Changelog
Everything new or changed in Gyronomics, from launch to today. Only changes that affect visitors.
Each entry describes the state as it was on its own date. Figures and claims in older entries may have been corrected by newer ones - for the current picture see the data releases page. Data releases
The first regular monthly measurement: gyro prices by city, September 2026
- 1,654 shops in 78 cities, from the three platforms, on 25 September.
- Median by city with its denominators, each platform's coverage and unread shops by name, on the Prices page.
- The measurement does not change the Index; the first monthly rate comes in October.
Read the full entry
We read the three platforms on 25 September, inside the 25th-28th window set by the protocol. There are 1,654 shops with a gyro pita in 78 cities, and for each city we publish the median, how many shops support it and from which platform. Eight shops on one platform were not read: seven have closed there, one refuses its catalogue. They are listed by name and not counted as shops without gyros. The rules were written the day before, ahead of the data: this month only the level is published, and the measurement does not change the Index or its 2026 figure. The first monthly rate comes in October, each shop against itself. Against the 8 September pilot, in the 14 cities with at least 20 shops in both, the median moved by more than EUR 0.05 in two: Heraklion from EUR 4.95 to 5.30 and Rhodes from EUR 4.40 to 4.50. This is a finding, not a rate, because the pilot did not keep the fields needed to compare the same product. Every shop and the table can be downloaded.
Original regional prices and regulated schedules from Heraklion and Messinia
- 1979–1982 price correction from four newly verified original pages.
- Separate Messinia 1980 calendar and review of the existing model source.
- Only relevant PDF pages are stored in the repo, with page mappings and separate fingerprints. Full originals remain local outside Git.
- New early-period evidence distinguishes pita, bread and pork skewers.
- A conditional annual Messinia regulated-price reconstruction on the 1979 page.
- Order, publication and commencement dates remain separate. The national Index is retained.
Read the full entry
Eight original Patris, Simaia and Tharros issues add regional price reports and regulations from 1976–1980. One named-shop report leaves the bread format unspecified and retains that limitation. The orders distinguish gyro in pita from gyro in bread, cooked gyro from raw pork skewers, and pita-bread weight from the whole wrap. Requirements are not presented as actual weighings. The Messinia order sequence supports a calendar reconstruction for 1979, explicitly conditional on no uncaptured amendments. The year page shows amounts, days, calculation and original sources, shared with its JSON. This is neither a new national price nor an automatic Index change. Relevant PDF pages are retained privately in a separate folder with distinct source and extracted-file byte counts and SHA-256; public references open library-provider files. A concrete follow-up targets the pita/bread identity in the existing November 1980 source. The new December order and recovered November scan add precise evidence for 1980. The Messinia regulated-price reconstruction now covers the full leap year, with separate commencement and pita prices. The existing Index source is shown in its actual model role with an unverified serving format. A separate folder retains exact originals and hashes. Numerical series values remain unchanged. Further research identified two numerical errors. The 29 November 1980 Rizospastis issue distinguishes bread amounts from pita, while the 8 February issue explicitly identifies gyro in pita with seasoning. The old bread-price anchor is replaced by a Messinia ceiling reconstruction with three periods covering every leap-year day. It is labelled a modelled regional proxy, with explicit continuity assumptions and the existing regulation uplift. The 26 June 1979 issue establishes a later rise and disproves the full-year freeze inferred from a May rule. Years 1979–1982 are recalculated; wages, composition and other years are retained. The 7 February forecast is recorded but excluded as a realised price. Affected year pages and the API provide before/after results, the calendar and original references. New dated sensitivity reports and an immutable dataset retain all previous results. The standalone reproducer now verifies the anchor source calendar too. Only the four relevant original page images were added, with dates, printed numbering and SHA-256.
Data updates
- Pita or sandwich, a protocol we were not following, and three whitepaper errors
- A fourth witness off the platforms: 5% cheaper, and yet it agrees
- Pita or bread: the word sandwich looked like a bug and was Thessaloniki
- A third platform, 1,288 shops, and the agreement held
- From September the price is measured every month, and the area map had left 92 readings out
- 2026 goes from 3 shops to 816, and its price did not move
- A second platform, 755 shops, and the two readings came out the same
- 2026 hung on one source, and the other reports were the same report
Read the full entry
Pita or sandwich, a protocol we were not following, and three whitepaper errors Three things in this update, and all three started from the same place: checking whether I actually do what I say. First, item names. The harvester was keeping the price but not the name of the item that gave it, and that had already cost me: to establish that one 8.90 reading was a folia spitaki and not a pita wrap, I had to re-read 406 catalogues. So I re-read all 1,287 shops, across all three channels. 1,113 now carry a name. And as soon as the names were in, something turned up that I was not looking for. In twelve cities the wrap is not called a pita, it is called a sandwich. Serres and Giannitsa at 100%, Thessaloniki 93, Veria 91, Katerini 83, and below them Halkidiki, Kastoria, Larisa, Florina, Alexandroupoli, Kavala, Xanthi. At the other end, in thirty cities with four shops or more not one shop uses it, not once. Athens sits at one per cent, three shops out of 445. You do not need a map to see what the twelve have in common: they are all in the north. The question that matters is not the word but the product, and price answers it. Where the word dominates and both names appear, the median ratio between them is essentially one. In Katerini it is four euros for both, exactly. In the south the opposite happens: where the word is rare, the few sandwiches really are a different and dearer product. Athens 6.80 against 4.20, Lefkada 7.00 against 4.80. I wrote an article about it. At the same time I corrected a figure of my own: I had written that removing the word would take Thessaloniki from 1.19 to 2.12. That came from a single platform. On the whole sample the factor goes to 1.24, but the city is left with seven shops out of ninety-eight and nationally nineteen per cent of the sample disappears. Second, the monthly reading. The first regular one falls on 25th-28th September, so I got ready what that day will need. The first thing I found was an inconsistency of our own. The protocol says a 25th-28th window. September's snapshot was read on the 7th, because the window was defined later. And since the filename gives only the month and not the day, the inconsistency was entirely invisible. The reading of the 7th stays and is declared a pilot, the first regular one happens inside the window. A register of readings went in carrying each one's date, and the test does not trust the declaration: it works it out from the date and fires if anyone writes inside for a reading taken on the 7th. I also wrote the engine that computes the monthly rate, now, before a second month exists. A tool you write at the moment you need it has never been tested. I exercised it with a synthetic October over real data, where 143 shops raise prices by 20 cents while the composition