How it works
The product runs one loop: import your posts, confirm the numbers, analyze, inspect the evidence, then test what you learned. Here is each step, and what the labels mean.
1. Import 10 to 20 recent posts
Enter each post as your platform app shows it: platform, format, publication time, caption. Concentrate on your primary platform and dominant format; comparisons need at least 8 comparable posts, so a scattered mix produces honest but empty results.
2. Confirm the metrics
Type the numbers you can see: reach, likes, saves, and so on. Leave anything unavailable blank; a blank never becomes zero. Your follower count at publication unlocks extra comparisons. Quick tags (how the post opens, its topic, whether it has a call to action) unlock more comparisons; they are optional. Nothing is analyzed until you explicitly confirm.
3. Run analysis
The engine compares groups of your own posts using medians, and only speaks when the evidence clears fixed rules: at least 3 posts on each side, at least a 20% difference, stable when the strongest post is removed, spread over at least two weeks, and metrics at least 72 hours old. Zero findings is a real result, and it always says exactly what would unlock more.
4. Inspect the evidence
Every finding shows its sample, period, baseline, effect, confidence, the posts behind it, and any counterexamples. Mark it useful or not; that feedback shapes the pilot.
5. Test it as an experiment
Convert a finding into a plan: one variable, a target, a planned sample. Publish on your platform, import the results, and let the predeclared rule judge the outcome: supported, not supported, or inconclusive.
6. Reports
Generate a frozen snapshot of findings, experiments, and data health at any time. Reports never change afterwards, even when your data does.
What the labels mean
- Platform Data: a number you read in your platform app and confirmed, stored with the exact label you read.
- Calculated Result: a deterministic formula over your confirmed numbers; the formula is always shown.
- AI Observation: not used in this version. The label exists so you can always tell where a statement comes from.
- Confidence: a transparent label built from sample size, effect size, stability, and time span. Higher labels are deliberately disabled until the method earns them.