Research
Research
AFP publishes one study so far, and it has no results yet: the comment-to-DM benchmark, whose method for measuring match rate, delivery rate, and time to first DM is written down before any data exists. No measured number appears anywhere in this section today. When the first run finishes, its results and the raw dataset are published on the same page as the method that produced them.
What is published here
Section titled “What is published here”| Study | The question it answers | Status |
|---|---|---|
| Comment-to-DM benchmark methodology | How should comment-to-DM automation performance be measured? | Methodology published; data collection not started |
What a study here has to do
Section titled “What a study here has to do”Five rules apply to every page in this section, and a page that cannot meet them does not ship as research.
- Definitions before numbers. Every metric is defined precisely enough that a stranger reading only the definition would count the same events.
- The procedure is published in full — sample, duration, environment, what was held fixed and what was not — so the study can be re-run by someone who does not work on AFP.
- The raw dataset is published, in a documented format, next to the summary. A summary with no dataset behind it is an assertion.
- Limitations are listed. Anything that would make a reader discount the result is stated on the page rather than left to be discovered.
- Every number is measured. No figure is estimated, rounded up from a hunch, or carried over from another study, and each one is dated and attributed to the run that produced it.
Why the method is published before the numbers
Section titled “Why the method is published before the numbers”A number with no method behind it cannot be checked, so it carries no weight — and a protocol written after the data is in can always be shaped to flatter the result. Publishing the procedure first fixes what counts as a success before anyone knows how the run went, and it lets a reader disagree with the design instead of arguing with a bare figure.
Nothing about an unfinished study is hidden here. The status column above says exactly how far each one has got, and it changes only when a run completes.
What AFP can already measure about itself
Section titled “What AFP can already measure about itself”Three surfaces in the product record what automation did, and every measurement in this section is built from them rather than from instrumentation a reader cannot see.
| Surface | What it records | What it is good for |
|---|---|---|
| Campaign execution history | The path each run took through the graph, the node it stopped on, and the error recorded against a failure | Whether a comment matched, and which branch it took |
| Activity feed | One row per comment AFP handled on a connected page, with the outcome recorded against it and an errors filter over the three failure kinds | Counting outcomes across pages, and finding comments AFP could not act on |
| Activity export | The filtered feed as a JSON download, with the export time in the filename | Analysis outside the product, and publishing a dataset |
The execution history shows the 20 most recent runs of a campaign and does not refresh itself, which is the main practical constraint on batch size in any study that reads it. Test and debug a campaign covers how to read both views.
Test executions are deliberately not a measurement instrument for delivery or timing. They walk the same graph as a real run but make no calls to Facebook or Instagram, so they can confirm which branch a message takes and nothing about how fast or whether a DM arrives.
What is not here
Section titled “What is not here”- No competitor performance numbers. AFP has measured none, and the comparison pages mark every competitor claim as unverified public material rather than dressing it up as data.
- No deliverability or reach claims. What Facebook and Instagram do with a message once AFP hands it over is outside anything measured here.
- No plan quota figures. Quotas are per-plan configuration that changes; Plans and quotas explains which limits exist without pinning numbers that would go stale.
- No translations. Research pages stay in English, so the method and the dataset have exactly one authoritative version.
Related
Section titled “Related”- Comment-to-DM benchmark methodology — the metric definitions, the procedure, and the raw-data format.
- Route comments to a DM — the campaign shape every measurement here is taken against.
- Execution states — what a run’s recorded state means when you read it back.
- Moderate comments — the activity feed and its export, from the operator’s side.