also changes. The level moved 2.38 per cent while the real increase was half a point, and the tool printed a warning. The dry run also showed something I had not considered: the median of the ratios read zero, while 143 shops had genuinely raised their price. As long as fewer than half of them move, the median stays pinned at zero, and that is precisely the normal state for a pita gyros. I would have published zero change for months on end. So two rates get published: the median, robust against outliers, and the geometric mean, which catches diffuse small moves. Third, I read chapters nine to fourteen of the whitepaper with the code open beside them, and found three claims that do not hold. The chapter on user reports said they capture geographic premiums, giving the islands as an example, and delivery premiums. Neither: the form has no field at all for whether the price is over the counter or for delivery, and the regions are three, Athens, Thessaloniki, rest of Greece. The islands fall into the third. The limitations chapter said five points on the meat cost share move the 1976 index by about seven per cent, while the sensitivity chapter three screens above said under three. The right answer is 2.2. Both now read the same computed value. And the reproducibility chapter promises the index comes out of the public data.json with three quantities, but said net monthly wage while the dataset carries two wages. With one the index comes out in fifty-one years of fifty-one, with the other it fails in forty-six. So anyone sitting down to check our work had even odds of concluding the numbers do not add up. The chapter now names the fields, and a test went in that takes the public dataset and recomputes the index for every year, the way an outsider would. I verified what does hold, which was most of it: all five public endpoints answer, the licence matches the one the dataset declares, and visitor reports really do not affect the calculation. Last, I looked at the scenarios page and found a trap I had built myself. The two pledges end in different years, one in 2028 and the other in 2027, and the page explained that correctly. Except that anyone pressing the two buttons in turn saw 202 and 209, and concluded the second pledge gives seven more wraps. It does not. In the same year the gap is three wraps; the other four are simply the one year of price rises separating the two horizons. So the comparison in the last year both of them speak for now appears, with a test requiring the naive gap to be genuinely larger: if it ever stops being, the warning is redundant and should go. I carried on to the limitations page, which is where an inaccurate claim costs most, because that is the page where we are supposed to say what we do NOT know. Two of the ten entries claimed less than we know. The first said documented prices exist for a handful of years, then listed them: twenty-two of fifty-one. We were understating our own documentation, which is the opposite of the usual error and just as inaccurate. The same entry also said the years in between are derived from official inflation. They are derived from the catering index, a different measure, and our own pages explain elsewhere why the distinction matters. And it ignored that the crisis decade is not interpolated at all, but follows an asymmetric pass-through rule. The second said that at a meat share of sixty or eighty per cent the older index values shift noticeably, when the number is known: the 1976 index goes from 274 to 286 or to 259, about five per cent. On a limitations page, an adjective where a measurement exists is a step backwards. The correction tripped one of our own tests, and it was right to: it read every four-digit number in the paragraph and demanded it be a documented year, so the new and correct reference to the crisis decade set it off. I did not loosen it. It now scans only the list inside the parentheses, leaving the rest of the paragraph free to mention years without being taken to claim them. I also looked at the gaps page, and it held: every year on it is derived from the confidence grades, the ranges form themselves, and an apparent mismatch in the tax gap resolves correctly, since the 1989 law applies retroactively from 1 January 1988. Finally, the map the naming article was missing went in. Fifty-one cities in their real positions, sized by sample, filled by how often the wrap is called a sandwich. The claim was geographic, so it had to be checked geographically, and that is exactly what the map did: it corrected my own text. I had written that the twelve cities are all in the north. True. The converse is not, and the map shows it at once: Ioannina, Igoumenitsa, Arta and Corfu are just as northern and sit at zero. The border is not the north, it is Macedonia, Thrace and Thessaly. It stops at the Pindus range rather than spreading down from the frontier, and the text now says so. I did not write the coordinates from memory, and just as well. I pulled them from Wikidata and five were wrong. English-name matching put Veria in the Peloponnese and Xanthi in the Aegean, because homonymous villages of twenty-six and thirty people exist, while Chania, Sparti and Argos landed somewhere else entirely. Had I not checked them one by one, the map would have put two of the twelve northern cities in the south and destroyed the very finding it is supposed to show. They were re-fetched by Greek name with a population criterion, and a test went in requiring every city where the word dominates to sit above thirty-nine degrees. I bit it with the real error and it fired. Last, the prices page, where I found the page contradicting itself. The reports table says explicitly in its caption that prices are medians, not averages. But the first sentence a visitor reads said browse averages by city and year, and so did the home page and the description that shows up in search. Six places in all, across both languages. This is not a matter of style. Median and average are different measures, and on a sample with outliers they differ noticeably. The visitor sample is exactly that kind, with reports running from one and a half euros to six eighty for the same year. The page promised one thing and showed another. A test went in that scans all four surfaces, and that writes the difference between the two measures as a number rather than an argument. The rest of the page held up, and well: the three-report threshold is the same in the table and the city map, below it only the count is published rather than a price, and the blend simulator says in capitals that it is a simulation and that in practice reports never enter the published series. Then I asked the data something we had never asked it, and the answer was better than I expected. This project's neatest argument is that Thessaloniki is not expensive, its wrap is simply bigger. The wrap-to-portion ratio inside one menu shows it, stripping out rents, wages and the city's price level. But it was also the weakest evidenced: eighteen pairs, collected one by one from photographed menus across two decades. One of the three platforms also gives a portion price, in the same shop on the same day. The test was already sitting in our own data, four hundred and seventy-two times, and nobody had asked it. The result confirms and widens the gap: Athens 0.381 on 137 shops, Thessaloniki 0.527 on 50. A ratio of 1.38, where the menus gave 1.24. And the best part: the 2014 test replicates. The page cited as its cleanest case two menus from that year, Thessaloniki against Ioannina, with the portion almost identical and the wrap twenty-nine per cent dearer. On sixty-five shops in those same two cities today: portions two per cent apart, wrap nineteen per cent dearer. Twelve years later, by a completely different method, the same shape. Divide the two and the premium turns around. The Thessaloniki wrap costs nineteen per cent more and holds thirty-eight per cent more of a portion. Per unit of content it comes out fourteen