Coffee Monitor: Ethiopia
Independent evidence on Ethiopia's coffee economy · Seattle, Washington
Global Coffee Observatory · Data and releases
The Observatory publishes its panel, controlling methodology, versioned data and verification manifests separately from the interface. A reader should be able to check what a map state rests on without trusting the map itself.
Frozen and publicly downloadable are two different things, and this page keeps them apart. An artifact that is frozen is fixed, dated and identified by a checksum; it is downloadable only once it is hosted or deposited. Files that are frozen but not yet downloadable are named here so that their absence is visible, and nothing on this page links to a document that does not exist.
Evidence Rule Book, Revision 2.7. Controlling since 1 September 2026. Publicly deposited on 14 September 2026 at doi.org/10.5281/zenodo.22740331. The first date is the controlling-version date and the second records the public deposit; they are not the same event.
Revision 2.7 is the only methodology that governs the live map. Draft Revision 3.x material is non-controlling and is isolated from live states. The rules are summarised on the Method page.
SHA-256: ef38c5e91eaf5769e12e274da6a4537988601aa39f2b85e1b14234a580b1ea9b
The hash identifies the controlling text and can be checked against the deposited copy.
The Observatory uses a fixed, versioned panel of 51 coffee origins. It is not an exhaustive enumeration of every territory reporting coffee production. Any change to panel membership is a versioned change and is published with a crosswalk.
Panel CSV and the 2024 producing-territory exclusion list: fixed and versioned, not yet publicly downloadable.
306 cells in the six-row matrix, none unreviewed. 259 reviewed with no determination, each carrying a named blocker and a reopen trigger. 44 held. 1 published rating. 1 rating disputed. 1 requiring re-adjudication.
One rating across 306 cells. That is the finding, not a reporting gap: a determination is reached far more often than it is publishable, and the gap between the two is the archive and verification work.
Producer-price slice. 102 cells. 63 reviewed with no determination, 37 held, 1 published, 1 disputed. Current determinations: 38.
An earlier pairing of 40 cells across 33 origins circulated internally and was withdrawn on reconciliation, because the two figures came from different moments and never described the same snapshot. The lower figure is not a retreat; it is what the dated cell-level record supports.
These figures are generated from the adjudication registry rather than maintained by hand. Where a figure on this page and a figure on the Observatory disagree, the registry governs and the disagreement is a defect to be fixed, not a judgement to be made between two accounts.
Research-cycle completion does not mean every cell is publicly rated. Publication eligibility is tracked separately from analytic closure. Of the 47 cells that left the research cycle with something other than no determination, 44 are held from public rating by a named publication, archive, evidence-chain, source-stability or independent-verification gate, 1 is published, 1 carries an unresolved rating dispute, and 1 requires re-adjudication. These are distinct release states and collapsing them into a single count misstates the record.
Research-final means the evidence search and adjudication cycle is frozen. That point was reached on 4 September 2026. No further country sweep is planned for this vintage, and a cell is not researched harder because its answer is unsatisfying.
Release-final is a later and separate point, reached when three things are done: the publication gates on the determined price cells are cleared or explicitly carried forward, the frozen 102-cell producer-price dataset is published with its manifest and checksums, and the prespecified independent adjudication has been run and its reliability result recorded.
Only at that second point does an edition exist. It is then named by its audit vintage, and a published edition never changes silently. Evidence arriving afterwards produces a later version with its own record, not a quiet amendment to an earlier one.
Where the project stands. Research-final was reached in September 2026. Release-final has not been reached. The closure-release v1.1 hashes below identify the frozen artifacts, but the files themselves and the release-qualified producer-price dataset are not yet publicly downloadable. Release-final also requires completion of the prespecified independent adjudication and the recording of the resulting reliability statistics.
A cell recorded as reviewed with no determination stays that way for the vintage. It reopens on one of four things and nothing else:
A cell is never converted to state one because the search stopped. State one is a positive finding and still requires the full structured negative-search record. The absence of a determination and a determination of absence are different results, and the map is not permitted to round one into the other.
Public Economic Visibility at the Producer Side of Global Coffee. The first comparative paper covers public reference price and farm-gate price across the fixed 51-origin panel.
Plan v5 was finalised on 7 September 2026, the registry was frozen on 11 September, and the plan was publicly deposited on 14 September 2026 at doi.org/10.5281/zenodo.22740569. The post-freeze deposit chronology is recorded as Deviation 1. Interim cell-level adjudications were already known during instrument development, so it is not presented as a blind preregistration.
The frozen producer-price dataset is not yet publicly downloadable. Independent validation is pending and adjudicator confirmation is awaited.
The third rule is the one that costs something. Of the 39 producer-price cells that reached a determination or a dispute, 37 are held, 1 is published and 1 is disputed. Reporting the held cells as failures would produce a dramatic number and a false one.
Coffee Monitor reports three separate questions for each producer-price row, not one score:
The denominator is fixed at 51 origins for all three counts. The thresholds are nested: everything reconstructible is recurring, and everything recurring is current. Cells unresolved at the relevant threshold are reported through identification bounds rather than counted as failures. No state numbers are averaged or converted into a composite score.
The counts are not published here and will not be until release-final. The dataset was frozen on 11 September 2026, which satisfies one condition and not the rest. Publishing the definitions first is the point: it prevents the thresholds from being chosen after the results are known.
The planned order put the deposit before the freeze. It ran the other way, by three days, and that is recorded as Deviation 1 rather than smoothed over. The cards are generated last, from the frozen dataset. No provisional count is published at any earlier step.
Every frozen release carries a file manifest and SHA-256 checksums. Released files are never silently replaced: a corrected release receives a new version and an explicit change record.
2026 closure release v1.1, frozen 11 September 2026. Controlling methodology Revision 2.7. The v1.0 artifacts are superseded and retained only as provenance.
These four artifacts are frozen. They are not yet publicly downloadable. The hashes identify them and can be checked against any copy supplied on request.
Source of truth. Public cell state for all 306 cells is generated from one status object. The explanatory prose attached to a cell may come from the research record, but it never decides public status. Where two accounts disagree, that is a defect to be fixed rather than a choice to be made.
Coffee Monitor preserves first-party evidence objects and their metadata for auditability. Third-party source files are redistributed only where rights permit.
Where redistribution is not appropriate, the public release provides the institution, the original address, the document date, the capture date, the artifact form and the cryptographic hash needed to identify the object.
Corrections and challenges
A new source or challenge triggers re-adjudication under the controlling rule book. Public state changes are logged with the old state, the new state, the effective date, the rule version and the decisive evidence. See Corrections.