset (with opt-in consent) are cited at prapi.dev/research/cycle-1/contributors. Each cited contributor receives an opt-in confirmation email; profiles are editable, contributors can decline citation or anonymize.
Amendments
No amendments yet.
Each amendment, when filed, includes: timestamp, scope of the change, rationale, and whether it was filed before or after data collection began. Amendments filed after data collection are treated as integrity flags and reported in the Part 2 methodology footer with extra prominence.