per cent cheaper. That assumes the portion is a comparable product in both cities, and I write it down beside it: the within-shop ratio does not need that, the between-city comparison does. It enters no observation. Something I was not looking for turned up too. Across seventeen cities, the correlation between the price of a portion and the wrap's share of it is minus 0.81. The islands sit at a 12.10 portion and a 0.384 share; the northern mainland at 8.68 and 0.469. The reading is that the wrap's price is anchored near a national norm while the portion follows local costs. Where costs rise the wrap does not get dearer: it gets smaller. It fits something else in the same data: within Athens, the spread of the wrap price between the twenty-fifth and seventy-fifth percentiles is only seven per cent. And I wrote it up, because the finding deserves an audience beyond people who read indices: Thessaloniki is not expensive, it is generous. It carries a scatter plot of the seventeen cities, portion price across and the share the wrap holds up. The cloud falls from left to right, so nobody has to believe the correlation: they can see it. I did not draw a trend line, deliberately. With seventeen points it would lend the eye a precision the sample does not have, and the bare number underneath is more honest. Writing it tripped three of our own tests, and all three were right. One forbids toFixed inside articles, because it prints 0.381 with a full stop inside Greek text. The other two saw the portion prices, from 7.55 to 14.50, as invented wrap amounts. I did not exempt the article: I taught the checker that a portion is a different product with its own table. Anything in neither table still fires. Last, I went to close a small mismatch and found something bigger. The harvest README said six per cent of items were on offer. I was measuring three and a half in the snapshot, so I went to see which held. Neither, and worse: the six per cent reproduced from nowhere. It had come from the platforms' raw files, which the repo does not retain. The snapshot holds one row per shop, not per item. In a README that insists on reproducibility, there was a figure with nowhere to come from. I re-measured it on seventy shops per platform, and what came out was worse than a wrong percentage. One platform gives 4.3% across 10,704 items, with a median discount of twenty-four per cent. The other gives zero across 9,761. And it is not that it runs no offers: the original-price field exists, we read it on every item, and it never differs from the current one. It is a copy. We were keeping it believing it protected us from offers, and the text told readers so too. In practice it means the protection exists on one platform out of three. If a shop on the other two runs an offer, we record the discounted price as the regular one, and the monthly rate sees a fake drop then a fake rise. I wrote it into the README, into the rate engine where it will bite, and into four tests. The first test I wrote was wrong, and showed it itself: it demanded the old phrase appear nowhere, and fired at once, because the README quotes it in order to explain what it used to say and why it was off. The quotation is exactly the point. The test now looks at the conclusion rather than the string. None of this changed a published figure: the price stays at 4.25 and the index at 218. A fourth witness off the platforms: 5% cheaper, and yet it agrees The three platforms I read agree with one another, and until now I treated that as strong evidence. There is a hole in the argument, though: they resemble each other closely. Same model, same commissions, largely the same shops. If the price level were an artefact of that way of selling, three agreeing platforms would never show it. I needed a different kind of witness, and found one in a small independent guide to grill houses that runs its own ordering system. I set a rule before looking at any numbers: keep only shops that accept live orders, not those with a plain listing. The reason is the same one that rejected an archived Attica panel earlier. A page that takes orders at a wrong price costs the owner money, so it gets updated. A plain listing can sit there for years. And plain listings are the overwhelming majority: of the guide's 2,153 shops, only 142 carry a menu. Of those, 118 give a price for a pork gyros in pita, and 115 are in Attica. Their median comes out at four euros, five per cent below the platforms' Athens median. The easy conclusion here would be that platforms mark the wrap up. They do not, and I checked two ways before writing it down. First by area: across eleven Attica areas where both are present, the median ratio is 0.988, a gap of about one and a half per cent rather than five. Second by shop, which is the cleaner test: across six shops present on both channels the ratio is exactly one, with half giving the identical price to the cent. The explanation appeared as soon as I looked at which areas these were: Perama, Drapetsona, Kaminia, Porto Rafti. The guide leans heavily to outer Attica, not the centre. The gap was geography, not channel. Along the way I had to fix my own matching. The first version produced fourteen same-shop pairs, which sounds better, but GYROVOLIES is a chain with four branches and prices from 3.50 to 5.00. I was comparing one neighbourhood with another and calling it the same shop. Requiring the area to agree leaves six real pairs. Fewer, and right. Those 115 shops do not enter the index, and I want it clear that this is a decision rather than an omission. Adding them to Athens would pull down the median of the city that is the base of every area factor, and would move every other city in the country for a reason that is sample composition rather than price. The temptation to merge all the data because it exists is exactly what must not happen without weighting, so I wrote it into a test as well, so it does not happen later. The guide did confirm something else found shortly before: that the word sandwich means different things in different cities. Being almost entirely Attica, only three of its 118 shops price from a sandwich item, and those sit at 6.70 to 6.80, which is gyros in bread. In Thessaloniki the same word gives five euros and is the ordinary wrap. None of this changed the published figures. Pita or bread: the word sandwich looked like a bug and was Thessaloniki I started by asking whether the filter I use to pick shops biases the price, because it is exactly the kind of assumption that stays invisible until it is measured. One platform lists 16,418 shops and the harvester keeps the 620 whose name says so, the ones called gyradiko or souvlatzidiko. I read 2,732 random shops from the rest. Four per cent of them sell an explicit pork gyros in pita, meaning the filter does not throw away gyros shops, it throws away pizzerias and cafes. And their median comes out at 4.20, exactly the same as the sample I use. Across twelve cities with enough shops in both groups, the median difference is zero. In Athens 4.20 against 4.20, on thirty-six shops. Two things had to come out of the sample: supermarkets, which also carry a pita gyros at 4.90, and one test shop. They separate cleanly, because a gyros shop has sixty to three hundred items and a supermarket over ten thousand. In the middle of that I ran into something that looked like an obvious bug and turned out to be the most interesting finding of the day. The list of words meaning wrap contains sandwich alongside pita. Gyros in bread is a different, dearer product, so my first instinct was to remove it. I measured before removing, and it was just as well. In Thessaloniki ninety-one of the ninety-eight shops with a named item price from a sandwich, because that is what the wrap is called there. Had I removed the word, Thessaloniki would have dropped from ninety-eight shops to seven, its factor from 1.19 to 1.24, and nationally 209 of 1,113 shops would have lost their price, nineteen per cent of the sample. I