Accuracy is the product. When we get something wrong, we fix it in the open, corrections are logged, dated, and published, never silently edited away.
19 October 2026, Myket, Filimo, Amnafzar, Eads and IDPay, read again at source: Myket’s public-company row also said it files on Codal under a symbol, was admitted to the Farabourse board on 12 August 2025 and had not yet traded; the gazette notice it cites prints only the conversion to a public joint stock company, so the row now says only that. Filimo’s parent row named Aparat Music among Saba Idea’s services; the Saba Idea site does not list it, and it was removed. Three rows on Amnafzar, Eads and IDPay had been written from search snippets, and their own pages returned a browser error on every direct read from Iran for three weeks, so they were withdrawn from public view.
5 October 2026, Iran AI Factory (AIIF), read again at source: The seed investment row said teams could receive up to 2 billion toman, structured as a convertible note with no valuation. AIIF’s autumn programme page now states up to 3 billion toman and does not print the convertible-note structure, so the row says only what the page says. The growth investment row and the programmes row were rewritten to match the current page in the same way.
2026-10-01, AzkiVam, SnappBimeh and SnappTrip, read again at source: Six rows said more than the page they cite. AzkiVam’s founding read 2020; Way2Pay’s company profile gives its registration as 10 March 2021. Its business model claimed a credit spread, software fees and merchant commissions, and its ownership row named MAX Holding as parent; the profile describes the services and a group relationship but neither how it earns nor who holds the shares, and both rows now say only that. SnappBimeh’s leadership row named Majid Hesami as chief executive and said Hossein Oliya had left for SnappPay; the linked profile is Oliya’s own, which still lists SnappBimeh. SnappTrip’s row named a current chief executive and a managing director from profiles that are not linked; the linked post names two leaders without titles. Tapsell’s Cafe Bazaar row keeps the purchase and drops a date, seller and price its linked report does not give. Two rows citing Instagram posts that do not open were withdrawn from public view.
2026-09-29, Digipay credit, Okala revenue and two Digikala counts, read again at source: An independent check found four figures whose quoted evidence did not hold them. Digikala’s 466,169 sellers and 10,510 staff for the year to March 2025 were right, but their passages quoted only the earlier report; both now quote the year’s annual report itself, which labels the 10,510 as the Digikala group’s headcount. Digipay’s monthly credit (about 20,000 billion toman issued and 7,000 billion used), carried on both the Digipay and Digikala records, rested on an English paraphrase of a spoken Persian interview, and Okala’s operating revenue (about 631 billion toman) on a mixed English and Persian transcription with no locator. All three rows are withdrawn from public view until the source’s own words are read. The Digikala dossier was regenerated.
2026-09-29, Digikala’s group companies and Digikala Jet, read again on the group’s own pages: The group’s English structure page lists eleven group companies, including Komoda; our record listed ten and left Komoda out on a claim that it had left the group, for which no source is on the record. Komoda is restored, and the unsourced exit and filtering-order statements are removed. The record also said the group holds shares in each company; the page says it owns all of the Digikala online store and all or part of seven named businesses, and the record now says only that. The group’s Persian page lists a different eleven, with Digikala Jet and without Komoda, and both lists are shown. Digikala Jet began operating in the Iranian year 1400 (2021-22) by the group’s Persian page, so its founding year is 2021, not 2022. A row that named Digikala Jet’s licence holder is withdrawn from public view: its passage concerns the Digikala online store and does not mention Digikala Jet.
2026-09-29, Miare, read again in the Pishgam venture fund’s filing: The fund’s cost for its Miare holding is 188,463,033,333 rial, not 188,462,033,333, and its total spend on the project is 378,463,033,333 rial, not 378,462,033,333, as printed in its audited statements for the six months to 21 December 2025. The 18.8 billion toman figure, the 5.00 percent holding and the 88,023 shares are unchanged.
