How the data behind EA Hospitality Pulse is built, verified and updated.
A market-intelligence product that asks you to trust its numbers should show its work. This page explains, plainly, how each dataset is built — what counts as a source, how confidence is graded, and what a stat on the homepage actually means before you act on it.
A fixed basket of 40 properties across six East African markets (Nairobi, the Kenyan coast, Zanzibar, Kigali, Kampala, and the safari circuit), tracked weekly at a constant booking convention so that a change in the index reflects a change in rate — not a change in what was measured.
Each market's figure is the median of that week's observed lead-in rates across its basket, not the mean — a median is far less distorted by one outlier property having a promotional rate that week. A market's week is only shown as confident once at least 3 observations (minN) were collected for it that week; below that, the figure is published but flagged unconfident, because a sample of one or two properties isn't yet a market read.
Properties were selected to span each market's actual segment mix — city business hotels, safari lodges and camps across price tiers, and beach resorts — rather than skewing toward the largest or most marketed names. The basket is fixed once set; we don't swap a property out because its rate moved in an inconvenient direction. Basket changes, when they happen, are logged with a reason.
A weekly sweep of the inputs that move a property's margin rather than its top line — fuel, currency, energy and food costs across the five markets. Currently 14 tracked line items. Every figure carries a named source and a date; where a figure is a same-day reported number rather than an official release, that distinction is stated rather than blurred.
Every entry rule, permit price, park fee, levy and surcharge we track across the five markets is marked verified or needs re-confirmation. As of this writing, 19 of 23 entries are currently verified against a primary source (a gazette, a regulator's own page, or an official fee schedule) within our re-check window; the remainder are marked to flag that they should be re-confirmed with the official portal before being quoted to a client. We would rather show that flag honestly than let a stale fee sit uncorrected on the page.
Every item we publish is graded, and the grade is printed on the item itself:
| Grade | What it means |
|---|---|
| Confirmed | Verified against a primary or official source — a regulator's own notice, a government gazette, an airline's own schedule filing. |
| Reported | Carried by credible press but not yet independently confirmed against a primary source. |
| Early signal | An upstream indicator — a filing, a schedule change, a procurement notice — that suggests something is coming before it's been reported anywhere. Labelled as inference, not fact, until it resolves. |
A recency gate runs before anything publishes: every item must carry a publication date, an event date inside the edition's coverage window, and pass a novelty check — if a well-informed operator already knows it, it's catch-up, not intelligence. Evergreen pages, undated aggregators and recycled press releases are treated as stale until proven otherwise, and we chase a claim back to its primary source before trusting the date a search engine shows for it.
Every forward call we publish is logged with a falsifiable resolution test and a resolve-by date, before we know the outcome. As of this writing: 43 calls logged, 41 still open, 2 resolved (1 correct, 1 partial, 0 incorrect) — a 75% accuracy figure on a genuinely small resolved sample so far. We show that number rather than waiting for a rounder one, because the point of a public ledger is that you can check it at any stage, not just when it looks good.
A call is only logged if its resolution criteria name a specific, checkable source in advance — "confirmed by KNBS's next release" or "resolved by the official gazette notice" — not a vague future vibe check. When a call resolves, the evidence and source are published alongside the verdict, including on calls we got wrong or partially wrong. Nothing is quietly removed or re-worded after the fact; corrections are published as corrections.
Undated content, listicles, evergreen SEO pages and copyright-year-only pages are not treated as sources regardless of how they rank in search.
See the FAQ for shorter, specific answers — including what the Rate Index is for, how advisories actually affect bookings, and whether this is free.