would have manufactured a whole finding out of a local name. The word stayed, with the reason written into the code and pinned by a test, so nobody tidies it away later. There was a real trap, just elsewhere: pita also matches inside spitaki. An 8.90 item reading FOLIA SPITAKI GYROS CHOIRINO was passing as a pita wrap, same family as the gyromprizolaki caught earlier. Measured impact: one shop in 1,250, no city median changed. The row came out and the reading goes from 1,288 to 1,287 shops. Finding out that those 8.90 were a spitaki took re-reading 406 catalogues, because I was keeping the price but not the item name. Now the name is kept too. The same pass turned up three errors in our own text. The first is the worst: our own source entry said 813 shops, 59 cities for three releases, while the reading had gone to 933 and then past 1,200. The scanner that catches stale numbers deliberately skips the source files, and rightly so: an August report says what it said in August. For our own reading, remeasured every month, that exemption worked backwards and kept the old number alive. A targeted test went in, in both languages, tied to the measurement file. The second is in the whitepaper. It said each year gets a confidence grade of A, B or C. There are four grades: seven years of the 1980s carry D, and their badge shows normally in the year table. So the most methodological page was not defining a badge the reader could see two clicks away. The census is now computed from the series and says how many years each grade holds. The third came out of the second: the D label read projection outside the documented range. Grade D covers two different things, interpolations inside the price-control era and projections outside the documented ends. All seven years carrying it are the first, not the second. The reader was being told projection about years that project nothing. The label now says both cases. None of this moved the published figures: Athens 1.00, Thessaloniki 1.19, Patras 0.95, Chania 1.13, middle city 4.20, price 4.25 and index 218. A third platform, 1,288 shops, and the agreement held A third platform joined the measurement, and the logic is simple: two channels that agree is good evidence, three closes the question of whether the platform is doing it. The third was found the same way as the second, by looking at what the shop page itself calls when it loads its menu. It also has a different geography from the other two, far heavier in Patras and Heraklion, so it adds more than volume. September's reading goes from 933 to 1,288 shops, and the city table to 52 cities with 1,229 shops. Most cities are now measured on three independent channels. The agreement held. In thirty-four cities there are at least three shops on at least two platforms, fourteen come out identical to the cent, and the median spread between them is ten cents. Ioannina gives 4.20 on all three channels, Katerini 4.00 on all three. The big figures did not move at all, which is the point: Athens 1.00, now on 496 shops, Thessaloniki 1.19 on 119, Patras 0.95 on 30, middle city 4.20, the 2026 price 4.25 and the index 218. Two things shifted, both in the right direction. Chania goes from 1.143 to 1.131, with its sample rising from seventeen to twenty shops. And Heraklion, which the first two channels put at 4.65 and 5.50, settles at 5.00 on seventeen shops. So that disagreement was a small sample rather than a real difference, as I suspected but could not show. The area map grew by sixty-one new names, mostly Thessaloniki suburbs such as Kalamaria, Evosmos, Pylaia and Stavroupoli, and Attica neighbourhoods in a third transliteration. Two are left out on purpose: rest-of-greece, a catch-all that matches no city, and Analipsi, which exists both in Heraklion and in Volos. I preferred to lose nine readings than to put them in the wrong city. In the same pass a contradiction in the whitepaper was fixed, which is the most methodological page and therefore the most awkward place to have one. It said the peak is 511 in 1983 and then in brackets that the highest documented is 511 in 1983, the same number twice. And further down that 1983 is interpolated, so measured between documented years the drop runs from 1983. The text was written for a time when the two years differed. Now, when they coincide, it says it once and plainly: 1983 has no price of its own, it is interpolated, and there is no higher year with a direct price to put in its place. From September the price is measured every month, and the area map had left 92 readings out From today the gyro price will be measured every month rather than once a year. The capability is only a day old: harvesting the public listings of two ordering platforms yields hundreds of shops in a single day, automatically. Before that there was no way to do this short of walking into a thousand shops. It cannot be done retrospectively, and I checked before saying so. There is not a single archived platform menu page in the Wayback Machine. The one archived directory that carried text menus produced an Attica panel from 2011 to today, and I rejected it: forty per cent of consecutive pairs showed exactly the same price, with some years reaching ninety-two. You cannot tell a price that did not change from a page that was not updated, so the chain understates. Monthly measurement does not have that problem, and that is the main reason it is worth doing: when we read the page today, we know the price holds today. So a no-change is real data. It even becomes a useful figure of its own: how many shops change price each month. Two decisions were written down plainly, because they are the first to be forgotten. The monthly rate will be chained within the same shop, keeping those present in both months, rather than taken from the median of whoever we catch. Shops join and leave platforms, and a change of composition would look like a change of price when it is not. And the year will close as the mean of twelve months, provisional during the year and final in January, because a single month, even December, would decide the whole year. 2026 does not change: it only has September onwards. The tools and the raw data went into the repo, side by side. A number in an observation without the data behind it cannot be checked by anyone, including us in six months. There is a test tying the two: the observation's count must equal the number of rows in the file. Along the way a serious fault of mine turned up. The map matching platform areas to cities was silently leaving ninety-two readings out. Nearly all of them were Attica written in a different transliteration, Acharnes, Pagrati, Vyronas, Glyfada, Palaio Faliro, along with thirteen cities missing entirely. It is fixed, and the reading goes from 813 to 933 shops and from 72 to 79 cities, with none now unmapped. The best part is that not one factor moved: Athens 1.00, Thessaloniki 1.19, Chania 1.14, Patras 0.95, middle city 4.20, the 2026 price 4.25 and the index 218. More data, the same conclusions. One thing did get worse and that is written down too: the median gap between the two platforms went from zero to seven cents. The reason is that three more cities entered the comparison, among them tourist ones, where the two channels genuinely disagree. The cities identical to the cent are still thirteen. 