2026-09-28, Basalam, Netbar and Tapsell, read again at their sources: Basalam’s revenue for the year to 20 March 2022 is 133,610,331,276 rial, not 133,610,331,326, as printed in the Pishgam venture fund’s audited statements; the 13.36 billion toman figure is unchanged. The company sold together with Netbar by the University of Tehran’s first venture fund is Mabna Jahan, not Mina Jahan, and the fund dates its Netbar project to 19 April 2022, not 20 April. The 2019 TechRasa interview we cite for Tapsell says its parent holding began formally about five years before the interview, not five years after Tapsell started, and Tapsell’s list of competitors comes from that 2019 interview rather than describing today.
2026-09-28, IranTalent and Nobar, read again in the funds’ filings: IranTalent’s comparative figures (revenue of 113,397 million rial, a loss of 16,706 million rial, and assets, liabilities and equity of 142,022, 136,139 and 5,883 million rial) were dated to the year to 19 March 2024. The Rooyesh Lotus venture fund’s table prints them under a column dated 20 June 2024, and neither column states how long a period it covers. On Nobar, the Arman Ati venture fund does not report parts of the company being dissolved: it reports exit negotiations with corporate venture arms, the transfer of the company’s platform and other intangible assets to the parties in negotiation, and a possible dissolution of the company as part of that exit. The profile, the company description and the venture profitability insight now say so.
2026-09-28, fifteen facts read again against the sources they cite: Pomegranate’s 2017 price of EUR 38 million for 10 per cent of the Cafe Bazaar group implies about EUR 380 million after the money, not before it, and its 2021 letter credits the growth in GMV of more than 50 per cent to the App Store, not to Divar. The Wikipedia article we cited for Divar’s origins does not name Divar’s founders, and the one we cited for its business model does not mention free listings with paid promotion. PitchBook names Tapsell as Cafe Bazaar’s buyer on 24 January 2025 but not Tapsell’s owner, so that ownership chain is withdrawn from public view. Hamkaran System states more than 100 subsidiary companies, not subsidiary and partner companies, and a sales network of 22 regional and 23 licensed companies; its product page offers Rahkaran in cloud and on premise models rather than a separate Rahkaran Cloud. The profile we cited for ZarinPal’s co-founder Ali Amiri does not call him chief financial officer, and ZarinPal’s gateway page describes the gateway alone, not a wallet, card or in-person payment. Azki’s share of more than 80 per cent of online insurance services comes from its investor MAX Holding’s portfolio page, not from the interview we had cited, and a count of 500,000 policies from 20 insurers is withdrawn because that interview does not give it.
2026-09-27, a listing board, and three sources that changed: Hamkaran System was described as trading on the second market; the exchange data shows it on the main board of the first market. Its market value is now the 27 September 2026 close rather than a July reading that the exchange page no longer shows. Fanava Group’s site no longer prints a founding year, ten companies or its licence list: it now states 20 years of experience and lists eight subsidiaries, and the record follows it. Padvish’s product site now redirects to Amnpardaz, which no longer publishes the sensor list we had carried, so it is removed. The GapGPT app listing is gone from both Android stores it was read on, so its figure is withdrawn from public view.
2026-09-27, registry records read in full, and a website that was never the company’s: ArvanCloud’s registered capital was published as 14,750,000,000 rial; the gazette gives 14,750,000,000,000 rial, a thousand times more, and the smaller number is its share count. From SnappPay’s gazette notices: the shareholders’ representative is Jalal Gharaei, not Gharapi; its gold and silver trading licence dates from 11 July 2026, not 16 July; ordinary documents can be signed by any one of the four officers, not the managing director alone, and binding ones need the chairman or vice chairman among the two signatories; and only the 2024 and 2025 capital increases are shown as paid in by its parent, so the record no longer says every one was. Torob’s chief executive was appointed on 8 March 2025, not 9 March. Blu’s page at Bank Saman moved; at its new address it does not date the launch or mention the deposit guarantee fund, so both claims are removed. The website listed for Max Digital, maxdigital.ir, belongs to a thermostat retailer; it was never this company’s site and has been removed from the record. Each record now carries the source’s own words, and each note says what was wrong.
