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 The strongest attack wins, revisions are dated, and superseded versions stay reachable.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: whether an outsider can reconstruct a given figure from public records, for a given crop year, as of a stated review date. It does not measure statistical quality, institutional capacity, honesty, or performance.
The instrument began as an Ethiopian one. It is now applied on the same terms to every coffee-producing origin through the Global Coffee Observatory, which runs this index across 51 origins. 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.
Not checked yet is a sixth marker but it is not a rung on this ladder. It flags a row whose state has not been established, either because the search is not complete or because it depends on a figure that is itself undocumented. It carries no position, high or low, and a row in this state never carries a 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 written publication, attributable to the institution responsible for the figure, that a third party can retrieve and cite. A press report of a speech is a record of the speech, not of the figure. A ministerial statement, a parliamentary transcript, a video, and a press briefing are all evidence that a claim was made, rather than an account of how the number was produced.
A row carries a numeric state only when two things are true: the row has been adjudicated, and the source document behind it has been captured and entered in the archive. An internal decision on its own is not enough. Coffee Monitor's own pages are never admissible as evidence for a Coffee Monitor rating.
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 headline figure is a count, not a composite score: how many of the six rows reach state five, published with a stated method. A count of zero is a factual statement under the definitions above rather than a grade.
Only when all six rows are settled. If any row is still marked not checked yet, the origin carries no count, and the audit is reported as in progress rather than as a score. A partial count would understate an origin by treating unfinished work as a finding of absence, which is the same conflation this index exists to prevent.
A row whose evidence chain breaks under verification is held, as is a row whose determination is made but whose document is not yet captured. In both cases the adjudication is preserved internally and the row displays without a rating. Researched-then-withheld is a different object from never-looked-at, and displaying one as the other would misreport the work done.
A derived row inherits the lower state of its inputs, and its evidence chain is whole only when every input chain is whole. Where any input row is held or unrated, the derived row is held 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 measured as US dollars free on board. Volume is measured in net kilogrammes, and shown additionally as 60-kilogramme bag equivalents by dividing by sixty. Implied export unit value is dollars per kilogramme, earnings divided by net kilogrammes; the bag conversion is not applied to it.
Heading 0901.90 is excluded outright, because it covers husks, skins, and substitutes containing coffee in any proportion, which are not coffee. Roasted coffee and soluble extracts are recorded as memo lines and do not score. Where a source offers more than one quantity, net weight is used, not gross weight and not a supplementary unit. Coffee re-exported from an origin that did not grow it is excluded from that origin's scored rows.
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.
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. Nothing else is. 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.
Send the publication. A qualifying source, captured and tested against the rules above, moves a row, and the change is logged in the corrections log with the source credited. Retrievability alone does not do it: a document has to pass scope, units and reconstructibility as well. 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.