2026 goes from 3 shops to 816, and its price did not move Yesterday this log said 2026 rests on three shops, while we hold 813 readings of our own that were going unused. The reason was specific: those readings come from ordering platforms, and we did not know whether a platform price is a counter price or something dearer. It has now been measured, two independent ways, and the number is 1.02. The first way is same-shop pairs. The Thessaloniki grill house O Giannis has a printed August 2026 price list reading all sandwiches, with or without spread, only four ninety, and the same shop appears on the two platforms at five euros and five ten. Two more shops with their own websites at four ninety also show five and five ten. The second way is same-city medians. Four printed 2026 Thessaloniki price lists were read and give a median of four ninety-five, while the two platforms across eighty-nine shops give five euros. The two ways give 3.1 per cent and 1 per cent. We take 2 per cent, between them, and the spread passes into the observation's uncertainty rather than being hidden. The price lists read are photographs of real menus: Giannis at four ninety, Stelios at five, Gyradiko at five, Express at four twenty, all in Thessaloniki, and Dokos in Euboea at three euros from a wall price list. A sixth was rejected because it is not a counter price: the Prassas leaflet is a delivery menu with a six euro minimum order. So our reading entered 2026 and the year goes from three shops to 816. The most important part, though, is what did NOT change: the year's price stays at four twenty-five and the index at 218. I write that with emphasis deliberately. When you add 800 readings to a year that had three, the normal outcome is that the number moves. It did not, because our reading agrees with the two journalistic sources: four twenty as a median per shop, four twelve after the channel adjustment, four twenty-eight if weighted by population. The year gained evidence rather than a new number, which is the best possible outcome. Two things are stated plainly, because 816 is easy to misread. First, the reading enters as ONE observation with its own weight, not as 813 independent sources. Second, its weight is not proportional to the count: it is a convenience sample of shops that happen to be listed on a platform, and those lean urban and newer. It carries a weight slightly above the two other sources, not ten times theirs, and a test holds it there. Finally, one test was reversed on purpose. Until yesterday it guarded that platform readings are not national evidence, which was right while the factor was unknown. It now guards the next thing that could go wrong, namely counting them as 800 sources. The reversal is written inside the test so it does not look silent. A second platform, 755 shops, and the two readings came out the same Yesterday a table of thirty-one cities went onto the regional page, measured from the listings of one ordering platform. The obvious question was whether the platform makes the geography by itself: each one has its own mix of shops in each city, and if it happens to carry the expensive ones somewhere and the cheap ones elsewhere, a difference appears that does not exist on the street. The only way to answer that is a second, independent reading. The second platform does not carry the menu inside the page; it loads it separately, so I had to find where from. Together with fixing a flaw in how shops were chosen, the table goes from thirty-one cities and 191 shops to forty-five cities and 755 shops, with Athens now resting on 237 and Thessaloniki on 89. The cross-check is the main finding, and it turned out better than I expected. In twenty-five cities there were at least three shops on both platforms. The median absolute difference between the two readings is zero, and thirteen of the twenty-five come out identical to the cent. Thessaloniki gives 5.00 on both, Chania 4.80 on both, Ioannina 4.20, Patras 4.00, Corfu 4.50, Kalamata 3.50. Two independent listings do not land on those numbers by chance, so the geography the page shows is real rather than an artefact of the method. The most useful part for me is that Chania was confirmed rather than changed. The factor I moved earlier, from 1.02 to 1.14, rested on five shops and it bothered me. It now rests on sixteen, from two channels, and produces exactly the same number. The country's middle city also stays at 4.20, so the index did not move at all: 4.25 for the price and 218 wraps. There is one area where the two channels disagree, and I write it beside the table rather than hide it: the tourist destinations. Mykonos comes out at 6.50 on one and 5.25 on the other, Rhodes at 4.80 and 4.45, Heraklion at 4.65 and 5.50. The explanation is simple: in a tourist town the shop on the square and the shop in the neighbourhood sit side by side with a huge gap between them, so a small sample gives a different answer each time. In ordinary cities that does not happen. With twice the data, the country's range grew at the top. Mykonos enters at 6.00 across nine shops, nearly one and a half times the middle city, and at the bottom sit Kalamata and Ptolemaida at 3.50. The very same thing costs almost twice as much between the two ends. Something I had not noticed in the first harvest was fixed too. I was choosing shops by keywords in their name, which caught the ones called gyros shops and missed the ones named after their owner. The list now comes from each city's souvlaki category page, that is from the platform's own classification. In Kefalonia the name filter gave four shops; the category gives nine. Finally, two smaller things. A fourth trap joins the recorded list: Neo Irakleio is in Attica, not Crete, and the first version of the analysis counted it on the island because the word matched, pulling Crete's median down. And the second platform gives both a current and a regular price, because six per cent of items were on offer. I keep the price the customer pays, and write down the percentage so the size of the issue is known. 2026 hung on one source, and the other reports were the same report The 2026 index said 215 wraps and from today it says 218. The interesting part is not the three wraps, it is why. The whole year rested on a single survey, the one broadcast in August. Looking for a second one turned up plenty of articles, but reading them one by one, they all reproduce exactly the same figures from the same survey. That is not a second opinion, it is one opinion under four headlines. A year hanging on a single reading has to declare wide uncertainty, and 2026 carried a plus or minus fifteen per cent band, the widest of the decade. The second source did exist; it just was not a survey. It is the president of the Attica grill owners, who on 2 May 2026 gives dated prices: a pork or chicken pita sells take-away between 3.70 and 4.20 euros, where before 28 February it was 3.50. The plain skewer went from 2.30 to 2.50. The take-away figure enters at a 3.95 midpoint, weighted below the broadcast survey because it covers only Attica and because these are averages a person states rather than menus somebody recorded. So the year's price moves from 4.30 to 4.25 euros and the index from 215 to 218. The more useful part is that the band narrows to plus or minus ten per cent, which means the 2026 index no longer ranges from 187 to 253 but from 198 to 242. The same source settled something that had been hanging for a while. The model carries a factor for how much dearer the same wrap is when you do not take it from the counter. It was eight per cent and it was our own number, justified as a typical difference. The same man, on the same day, for the same product, gives the second price too: by delivery, 4.00 to 4.50 euros. The ratio of the two midpoints comes to 7.6 per cent. The assumption was nearly right, but it has stopped being an assumption. And one detail changes what the factor means: the source says explicitly that it costs the same to sit and eat it in the shop. So the factor is not a platform commission; it is the gap between taking it in your hand and anything else, the table included. Two stale spots turned up along the way. The box explaining the last revision had a heading from one revision and figures from another, claiming the index went from 225 to 215 while showing 4.81 and 192 underneath. And the example on the limitations page had 4.30 euros typed into it by hand. Both are fixed, and the second is now computed from the series so it cannot go stale again. Finally, the same article forecasts a five to six per cent rise during May or June. It is used nowhere, because it is a forecast and not a measurement.