2026-09-27, fifteen more facts, read again on the source’s own pages: reading Iranian company sites from inside Iran to attach each fact’s exact source words found fifteen that said something else. Basalam’s 2021 round named Mabna Holding as an investor; Digiato’s report names Mobin Holding. Asan Bime’s brokerage licence code is 1998, not 1049. Bankino is a subsidiary of Middle East Bank, but its page does not say it operates under the bank’s licence, and we had said so. Finnova was called a subsidiary of Ertebat Farda; its page lists Ertebat Farda Holding as a partner. Iva’s publisher is Sadad Electronic Payment Company, not Sadad Informatics, and it sells air and hotel bookings, not bus tickets; our 1 September post on X and Telegram used the wrong company name. Melli Gold has branches in 19 cities, not 20, and Mashhad is not one of them; its 36 per cent figure for the 5-gram bar is a share of users, not of delivery orders. Daric’s page does not say gold is stored free of charge, and prints its 0.5 fee with no unit. Talasea does not name Parsian coins or bars as delivery options. APK’s founding year runs from March 2007 to March 2008 in the calendar it states it in, so it is not 2007 alone. A clause about Arian System’s products and one about APK’s maturity model were not on the cited pages and are removed. Where a page had changed since we read it (DigiExpress’s volumes, Delino’s cities, Keepa’s merchant count, Farabi’s assets under management), the fact now gives the current reading and its date. Each record now carries the source’s own words, and each note says what was wrong.
2026-09-27, forty facts that their own sources contradict: writing each source’s exact words beside its fact, from copies of the pages the record already cited, found forty that said something else. The larger ones: Irancell’s 13 years belonged to its founding chief executive, Alireza Ghalambor Dezfouli, not to his successor Bijan Abbasi Arend, and the 11 per cent growth was in job positions, not employees. The Telecommunication Company of Iran’s ownership breakdown dates from May 2013, and its fourth 20 per cent is the government’s, not public float. SEP’s 625 billion toman month is the month to 20 March 2025, not the last month of a year to December 2024, and a 20.82 per cent market share its source does not carry is removed. Hamrah-e Aval’s subscriber base grew 6 per cent over nine months, not about 4 million over a year. Tara’s 350 merchants are cashback partners, not its credit network. Doctoreto’s 46,000 includes dentists and psychologists. Myket’s figures are about 800 billion toman of operating revenue and more than 340 billion toman of net profit, not exact figures. Digikala: the head of Ganje is Parham Kazemi, not Arian Bagheri; Pomegranate’s holding in IIIC is 28.72 per cent, not 28.62; and Pomegranate’s own report gives Digikala’s growth in net merchandise value as 71 per cent in one place and 69 in another, and both are now shown. Tosan’s chief executive has held the post since late 2022, not December 2025. Where a page had changed since we read it (app store listings, price lists, product and portfolio lists, Torob’s 15 million monthly users, Digishahr’s 400 million toman ceiling), the fact now gives the current reading and its date. The rest were qualifiers lost in transcription: “more than” added or dropped, a date a few days out, a claim widened beyond its source. Each record now carries the source’s own words, and each note says what was wrong. The Digikala dossier was regenerated from the corrected record.
2026-09-26, six figures that their own reports contradict: writing the source’s exact words beside each figure, from the report the record already cited, found six that said something else. Divar: “44M unique phone numbers” counted handsets that opened the app, not phone numbers (the report puts phone numbers that posted ads at 18.5 million); and “141B+ ad views” was 41 billion unique views, the report’s plus sign having been read as a 1. Cafe Bazaar: food and restaurants “up 160%” is a worldwide Amplitude figure for 2020 to 2021 that the report quotes, not Iranian data, and is removed. Lendo: the month of more than 700 billion toman in loan requests was never called the peak, and the 67 per cent paying on time is a share of customers, not of instalments. Karnameh: the 70 and 94 per cent damage rates cover inspected cars above 100,000 km, not every car inspected, and the 709 million toman average covers all new and used cars, not domestic ones. Taline: the busiest day’s gold volume is about 37,518.55 grams, not 37,518,555; the report uses the slash as its decimal mark. Each record now carries the report’s own words and page.
