Coffee Monitor: Ethiopia
Independent evidence on Ethiopia's coffee economy · Seattle, Washington
Method
Everything on this page is a first draft, published so that it can be attacked. The strongest attack wins, revisions are dated, and superseded versions stay reachable.
One thing only: what the public record of an origin lets an outsider establish about a named economic quantity, for a stated period and review date. The instrument does not measure statistical quality, institutional capacity, honesty or economic performance.
The instrument began as an Ethiopian one. It is now applied on the same terms to a fixed panel of 51 coffee origins through the Global Coffee Observatory. The panel is fixed rather than exhaustive: it is the set of origins the Observatory follows, not the set of countries that grow coffee. Ethiopia is assessed by the standard applied everywhere else, not by one written for it. That matters: a standard that only ever pointed at one country would be an accusation rather than a measurement.
A row marked no public series identified means that Coffee Monitor searched and did not find a recurring national publication. It does not mean that no ministry, regional bureau, cooperative union, research institute, or price desk collects the figure. Establishing the absence of any collection anywhere is not something an independent publication can do, and this index does not pretend to.
Equally, a low position on the ladder is not an accusation. A figure can be entirely accurate and still have no reproducible public account behind it. Publication and truth are different properties, and conflating them is the error this site exists to correct.
These five are analytic states. Whether a state appears publicly is a separate question, governed by the publication status rules below. A row can hold a settled state internally and still show no number.
The distinction between states four and five carries the weight of the whole instrument. Four means a number is published. Five means the number can be checked. Almost everything that circulates as an authoritative farm-gate figure, anywhere in the world, sits at four.
A positive rated cell rests on a dated first-party object published by a named institution, preserved as the published file where one exists or as a complete page render, with stable metadata and an archive record. A search result, assistant-surfaced link, news report, quotation, machine translation or Coffee Monitor page is not a substitute for the underlying first-party object.
State one uses a structured search record rather than a positive source document. It is still an evidence packet and must pass the archive gate before publication.
A numeric state does not automatically appear on the public map. Published means the analytic determination and its publication gates have cleared. Held, evidence pending means a current state one to five exists internally but a named evidence, archive, source-stability, right-of-reply or verification gate remains open. Reviewed, no determination means the row has been worked but no current numeric state is settled.
The archive guard is the final publication veto. It may withhold a determination whose evidence packet is incomplete or defective. It never promotes a row.
Rows without a number are not all in the same condition, and the Observatory does not display them as though they were:
Collapsing these into a single blank would hide the difference between work not done and work done but not yet provable, which is the distinction the whole instrument rests on.
The permanent archive is being built, not finished. Every cited document is captured as a file and hashed, so that a citation cannot quietly rot. Most rows that have been adjudicated are not yet through that step, and the consequence is deliberate: they publish no number. A row is held rather than rated until its document is in the archive, which is why the Observatory currently publishes a single price rating across 102 cells while many more have been decided internally. The gap between what has been determined and what has been published is visible on the page rather than hidden by it.
They are the minimum set required to answer one question: what share of an origin's coffee export earnings reached the people who grew the coffee? Two describe what left the country. One is derived from those two. One is the official benchmark price. Two describe what the grower received. Remove any of the six and the question cannot be answered. Additional rows will be added when they are needed to answer the question, not to lengthen the index.
Because weighting would require a judgment about which absence matters more, and that judgment would be mine rather than a finding. The formal six-row summary is a count of how many rows reach state five after all six rows are settled and publishable. It is not an average of the state numbers and it is not a grade. A held or not-determined row prevents a final count.
The Observatory may also show publication thresholds or status views as display aids. Those views do not change the underlying states or the controlling methodology.
Only when all six rows are settled and publishable. If any row is unrated, for any reason, the origin carries no count. That is not a statement that work is ongoing: the 2026 research cycle closed on 4 September 2026, and a row can be unrated because it was reviewed and yielded no determination rather than because it is still being looked at. A partial count would understate an origin by treating an absent determination as a determination of absence, which is the conflation this index exists to prevent.
A held row is researched and adjudicated, but not public-ready. Its internal determination is preserved for provenance while the public interface shows held, evidence pending and no numeric rating. A held row must never be displayed as not checked yet.
Implied export unit value inherits the lower state of export earnings and export volume, and its evidence chain is whole only when both input chains are whole. If either input is held or not determined, the derived row is held or not determined with it.
A derived figure is Coffee Monitor's own arithmetic. It is admissible at all only because arithmetic is self-checking: any reader can repeat the division and get the same answer. That only works if the reader can reach the two numbers being divided. When both inputs are held, the quotient stands on Coffee Monitor's authority for its own inputs, which is exactly what the self-citation control refuses one step earlier. A derived row can never make an origin look better documented than its sources are.
A document published by a named institution, reachable at a stable address, carrying a date, and captured as a file rather than as a link. The published PDF where one exists, otherwise a full page render, and not a snippet, a search result, or a quotation of the document.
Four things are not admissible, however sound the finding behind them:
A machine translation is not a document either. The source-language publication is captured as primary; an institution's own translation may accompany it.