- Datav2026.09-odigoi
Data updates
- 1984 turned up in an expense ledger, and 1979 turned out not to have risen at all
- A 1983 price turned up, and 2002 turned out never to have been measured
Read the full entry
1984 turned up in an expense ledger, and 1979 turned out not to have risen at all The best find this time was neither a newspaper nor an order, it was a diary. A man called Louis Grivetti cycled across Greece in the summer of 1984 and wrote down what he spent every day, in drachmas and in dollars. In the list, between the newspaper and lunch, sits the line: souvlaki sandwich, thirty-five drachmas, thirty-two cents. Not an estimate, not a ceiling, a price actually paid. And the nice part is that the line carries its own check. Thirty-five drachmas at thirty-two cents means one hundred and nine drachmas to the dollar, which is exactly the 1984 rate. We needed that: the same line appeared in the index also attributed to a travel guide from 1979, where the rate was thirty-seven and the figure simply impossible. The arithmetic showed which attribution was right, without having to trust the catalogue. A second source says the same: the Let's Go guide of the 1985 edition writes that souvlaki, on a skewer or wrapped in pita, is a bargain at thirty to thirty-five drachmas. We did not make 1985 an anchor, and it is worth saying why: guides report a survey from the previous year, so the 1985 edition speaks about 1984 prices. Entering it as a 1985 price would have manufactured a decline that never happened. With 1984 documented, the long gap has been cut a third time. Two four-year runs remain, 1979 to 1982 and 1985 to 1988. Two releases ago it was eleven consecutive years with no source at all. The second find is of a different kind and perhaps matters more. Searching not for the word souvlaki but for the price orders themselves turned up the front page of Makedonia of 30 December 1979. It says the selling price for souvlaki made from pork will rise by thirteen percent, and that the relevant order is expected. Put it beside the order of May that same year, which raised soups and pasta by forty percent while stating that souvlaki prices remain unchanged, and the year is documented end to end. Within 1979 the price of souvlaki did not move, and the first increase falls at the turn into 1980. Until now we gave 1979 headline inflation, nineteen percent. That was not a bad assumption, but it was an assumption, and two pieces of paper undo it. The year's weight was set to zero. The 1979 index rises from six hundred and one to six hundred and ninety-four wraps, the largest figure this update moved. The third part concerns 2000. Four travel guides, naming shops and streets, give a wrap at three hundred and fifty to four hundred and fifty drachmas in central Athens. Thanasis in Monastiraki at three hundred and fifty, Pitta Pan at four hundred and twenty-five. These are central prices, so they are rescaled to the national median exactly as we do for the Athens sources of recent years. 2000 has stopped being an interpolation. The same guides also gave something that had been missing for a long time: a third documented point for the skewer-to-wrap ratio. We had two, from 1976 and 1978, and then a pinned figure for 1998 that was pure assumption. The 2000 measurement contradicted it, so the assumption went. And with it went a standing disagreement: the undated 1988 clipping gave a ratio of seventy-eight percent while our model gave fifty-eight. The model now gives seventy-four, and the clipping was not used at all to get there. We also searched two official routes that gave nothing, and we write that down so nobody searches them again. The Hellenic Parliament digital repository has character recognition and twenty-two items containing the word souvlaki, in sittings from 1975 to 1995. We pulled the text of eleven sittings. No price, all rhetorical. And in the government gazette, full-text search does not reach before 1992, so the price-control orders of 1984 to 1988 cannot be found by word. Finally, something about the checks. Seventeen new ones were written, and three of them did not bite. We found that by deliberately breaking the code to see whether they would fire. Their shared mistake was looking at the output instead of the source: if someone typed the thirty-five by hand instead of reading it from the constant, the result came out the same and the check passed. They were fixed, and ten of ten deliberate defects now fire. A 1983 price turned up, and 2002 turned out never to have been measured Three documents, all three from newspapers of the day. The first is Rizospastis of Sunday 14 August 1983. In a short column about cigarettes that had suddenly gone up, it also says this: the kebab-type souvlaki sells for two drachmas more, that is thirty drachmas. The verb is sells, so it is a market price rather than a ceiling, and it is the first free price we have found inside the big gap. The same sentence gives the previous level too, twenty-eight drachmas, so we know both of the year's figures. The annual price is their weighted mean, because the rise came in mid-August and covers four and a half of the twelve months. The project's longest gap was cut in two and fell from ten years to five. And something happened that did not have to. The interpolation between 1977 and 1989 already gave a price for 1983, without ever having seen the report: twenty-eight point six. The document says twenty-eight point seven five. So the documentation did not correct the series, it confirmed it. The peak of the whole classic series, which is 1983, stopped resting on a calculation and now rests on paper. The second document is Makedonia of 18 May 1979, front page. Two price-control orders raise soups and pasta by forty percent, pulses by up to fifty, café items by up to twenty-five. And about our item they say the opposite: prices remain unchanged. We did not make it an anchor, because it gives no price level, only a change. It did go into the limitations, because it says something that needed saying: the official catering index measures a whole basket, and the basket and the item can part ways. The third is Rizospastis of 8 January 2002, with its own market survey. On souvlaki, it writes, increases of up to forty-six percent were observed, from three hundred and fifty drachmas to one and a half euro. The source's arithmetic checks out on its own. That forced us to look at what we had been saying about 2002, and it was not right. 2002 carried exactly the same price as 2001 and was declared documented, so the series was saying the euro changeover changed nothing. Except that one euro seventeen was not a 2002 measurement: it was the conversion of the dual-pricing campaign's three hundred and ninety-nine drachmas, a 2001 price written in euros. Now 2002 is an estimate with a range, from one euro seventeen to one and a half. The lower end means no change, the upper is the largest increase the paper recorded. Up to forbids taking the upper end as the central value, so the centre comes from 2001 on the official 2002 sector index and nothing else. The 2002 index falls from four hundred and ten to three hundred and eighty-seven wraps, and the years to 2010 follow with smaller steps. 2026 did not move. Along the way a citation turned up pointing at an article that does not exist: the identifier was 1234567, a sample rather than a real one. The actual report was located and a check now scans every source for such patterns. A fake identifier in a citation is worse than no citation, because it looks like documentation. Finally, how they were found. Last time meant reading scans by eye and stopping at the easy ones. This time six hundred and eighty-nine articles were collected and scored straight from the character recognition, without opening a single scan. The key was the negative weights: for sale next to grill house is a classified ad for selling a shop, and confiscated is a price-control raid, which never prints a price. The scoring was tested on known answers first: the two finds we had already confirmed came out first and third of six hundred and eighty-nine. Without the negative weights, one of the two fell to fifty-eighth.
- Datav2026.09-arxeio
Prices for 1976 and 1978 turned up, in the newspapers of the time
Read the full entry
We had been looking for prices in the years with no source at all. The route that worked was the National Library digital newspaper archive: free, no account, full-text search, and it holds Makedonia, Rizospastis, Eleftheria, Embros and Tachydromos. The word "souvlaki" returns two hundred and sixteen articles inside our gaps. The first two we opened both carried a price. In Makedonia of 24 October 1976, a ministry of commerce order sets the gyro wrap at seven and a half drachmas with pita, and the same text records that the price already formed in the shops was eight drachmas. In Rizospastis of 17 June 1978, a price-control order sets the doner wrap with pita at ten drachmas. The second one matters most: it is the first documented wrap price inside the eleven consecutive years that had no source. The project's largest gap falls to ten, and documented years go from sixteen to eighteen. The same documents also corrected two constants that had been assumptions. The first is the factor that turns a regulated ceiling into a market price. We had measured it once, from a 1989 report, at the moment price control was lifted and prices typically overshoot. The 1976 text measures it a second time, inside the price-control era and in Athens, and comes out lower. We take the mean of the two. And here something unexpected happened: with the new factor, the 1977 Chania order carried forward to 1989 on headline inflation alone lands half a percent from the documented price, where before it missed by three and a half. We tuned nothing, it came from an independent document. The second constant is the skewer-to-wrap ratio. It was pinned at sixty percent with no source. Both documents print the two prices side by side and say ninety-three and eighty-five. That makes sense: back then the wrap was nearly the same thing plus the pita, while today it carries fries, tzatziki and three times the portion. A third document turned up, from 1983, which gives no price but two rules: that the 37 to 40 gram portion held in the middle of the period as well, and that the price of souvlaki was tied to fresh pork. The figures: prices from 1976 to 1987 fall by one to four percent, so the index for those years rises a little. The peak stays in 1983 and goes from 500 to 511 classic wraps. 2026 stays at 215. The remaining two hundred and thirteen articles were not mined. Reading is scan by scan, by eye, and 1977, 1979, 1980 and 1981 hold dozens of hits each.