2026-09-21, the App Index, a platform gap we never disclosed and one we described wrongly: the App Index carries install and rating figures from Cafe Bazaar and Myket. Both are Android stores, so every figure in the index has been an Android figure, and the page did not say so. A reader comparing two companies there was comparing their Android traction and being told it was their traction. Worse, the page listed Sibapp and Sibche among “several Iranian Android stores” still to be added. Both are iOS stores. Iranians cannot use Apple’s App Store normally, so iOS distribution in Iran runs through domestic stores, which makes those two the other half of the market rather than a long tail. Every figure is now labelled Android on the page, in the dataset download and in the licence line, and the iOS absence is stated in the open. The platform of each store is now held in one table in the code, and a build gate fails the deploy if a store reaches the data unclassified or if a source file files an iOS store under Android. The gate was tested by making it fire on the exact sentence that shipped.
2026-09-20, System Group, a net profit wrong by a transposed digit: the record gave net profit for the year to 20 March 2026 as 33,062,411 million rial. The company’s own annual general meeting filing on Codal (tracing number 1547991) prints ۳۳,۰۶۴,۴۱۱. Ours was 2,000 million rial low and matched neither the parent figure nor the consolidated one, which is 35,592,290 million rial. This is a company listed on the Tehran Stock Exchange and the figure came from its own filing, which makes it the most serious error this record has carried. It was found the first time anyone pulled the filing’s exact words rather than a transcription of them. The record now states that the figure is the parent company’s and gives the consolidated number beside it.
2026-09-20, System Group, a chief executive’s name misspelled: we romanised the chief executive as “Mohammad Azirollahi”. The Persian is محمد عزیزالهی and the filing’s own English reads Azizollahi. The Persian is now carried beside the romanisation so a reader can match the name to the filing instead of trusting our transliteration of it.
2026-09-20, System Group, what “unaudited” actually means here: our consolidated revenue and profit figures were already labelled as coming from unaudited statements, which was correct. Reading the filings established something we had not said: the audited filing for the same period contains no consolidated statements at all, every section blank, with only the parent company’s statements audited. So there is no audited consolidated figure to compare against, for this company, for this year. A reader relying on those numbers should know that, and the record now says it.
2026-09-20, Digikala, a number whose meaning we reversed: the record carried “E-commerce market share: approximately 30 per cent of Iran e-commerce (CEO estimate)”, sourced to a televised interview. Transcribing the cited moment for the first time, the chief executive says the opposite. His words are بزرگترین ایکامرس کشور، ۳۰ درصد مارکتو نتونی بفروشی، بگن اینا قاچاقه: that Digikala cannot sell roughly 30 per cent of the market, because goods sold there are labelled contraband. It was a statement about market access, never about market share. The figure was published for about three months on one of our most-read records. The cause is specific and worth naming: the claim was written from a paraphrase of spoken Persian rather than from a transcript, and nothing in our process required the speaker’s own words to be recorded next to the number. That requirement now exists.
2026-09-20, four claims that went further than their sources: the same review found a DigiExpress daily average that was our own arithmetic presented as the source’s, an AzkiVam partnership with Bank Mellat that the cited article never mentions, and two Digikala multi-year series whose earlier years cannot be reproduced from any readable source. The unsupported parts are removed or labelled as ours. Where we have kept the earlier figures in the text it is so a reader can see what we previously asserted, not as a current claim.
2026-09-20, personal identity numbers we should never have published: a fact on the SnappPay record listed four registered officers with each person’s national identity number beside their name. Company officers are public through the gazette and their names stay. A private individual’s identity number is not corporate record and has been removed. Our own rules already said so; nothing enforced it, and a build check now does.