The scored export object is green coffee only: tariff headings 0901.11 and 0901.12. Earnings are US dollars free on board. Volume is net kilogrammes, with 60-kilogramme bag equivalents shown only as an additional display. Implied export unit value is dollars per net kilogramme.
Where the public system distinguishes domestic exports from re-exports, the scored account uses domestic exports. Roasted coffee, soluble coffee, husks, skins and substitutes are outside the scored green-coffee account and may appear only as non-scoring context.
This matters when comparing a cell against a national headline figure. For the largest exporters the two will differ, sometimes substantially, because a national headline usually covers all coffee products while this index covers green coffee alone. That is a difference of scope rather than an error in either number.
An internal price desk that an institution operates, and a benchmark price that it publishes, are different things and are adjudicated separately. Ethiopia has both. Neither currently carries a rating, and the reasons differ: the desk because no institutional document for it has been located, the published benchmark because its adjudication is open.
An externally generated market quotation qualifies for the public reference price row only when an authority of the origin country formally adopts or operationalises it in a domestic pricing, registration, settlement, regulatory or comparable institutional process. Mere quotation or market context does not qualify.
The scored row uses the best qualifying public account on the fixed green-coffee scope, whichever institution publishes it. An origin is not marked down because a coffee authority issues a second, less reproducible series alongside a reconstructible customs account. Every other compatible official series is retained as a comparator, and any material disagreement between them is recorded as a defect against the origin rather than silently resolved.
The revision number at the top of this section is the controlling text, and it has a permanent identifier: Revision 2.7 is deposited at doi.org/10.5281/zenodo.22740331, where it supersedes Revisions 2.6 and 2.5 and Addendum 1. A draft numbered 2.6.1 circulated internally and was never operative. Nothing other than the deposited text is controlling. A rule that has been discussed, drafted, or agreed in principle is not yet a rule, and no row anywhere on this site may be adjudicated under one.
That matters because the pressure runs the other way. A rule is usually proposed because a specific origin has exposed a gap in the current text, which means the tempting order is to settle the origin first and write the rule afterwards to fit it. Doing that would make the rule an account of a decision already taken rather than the standard the decision was tested against.
So where a determination depends on a provision that is under discussion but not yet published here, the row does not take that determination. It stays unresolved, and the reason is recorded against it. Amendments are drafted from the controlling text, reviewed as text, published here with a new revision number and date, and only then applied to adjudications. Rows decided under an earlier revision are re-tested against the new one rather than inheriting their old state.
This is the same standard the coverage index applies to everyone else. An institution that announces a figure has not published an account of it, and a rule that has been settled in conversation has not been adopted.
Revision 3.x is being developed as a replication and research standard. It is non-controlling. Candidate rules and candidate country outcomes are isolated from the live generator and do not change a public state unless and until a new version is formally adopted after its validation gates pass.
Send the first-party publication or evidence route. Coffee Monitor captures it, tests it against the deposited rule book, and re-adjudicates the row. A new source does not move a state merely because it is retrievable: it must pass the applicable authority, scope, economic-level, currentness, recurrence, units, flow, reconstructibility and archive tests. Any public state change is logged in the corrections log with the decisive source and the rule book revision. This is still the fastest way to be right about coffee data in any origin, and it is offered in earnest.
The price actually received by the grower at the point where the coffee leaves the grower's control, net of any deduction taken at that point. Recorded with a date, a location, a form of coffee, a unit of weight, and the class of counterparty. It is not the price a buyer says was offered, and it is not the price at the next node up the chain.
The definition matters more than the sampling. A weak sample of a well-defined quantity can be improved. A large sample of an undefined quantity cannot.
The farm-gate price expressed as a proportion of the free-on-board export price for comparable coffee in the same crop year, adjusted for the conversion ratio between the form sold at the farm gate and the form exported.
No reproducible farmer-share series has been published for Ethiopian coffee, because the first input has no published series either. Percentages do circulate, and some are widely cited; what is missing is a figure whose definition and inputs a reader can retrieve and check.
The farm-gate price below which continued production stops being rational for a typical smallholder in a given region, given local labour and input costs. A threshold, not a target, and it will differ by region. It is the trigger the second window of the proposed fund would use. See Beyond the Coffee Record.
Why publish a method at all
Asking an institution to publish a number is easy to say. It carries weight only if you have published your own definitions first, tried to collect the number yourself, and stated honestly how far you got and where you failed. That is the standard this page holds itself to, and it is the standard it asks of others.
An independent collection effort is under way, using a standard instrument and a small number of cooperating growers and cooperatives. It will not produce a national series and is not intended to. It is intended to demonstrate that a national series is buildable, and to make the cost of not building one explicit.
Results will be published here with the sample size, the regions covered, the collection dates, and the confidence a sample of that size can honestly carry. If the collection produces nothing usable, that will be published too, with the reasons.
Growers, cooperatives, unions, and exporters willing to report a price: the instrument and the reporting terms are available on request. Individual respondents are not identified. Write to wondwossen.mezlekia@coffeemonitor.com.