The index we were using was also measuring hotels
Read the full entry
To fill the years that have no documented price of their own, the model uses the official catering price index. The index we had been using is called "Restaurants and hotels", and the second half of that name is the problem: it also measures hotel rooms, which swing with tourism and have nothing to do with a souvlaki shop. In 2022 that index showed a rise of 8.7%, while the sub-index covering only restaurants and cafes showed 4.7%. Half of what we were counting as food inflation was overnight stays. The index has been switched to the narrower one, the one that literally covers the shops where you buy a gyro. The change moves ten years, seven of them by a cent or less. The three that matter are 2018, 2019 and 2022, which become five cents cheaper. 2026 does not change at all: 4.30 euro and 215 gyros, as before. There is an uncomfortable thing to say here: the correction makes the site's case stronger. The hypothetical scenario "if the gyro had followed official catering inflation" now shows a wider gap than before. When a change suits you, you owe it to the reader to show it: the live Eurostat check now also downloads the old index and permanently states in how many years it differs and by how much, so anyone can judge for themselves. The second correction is in the text. The methodology page said the net wage is derived by deducting social contributions and income tax. Income tax is not deducted, and that is a deliberate choice explained in four other places on the site. So the page that exists to explain the method was contradicting the method. An outdated justification fell with it. The site said in six places that the minimum wage sits below the tax-free threshold, so deducting tax would be pointless. That held for 47 of the 51 years. Since 2023 it does not: today the minimum wage pays about 342 euro of tax a year, and deducting it drops the index from 215 to 208. The site now says so with numbers instead of skipping past it. Finally, the sensitivity table gained a column with each year's grams of gyro. It was showing "equivalence 114%" without saying that means 106 grams against 88, which became unreadable once visitors could change the reference year themselves. All three findings came from someone else's comment on LinkedIn. All three were right.
- Datav2026.09-sources
A 2019 price was counting the same vague claim twice
Read the full entry
The series carried a 2019 observation of "€2.00 at a souvlaki shop in Pangrati", sourced to a report that tracks that same shop year by year. We went back to the article: it gives no price for 2019. Its series starts in February 2020 at €2.50, and the "two euros" appears only as a loose phrase, "around 2 euros", for 2019-2020. That same loose phrase was already counted separately, correctly flagged as a rounded retrospective figure with a range. So one statement was entering twice, once disguised as a measurement of a specific shop - and it conflicted with a contemporaneous May 2019 report putting that same Pangrati shop at €3.00. The false measurement was removed. The 2019 price moves from €2.30 to €2.35 and that year's index from 308 to 302 classic wraps. No other year changes. Second: the 1977 Chania price order, on which the whole first decade rests, was checked against the original again. It has three price lines - meatball, gyro, pork skewer - and the figure we use is indeed the gyro line, not the cheapest item on the list. All three are now recorded. Third: two more sites that republish our index were added to the exclusion list, because a search for historical prices now returns our own figures - frozen at an old release, at that.
- Datav2026.08-attica
Data updates
- Counter prices for 2025 - now the best-documented year in the series
- 2019 gains a nationwide source - and what we did not find for 2003-2010
- The 1989 law itself was found - and it corrects something we had written
- One survey was filed under the wrong year - now fixed
Read the full entry
Counter prices for 2025 - now the best-documented year in the series A gap we had not noticed: every source we had for 2025 was either a delivery price or a single shop. None was a counter price from multiple points - which is exactly what the index measures. A May 2025 report was found with prices across eight areas of Attica: Dafni 3.80, Neos Kosmos 3.90-4.00, Ilioupoli 4.20-4.40, Gyzi 3.70-3.90, Kolonaki 3.90-4.40, Kifisia 4.10-4.30, Vouliagmeni 4.00-4.20, Piraeus 3.70-3.90 euros. The 2025 price moves from 4.22 to 4.16 euros and the year now rests on four sources with a confidence score of 98 out of 100 - the highest in the entire series from 1976 to today. The 2026 index is unchanged at 215. The report states no methodology and no number of shops per area, so it entered with a reduced weight - we note that because it matters. 2019 gains a nationwide source - and what we did not find for 2003-2010 Until now every source we had for 2019 was either from Athens or retrospective. An April 2019 report was found quoting regional catering trade associations across six areas outside Athens: Thessaloniki 2.40-2.80, Patras 2.30-2.50, Chania 2.30-2.50, Kozani 2.70-2.80, Volos 2.80-3.00, Corfu 2.40-2.80 euros. The 2019 price does not change - it stays at 2.30 - but the year now rests on four sources and its confidence score reaches 92 out of 100. One detail is worth stating: the 3 euro figure quoted for Athens was PROPOSED by the association, which is exactly why the Competition Commission opened an investigation into concerted practice at the time. It is not a price observed in the market, so we left it out. We also record a negative finding, because that is information too: for the years 2003 to 2010 we searched the press around the VAT rises, the "ten years of the euro" comparisons, the ESYE bulletins and the INKA tables, and found no usable price. Those eight years remain an interpolation between the 2002 and 2011 anchors, and we say so openly. The 1989 law itself was found - and it corrects something we had written The primary text has been located: Market Order 14/1989, article 201 "Souvlakia" (Gazette B 343, in force 11 May 1989). It forces us to correct our own wording. We had written that 1989 was when "price control on souvlaki was lifted". It was not: article 201 KEEPS the regulation and merely replaces the fixed ceiling with a formula tied to cost - the price is derived from the meat purchase price divided by fixed factors. The 1989 newspaper report we had found concerned soft drinks, souvlaki with a BURGER PATTY, patsas and offal dishes; gyro appeared only as a benchmark. Regulation in fact continued afterwards, amended in 1992. The figures in the series do NOT change - the 70-80 GRD level for a 1989 wrap still stands. What changes is the explanation, and for the better. The law brings three bonuses: it confirms the 37 g minimum weight at national level, sets a pita at no less than 65 g, and - most importantly - states plainly that its prices apply to Attica and Thessaloniki and that other regions may not exceed them. So the 1977 Chania order was by definition below the Athens level, and raising it is not our assumption: it is the direction the law dictates. One survey was filed under the wrong year - now fixed Pricefox runs its nationwide price survey every year with the same method across six cities. The edition of 25 October 2024 was recorded in our series as "December 2025", so 2025 rested on 2024 data and came out cheaper than it really was. The two editions are now separate and in their correct years, and the 4 December 2025 edition has been added as well. The gain is not only the correction: 2024 and 2025 each had a SINGLE source until now, one souvlaki shop in Pagrati, and now have two and three respectively - both move up a documentation grade. The prices: 2024 3.70 -> 3.71, 2025 4.01 -> 4.22, 2026 4.20 -> 4.30 EUR. The 2026 index goes from 220 to 215. Note that 2023 remains the only recent year with a single source: Pricefox ran no survey that year.
