Methodology
How to read this page
These rules apply to every figure here. They are on the page and not in a footnote because none of the numbers can be judged without them.
01
Recorded figures and live figures
Live figures are recomputed from the pipeline's stored output every time the page is built: collection volume, enrichment lag, the daily run. Recorded figures are measurements made once, by hand or in a one-off study, and transcribed with the document and date they came from: the benchmark reproduction, the filter error rate, the blind reading, the negative results. A recorded figure is never refreshed, and its date is printed beside it.
02
Every number states its basis
The line marked Basis above each table says what the number is a fraction of, what it counts, and what it does not. A row count is rows written, not distinct posts. A precision is precision on 88 labelled posts, not on the corpus.
03
Missing means "Not available"
If a source cannot be read, the card says so and shows no number. There is no sample data on this site and no fallback value. If the page cannot read a summary at all it says that and shows nothing.
04
What the collection is
Collection samples X; it is not a census and not a random sample of Kenyan discourse. Several partitions are collected because something about them looked interesting, so no rate over them has a denominator. That is why no prevalence figure is shown, and why collection volume is split by the partition that produced each row.
05
The fourteen-day horizon
X search reaches back 14 days. Anything not collected inside that window is gone, so a day with no data is a permanent hole and is drawn as one.
06
Triage, not verdict
The detector ranks accounts for a human to look at. It does not establish that any account is inauthentic, and the coded-term register it uses has innocent everyday senses. The listing it produces is account level and is not published.
What is not published
- Any prevalence rate. Nothing in the collector samples the discourse at random, so no rate over it has a denominator. A control arm is needed first.
- Any account, handle or per-account claim. Detection says accounts act together; neither it nor a reader's judgement establishes that a named account is inauthentic.
- Counts of coordinated accounts or communities from the live run. The listing length is a triage budget and the community count depends on a clustering resolution and seed. Neither is a finding.
- A toxicity series over time. The raw series moved with the collector's own target list. Only a composition-standardised or within-partition rate may be shown, and none is shown yet.
- Cross-channel corroboration of clusters. The earlier statement about it is false and was retired.
What we will not claim
- Not a vote forecast. No model, no voter file, no backtest, no win probability.
- Not Kenyan opinion. This is speech observed on X, which is not the Kenyan electorate.