2026-09-16, corrections we published that never reached this page: an external reviewer told us the corrections below were not visible on the live site. They were right, and the cause was ours: every production deployment since 14 September had failed a build check, so two days of corrections sat in our repository and never reached a reader. We had told the reviewer the changes were published without once opening the page. A correction that does not deploy is not a correction. Both failing checks are fixed and this page is the first deployment since. We are also recording the gap itself rather than quietly catching up: anything dated 15 September below was written then and became visible only today.
2026-09-16, how we counted the link between a brand and its company: for a company to count as fully identified, we require both a registered legal entity and a source that ties the consumer brand to it. The second test was being applied by a text match on the title of a claim, so an Official Gazette notice that names a company and never mentions the brand was counted as proof of the relationship. That is the same shortcut we withdrew on 15 September, hidden one level further down. It is now a judgement recorded claim by claim, from the words of the source. Snapp’s link is established by Snapp’s own terms of service, which name the operating company. AbanTether’s is not established at all: the site names no legal entity anywhere we can read, so that record no longer counts as fully identified and we are not filling the gap with an inference.
2026-09-15, a withdrawal we should not have made, and the false reason we printed with it: earlier the same day we removed Tapsi’s figure of 1.5 million registered drivers from the record and published the reason that the company’s own acquisition announcement “does not state” it. That reason was untrue. The announcement does state it, one paragraph below the one we had read. The figure is restored with the sentence that carries it, described as the company’s own statement rather than a measured count, and the withdrawal is itself withdrawn. We are recording this prominently because a wrong reason attached to a correction is worse than the omission it was meant to fix: it asks readers to trust a claim about a source we had not finished reading. The same announcement gives the company’s founding year as 2016 in two places and 2017 in a third, so we take no founding year from it.
2026-09-15, Tapsi, a unit question we said was resolved and was not: one row on the record reported that the unit behind two early revenue figures had been resolved to million rial and “confirmed”, while the row carrying those figures said the opposite, that no unit is printed and our reading is an inference. Both cannot be true. The investor page prints no unit; our reading comes from the unit used in a different document for a different period, which is an inference and not a confirmation. The gap is reopened and published as open, and the figures continue to carry no unit, no normalised value and no currency equivalent.
2026-09-15, replacement pointers that named the wrong company: when a claim is withdrawn and replaced, we record which claims replaced it. A check across the record found eight of those pointers, on five rows, naming claims belonging to other companies, an OKala row pointing at two Alibaba claims, a Nobitex row at an OKala claim, and so on. The written explanation beside each was correct throughout; only the machine-readable references were wrong, from a single mistake made when they were entered by hand. All eight are corrected, and the check that found them now runs every time a validation package is built, so a replacement must name a claim that exists, belongs to the same company, and is actually published.
2026-09-15, two records published without a registry identity, now said so: the *780# and Tad Studio records carry no national identifier or registration number for the company behind the brand, and until today did not say so. Both now publish that gap and name the route we tried: the Companies Registration Office’s public lookup accepts an identifier, not a company name, so it cannot be used to find one. We do not substitute a commercial registry aggregator for the primary notice.
2026-09-15, forty assertions formally withdrawn, and a rule we had wrong: in the course of an external validation exercise we removed fifty-eight claims from our delivered record because their evidence did not meet our own published standard, thirteen of them resting on a registry intermediary we had already ruled was for discovery and never for citation, three citing this website as the source for its own claim, and the rest sourced to pages that no longer resolve, to video, or to login-walled surfaces. We described those rows as “withheld, with a published reason”, and treated that as sufficient. It is not. A withheld claim is one we still hold and have neither evidenced nor retracted; hiding it changes what a reader sees rather than what we are asserting. There are two honest outcomes and no third. The full list, each with the claim as it stood, the source it had and its resolution, travels with the validation package.
Counts as at 2026-09-16, across the eighteen records in the validation sample: of 59 removed assertions, 12 are restored (4 unchanged and 8 restored more narrowly than they stood) and 47 are withdrawn. This entry first published 18 restored and 40 withdrawn on 2026-09-15. It changed because we reclassified three resolutions the following day, after an external reviewer showed that pointing a withdrawn claim at a valid replacement does not make it restored: two rows had been recorded as restorations when nothing of what they asserted came back, and one was labelled unchanged while its replacement dropped part of it. The earlier figures were a fair account of what we believed on the day and are wrong now, which is why both are printed. These four numbers are read from a single constant that the build checks against the database on every deployment, so this paragraph cannot drift from the record again without the build stopping.