- Data
Data updates
- Two price-control orders rewrite the whole pre-2002 period
- Contemporaneous sources for 2018-2026 and two biases corrected
Read the full entry
Two price-control orders rewrite the whole pre-2002 period A 1989 newspaper report surfaced describing the moment the Ministry of Commerce lifted price control on souvlaki. It quotes both the regulated price (65 GRD) and the first free-market price (70-80 GRD), and explicitly extends that to "the skewer or gyro souvlaki". The resulting check is the strongest we have ever had for the period: the 1977 Chania order carried forward on the general CPI ALONE predicts the 1989 order to within 3.4%, with no free parameters. Consequences: the price-control era now runs to 1989, the regulation uplift becomes measured (1.154) rather than assumed, and the old 1985 anchor of 60 GRD is withdrawn - it rested on recollections and would imply an 8% rise over four years in which inflation ran 85%. 1998 moves from 350 to 280 GRD with a wide 200-360 range. NOTE ON THE NUMBERS: the index peak moves from 2009 to 1983 (439 to 500 classic wraps) and the 2026 index goes from 225 to 220. Contemporaneous sources for 2018-2026 and two biases corrected Added a contemporaneous May 2019 report on four Athens souvlaki shops quoting both the old and the new price, which yields observations for 2018 and 2019; a March 2022 shop-owner account; and a nationwide August 2026 survey, thanks to which 2026 stops being a projection and becomes a documented year. Two biases were also corrected. First, the "SKAI 2022 survey" was in fact a Pricefox survey republished by five outlets - one source, not five confirmations. Second, the Athens price premium is not constant: the 2025-2026 nationwide surveys place Athens at the median with Thessaloniki and Heraklion above it, so the factor now decays from 1.08 to 1.00. Finally, a bug was fixed where the PDF export could show an index up to 2 points different from the page.
- Data
Data updates
- Classic-equivalent rescaling
- Price-control era 1976-1985: explicit uplift and a wide band
- Reference basket, Pricefox as a second source and three new layers
- Price bias correction: from Pagrati to a national median
- The series now starts in 1976
Read the full entry
Classic-equivalent rescaling The index now states its goal explicitly: how many classic present-day wraps (~88 g gyro, fries inside, tzatziki, tomato, onion) each year's salary could buy. Since the 1977 wrap held 37-40 g and no fries, a continuous equivalence factor was added: 0.70 x (year grams / 88) + 0.30 x (0.65 + 0.35 x fries adoption). A 1976 wrap therefore counts as ~0.50 of a classic wrap and the era's index roughly halves. New Classic-equivalent tab, a table badge and extra export columns. Price-control era 1976-1985: explicit uplift and a wide band The first years of the series were not free market prices. The 3/1977 price control (8.50 GRD, Chania) was a ceiling on a 37-40 g portion, so it is now rescaled to an estimated market price with an explicit 1.20 regulation factor (10.20 GRD, range 8.50-11 GRD). Across 1976-1985 the uncertainty band widens to plus or minus 35%, interpolated years drop to confidence D (experimental), and the years are flagged with a price-control badge in the table and a shaded zone on the chart. For long-run comparison the recommended measure is the index per 100 g of gyro, which neutralises the smaller portion of that era. Reference basket, Pricefox as a second source and three new layers The reference basket is now explicit: pork gyro wrap, counter price, national median. Every price observation is normalised to that basket with stated factors (delivery 1.08, Athens 1.08, mixed wrap basket 1.10) and multiple sources for a year are combined with a weighted geometric mean. The nationwide Pricefox survey (6 cities, 2022-2025) joins as a second independent source, so the last five years no longer rest on one shop. 2025 settles at EUR 3.92. Also added: an uncertainty range on the index itself, the ex-VAT catering price, an index per 100 g of gyro (portion size) and a disposable-income-after-tax scenario with 14 salaries. Price bias correction: from Pagrati to a national median The 2019-2025 prices came from a single shop in Pagrati, an expensive central Athens neighbourhood. We now keep the source's rate of change but rescale the price level to a national median by dividing by 1.08 (the Athens versus rest-of-Greece gap seen in visitor reports and local reporting). 2025 moves from EUR 4.70 to EUR 4.35. Single-source years are also downgraded to confidence "B" and their uncertainty band widens to plus or minus 15%. The series now starts in 1976 A major data update. Minimum wages for 1976-1985 now come from the primary texts of the national collective agreements and arbitration awards of the period (6,121 GRD in 1977, 10,240 GRD in 1980, 31,932 GRD in 1985 instead of the old 34,390 GRD estimate). A documented 1977 gyro price was added (8.50 GRD, Chania prefecture price order 3/1977) and 1978-1984 are interpolated between the documented 1977 and 1985 anchors. For years up to 1992, employee contributions use the IKA mix of that era (12.22%).
- Data
Three price scenarios with real Eurostat data
Simplification to three scenarios (actual price, general CPI, catering CPI) using official Eurostat HICP catering data since 1999.
- Data
1985 data fix and consistency validation
New 1985 anchor (60 GRD per gyro) and automatic consistency validation with 11 data integrity checks.
Price and wage coverage: price collection of September 2026, catering CPI to August 2026, minimum wage in force since 1 April 2026. Dataset version 2026.10.10-monthly-bulletin, last updated 2026-10-10.