One Concordance, 172 Decipherments, And No Two Agree On A Sign.
Sixty-one sign-by-sign readings, 1,830 pairs tested. The best agreement between any two is four signs out of 417. Seven pairs agree on nothing. The field's working text has nineteen typos.
A months-long reporting project. Documents cited below are held in The Vault and available to readers.
CHENNAI, India — Four signs.
Out of a sign list of four hundred and seventeen, four is the largest number that any two independently produced readings of the Indus script have ever agreed on.
Seven pairs agree on nothing at all — not one sign in common.
We arrived at those numbers by treating a hundred and fifty years of decipherment as a dataset rather than a debate.
What we collected
A hundred and seventy-two published claims to have read the script, from 1875 to 2025. Journals, monographs, pamphlets, two doctoral theses and one patent application.
Sixty-one of them assign a value to a numbered sign, which is the only kind of claim that can be checked against another. The rest describe a method, or a language family, or an idea.
Sixty-one readings make 1,830 pairs. We compared all of them, counting a match only where two readings give the same value to the same sign number.
Best pair: four signs. Median pair: one. Seven pairs: zero.
Twenty-one of the sixty-one propose Dravidian, sixteen Indo-Aryan, four Munda, three a link to Sumerian, six that it is not writing at all, and eleven that it is a language nobody has named.
The book underneath all of it
Eighty-four per cent of the work published since 1977 is keyed to one concordance, printed in an edition of one thousand and out of print since 1989.
That book is why the field can talk to itself. It gave every sign a number. Without it, two scholars describing the same mark would be describing different things.
We licensed photographs of 300 seals and collated them against it, sign by sign.
Nineteen errors.
Not interpretive disagreements. Transcription errors: a sign number printed where the photograph shows a different sign, a stroke count of five entered as four, one line reversed, and on seal M-1103 an entire terminal sign that is not on the object.
All nineteen are listed with this story, with seal numbers, so anyone may check us.
Errors that travel
Three papers published this year — 2026 — reproduce the M-1103 reading with the sign that is not there.
Their authors have not been careless. They used the standard reference, as everyone is told to. The photograph they would need to catch it sits in a collection whose reproduction licence lapsed in 2011.
The edition that has been in preparation since 2004
A corrected digital concordance has been described as forthcoming for twenty-two years.
We asked why, and got three answers.
The concordance's custodian, Prof. Aravindan Seshadri-Roche, says three institutions hold the photographs, one licence has lapsed, one collection refuses reproduction of objects it considers unprovenanced, and he is seventy-nine.
A funder told us the project has been supported twice and delivered neither time.
A former research assistant told us the corrections exist on paper, in a filing cabinet in Chennai, and have existed since roughly 2009.
We cannot reconcile the three and we are not going to pretend we can. All three are printed here as we received them.
Why this is not a secret
The thing worth saying plainly is that nothing is being withheld.
The median Indus inscription is five signs long. The longest known runs to twenty-six, on a three-sided prism. The entire corpus — every seal, tablet and potsherd, added together — comes to fewer sign occurrences than one week of a Mycenaean palace's grain accounts.
Dr. Nalini Kothari-Bergström put the threshold plainly. No script has ever been deciphered from a corpus this short without either a bilingual text or a known underlying language. Linear B had thousands of tablets and Greek waiting underneath. Egyptian had the Rosetta Stone. Maya had Landa and long inscriptions on stone.
The Indus corpus has none of those, and it is short.
What this desk takes from it
A field that everybody assumes is blocked by mystery is in fact working from an out-of-print book with nineteen typos in it, and it knows.
The 172nd decipherment will be announced, probably within the year, and it will agree with the 171 before it on approximately one sign.
That is not a scandal. It is a corpus problem, and it has a boring remedy: a corrected edition, three licences renewed, and photographs anybody can open. Two of those three are administrative.
Professor Seshadri-Roche's reply runs below, at his length. He thinks our arithmetic is a conceit and says so at some cost to us. He is partly right, and the sentence in it this desk cannot get past is the last one.
The six o'clock edition
Three stories a morning, in your inbox before the coffee is made.
Sources & Method
We did not attempt to read the script. We treated the decipherments as data: collected every published claim we could find from 1875 onward, kept the 61 that assign a value to a numbered sign, and compared all 1,830 possible pairs for agreement, counting a match only where two readings give the same value to the same sign number. Separately we licensed photographs of 300 seals and collated them, sign by sign, against the printed concordance almost every modern paper is keyed to. Our collation method was shown to an outside Assyriologist and criticised before we used it.
Who we spoke to
- 172 decipherment claims, 1875–2025, Journals, monographs, pamphlets, two theses and one patent application. Assembled by this newspaper over fourteen months; the bibliography is published in full 2025–2026 Eleven claims were excluded as duplicates of an earlier claim by the same author. The exclusions are listed.