2026-09-16, one withdrawn claim was missing from every register we have sent: the check we added for the paragraph above disagreed with our own database by one row, and the check was right. The list of removed assertions that travels with each validation package was being filtered by a case-sensitive text match, so a claim whose internal note reads “Superseded” rather than in capitals never appeared in it. One claim was affected, an Alibaba chief-executive entry withdrawn on 7 September, and it was absent from every package we issued between 7 and 16 September. The claim itself was off the public record throughout, so nothing unsupported was ever shown to a reader. What was wrong is more particular than that: the register exists to prove we did not reach an accuracy figure by quietly deleting claims, and a silent omission from that file is the one place in the delivery where an omission undermines the thing the file is for. Fixed, counted, and the corrected totals are the ones printed above.
2026-09-15, Alibaba officers corrected from the primary notice: further to the entry below, we have now read the latest Official Gazette notice directly (gazette 23666, 9 Tir 1405, page 24). It records Majid Hosseini Nezhad as managing director and chairman of the board to 4 Tir 1406, with Mir Nima Ghazi as vice-chairman representing Divan Ghavam, and Saeed Hosseini as a board member representing Khalgh-e Ayandeh-ye Ali Baba. Our earlier row naming Ghazi as registered managing director is withdrawn and replaced by the notice.
2026-09-15, a correction of our own correction, withdrawn the same day: the Alibaba entry below named Mir Nima Ghazi as registered managing director and Saeed Hosseini as registered chairman, read from the Official Gazette’s board table. An external evaluator reports that a later gazette notice gives different officers. The board table lists historical seats beside current ones and we read terms rather than the most recent notice, which is a method error and ours. The claim is off the record and nothing is asserted about this company’s officers until the latest notice is read directly. The entry below stands as the history of what we published and why, not as a current statement.
2026-09-15, five further claims withdrawn after an external substantive review: a row-by-row review of 111 claims by the same evaluator found defects we had not caught. Snapp’s operating company carried a registered capital of 12,097,381,999,290 rial from an intermediary, against which a later gazette notice reduces the capital; the figure is withdrawn and none is asserted. A Tipax board entry listed five corporate members as current, one of whose registered term had already ended; corrected to four. The AbanTether trademark registration date, published as 9 August 2022, rests on a Persian date the evaluator reads as 18 Aban 1401 rather than 9 Aban 1401; the date is withdrawn pending a re-read. A claim that a co-founder put Alibaba.ir at about 70 percent of Iran’s online travel market is withdrawn: the evaluator reports the interview passage concerns Jabama, a different company in the group. Each carries a dated re-verification item.
Superseded 2026-09-15, kept as the history of what we published, not a current statement. The officers named in this entry were withdrawn the same day; the gazette 23666 entry above carries the current ones.2026-09-15, Alibaba, who runs the company: the record named Tohid Aliashrafi as chief executive since August 2024, taken from a self-declared professional profile, and in August it withdrew a reading that named Mir Nima Ghazi. Read on the Official Gazette’s board table, the desk took Mir Nima Ghazi to be the registered managing director with a term to June 2027 and Saeed Hosseini the registered chairman, a reading since withdrawn. The withdrawn reading was right about the registered office. The record now carries the registered officers as a registry-sourced claim and the publicly presented chief executive as a separate claim marked not independently sourced, because no company announcement, filing or registry record naming him in that office has been read. The public chief-executive field is cleared until one is. A registered managing director and an operating chief executive are different offices and the record no longer merges them.
2026-09-15, Nobitex, ownership: the record said the operator had no parent and was “not part of any known group”, citing a company registration record. A registration record shows who is registered, not who controls, and the share register of an Iranian private joint stock company is not public. The claim of independence is withdrawn. The record now states what the gazette board table shows, that the board seats only individuals with no corporate member, and says plainly that no parent or controlling shareholder is established either way.
2026-09-15, Respina24, registered capital: the record gave the operator’s registered capital as 24,750,000 rial. Read on the Official Gazette’s own portal, the founding notice states 25,000,000 rial; 24,750,000 rial is the share one of the two partners holds after a 2014 transfer, and 250,000 rial is the other’s. A partner’s holding was published as the company’s capital. Corrected, with the founding notice and the transfer notice now cited on the record.
2026-09-15, Tipax, registration number and registered capital: the record paired the operating company’s national identifier with a registration number belonging to a different, older Tipax entity, and carried a registered capital of 2,000,000,000,000 rial that an increase of August 2025 had superseded. The gazette gives registration 518700 for Tipax Express Pars and a current capital of 4,000,000,000,000 rial, raised out of retained earnings. Both corrected. The brand’s 1973 founding date refers to the brand, not to this company, and the record now says so.
2026-09-15, Maktabkhooneh, registration number: the record held a registration number one digit away from the one the Official Gazette’s register prints for the operator. Corrected from the register row.
2026-09-15, four monetary presentations withdrawn or relabelled: a review of how figures are presented, prompted by an external evaluation of our own evidence package, found four that claimed more than the evidence supports. Snapp’s driver-income figures and Tipax’s declared-value limits were held from dated readings of pages that no longer show them, with no archived copy; both are now off the public record entirely, held internally until they can be read at source again, which is stricter than the caveat first applied to them. A Tapsi early-revenue claim was labelled “rial basis confirmed” when the source page prints no unit at all and the unit is the desk’s inference; relabelled. A Baloon loan ceiling carried a dollar conversion that our own published convention does not permit below one billion toman, and the arithmetic beside it still used a rate that had already been corrected; the conversion is withdrawn and both superseded figures are kept as history.
2026-09-15, Fidibo, the registered entity: the record’s entity note described the registered company behind Fidibo as a dormant shell registered in 2014 with no filings since, and suggested the business might sit under the Digikala operating company. Read on the Official Gazette’s own portal, two companies carry the Firouzan name. The 2014 company is not the operator. The operator is Afarinesh Ketab Electronik Firouzan, registered in September 2015 and active, which owns the registered Fidibo trademark, seats Digikala E-Commerce Group on its board, and has Fidibo’s chief executive as its registered managing director. The record had also paired one company’s national identifier with the other’s registration number. Entity note rewritten, identifiers corrected, four gazette-sourced claims added. Found in the desk’s own work for the validation pilot.
2026-09-15, Tapsi, standalone net loss for the year to 20 March 2026: the record carried a standalone net loss of 2,378,611 million rial (about 238 billion toman) beside a loss per share of 1,819 rial, and a note that reconciled the two by assuming a mid-year weighted share count. Re-read against the audited statements on Codal, the standalone net loss is 4,478,611 million rial (about 448 billion toman), and the per-share figure reconciles with it on the full 2,462,567 thousand shares. The loss figure was mistranscribed on 10 July; the note built on the error is withdrawn. The consolidated figures (revenue 35,489,234 million rial, net loss 5,495,331 million rial) were confirmed against the same audited letter. Found in the desk’s own re-read for the validation pilot.
2026-09-15, Nobitex, volume processed, a misattribution: the record said about $7.2 billion was processed in 2025, attributed to Reuters. Reuters’s investigation of 1 May 2026 gives a cumulative total of $11 billion processed and does not print $7.2 billion; that number appears in a syndicated commentary piece with no stated period or method. Corrected to the Reuters figure, with its period stated as unspecified by the source, and the earlier figure withdrawn.
2026-09-14, Nestlé Iran, share capital, a unit error: the Foreign Exposure table gave Nestlé Iran’s share capital as 358,538,000 rial. Nestlé’s 2024 financial statements print it as KIRR 358 538 000, in thousands of rial, so the figure is 358,538 million rial, one thousand times what we showed. The quotation beside the field carried the K throughout; the normalised field dropped it. Found in an external review of the public site run with an AI model, which checked the row against the filing; confirmed by the desk on the printed page. Corrected on the table and in the insight that cites it. Every monetary row on the table now carries the value exactly as the filing prints it and the scale the filing applies, beside the normalised figure, so the transformation can be checked without opening the document; the Roche row, printed under an “in millions” header, was re-read and stands.
2026-09-07, Irancell, subscribers: our record carried more than 69 million subscribers, sourced to Persian Wikipedia citing the Communications Regulatory Authority. MTN Group, which owns 49 percent of Irancell and reports it as an equity-accounted joint venture, publishes a quarterly subscriber series in its audited results data sheets. It gives 56.2 million at 31 March 2025 and 57.6 million at 31 December 2025, roughly 12 million below the figure we carried. The record now uses MTN’s series, with the date of each reading. The regulator’s own count is expected to be higher because it reports registered mobile subscriptions rather than active customers; the two definitions will be set out side by side once the regulator’s statistics page can be read directly.
2026-09-07, Takhfifan, Pomegranate’s stake: our working record held a 2.3 percent direct holding by Pomegranate Investment AB, taken from a footnote quoted in part. Read in full, the footnote assigns 2.3 percent to Digikala and 21.3 percent to Takhfifan at 31 October 2023, rising to 21.9 percent at 30 April 2024 and unchanged at 30 April 2025. The published ownership entry, which had given no percentage, now carries 21.9 percent direct plus 3.2 percent through the Dutch holding company.
2026-09-03, Takhfifan and Netbarg, the Competition Council ruling: our record, sourced to the regulator’s press announcement, dated the decision 25 December 2022 and said the fine was not disclosed. The numbered decision, session 531 of 20 December 2022, gives the date and the fine: 5.962 billion rial. More materially, the record presented the order to unwind the Netbarg transfer as standing. It was quashed in full by the Council’s appeal board on 21 April 2025 and the case remains undecided between the two bodies. Both entries on the Takhfifan record now carry the decisions themselves.
2026-09-03, Digikala, exchange admission: our record and two insight answers said Digikala had never completed a listing application, and a 31 August signal said no exchange filing had been published. The Tehran Stock Exchange’s admission board approved Digikala E-Commerce Group’s shares on 8 January 2025 and the exchange published the decision that day. Corrected to admitted, not yet listed or offered.
2026-09-03, Snapp, MTN shareholding: the 43 percent MTN stake was presented as a current holding. It rests on one 2020 interview with MTN’s chief financial officer and appears in no MTN filing from 2022 to 2025. Reattributed to the quote, with its date.
2026-08-06, Five insights and two records, perimeter and figure revisions: an external review found language in several earlier insights that crossed our stated coverage perimeter (business, technology and markets only), plus two figures that had drifted from the live registry. The field guide, the payments piece, the e-commerce piece and the travel piece were revised to state the access constraints factually and drop speculative framing; the Digikala group-company count was aligned with the live dossier (the group’s own materials list 10 or 11 companies, not 16), and the fintech firm count is now anchored to the registry (149 tracked at company level) alongside the industry estimate (~275). A Snapp operating fact was rephrased to describe the 2025 service disruptions operationally. Each revised piece carries a dated revision note.
2026-07-21, Alibaba (Safarhaye Alibaba): our record carried the founder as chief executive. Corrected to the current managing director, appointed August 2024, after a leadership-freshness review against Persian press and company sources.
2026-07-21, Sheypoor: our record carried the founder as chief executive. Corrected to the current chief executive, appointed July 2023 (per the investor’s primary announcement); the founder serves as chairman.
2026-07-21, ArvanCloud: the profile labelled the company independent. That label was removed: reporting and registry-based sourcing document a major institutional shareholder, while the company publicly disputes the extent, the ownership field now carries the sourced account of both positions rather than a label we cannot stand behind.
All three were surfaced by our own published benchmark run , the process working as designed.