- Prof. Aravindan Seshadri-Roche, Epigraphist; custodian of the concordance and its unpublished revisions. Interviewed twice in Chennai; read the full draft and the error list April and June 2026 Accepts eighteen of the nineteen errors and disputes the framing of all of them. His reply is printed entire.
- Photographs of 300 seals, Three museum collections; images licensed for this collation. Licensed at commercial rate and collated sign by sign against the printed concordance February–May 2026 Two of the three collections permit us to republish the images. The third does not, and that is stated on each entry.
- Prof. Ekundayo Adeleke-Sørvaag, Assyriologist; published the Ur III closing-formula concordance held in this newspaper's vault. Asked to inspect and criticise our collation method before we ran it January 2026 Told us our method would find typographical errors and would miss editorial judgements, which is exactly what happened.
- Dr. Nalini Kothari-Bergström, Computational linguist; works on corpus-length thresholds. Interviewed by video call three times March–June 2026
Documents
- PX-1908 — Adeleke-Sørvaag, concordance of Ur III administrative closing formulae (1998), §4 accepted
What we could not confirm
- Whether the corpus encodes a language at all. Six of our 172 claims say it does not, and the conditional-entropy work cited by both camps is, on our reading and on Dr. Kothari-Bergström's, compatible with either answer.
- Whether the nineteen errors are all of them. We collated 300 seals against a concordance of roughly 3,700 entries — eight per cent of it. The remaining ninety-two per cent has not been checked by us or, so far as anyone would tell us, by anybody.
- Why the corrected digital edition has not appeared in twenty-two years. We have three accounts from three people and they do not agree with one another. All three are printed in the story.
How Others Covered This
The same events, as reported elsewhere on the same day. We list what each outlet had that we did not, as well as what we had that they did not — including where we come off worse. Why we print this.
-
The Meridian TelegraphIndus Script Cracked At Last, Says Researcher
Ran the 172nd claim as news, on the day of the preprint, with no comment from anyone else.
Had that we did not
The researcher, at length, and a clear account of his method.
Left out
The other 171. A reader of that piece has no way of knowing this happens roughly twice a year.
-
Signal & LedgerCorpus Digitisation Enters Its Twenty-Second Year Of Preparation
Trade press, following the money: grant rounds, institutional permissions, image rights.
Had that we did not
The permissions tangle, better than anyone. Three collections, three licences, one of them expired.
Left out
That the delay has a cost measurable in published papers. It treats the digital edition as a project rather than as the thing the field is waiting on.
-
They Buriedthis newspaperOne Concordance, 172 Decipherments, And No Two Agree On A Sign.
Collected every claim, tested the testable ones pairwise, and collated the field's working text against photographs.
Had that we did not
The full bibliography, the pair matrix, and all nineteen errors with the seal numbers.
Left out
Our headline says no two agree on a sign. Four pairs agree on as many as four. The second paragraph says so; the headline does not carry the qualifier and it should. — V. Ashcombe-Doyle, standards editor
Right of Reply
They Buried contacted Prof. Aravindan Seshadri-Roche, epigraphist given the draft and the full error list on 19 June 2026, with three weeks. Replied on 8 July 2026. Printed in full and unedited, including the paragraph about this newspaper's own error rate.
Eighteen of your nineteen are errors and I thank you for them. The nineteenth is not an error, it is a reading, and I will come to it.
First, your arithmetic.
You compared one thousand eight hundred and thirty pairs. Among those pairs you have set a 1911 pamphlet by a retired railway engineer against a 2019 paper using a hidden Markov model, and reported, with an air of discovery, that they disagree. Of course they disagree. One of them is not a decipherment; it is a hobby with a printer. Your matrix treats every claim as an equal citizen of the same republic and then expresses surprise at the noise.
A scholar in this field could sort your sixty-one into three piles in an afternoon, and the top pile would agree with itself considerably more than four signs. You would then have a much less interesting number and a much more honest one.
Second, the errors.
Nineteen in three thousand seven hundred entries is an error rate of one half of one per cent, in hand-set type, proofed in 1976 by two people, one of whom was me and one of whom is dead. I would ask what your own rate is over a comparable body of set text, and I would ask it in public, because you have asked me in public.
That is not a defence. It is a proportion, and your piece has none.
Third, and this is where I will give you more than you asked for.
The edition of one thousand was a decision. In 1977 a thousand copies went to the institutions that would use them and the price was kept low deliberately. It did not occur to me, and I do not think it occurred to anyone, that a printing of a thousand would still be the field's working text in 2026, or that the errors in it would be photocopied forward for half a century by people who had never seen a seal.
The digital edition is not delayed by secrecy. It is delayed because three institutions hold the photographs, one licence lapsed in 2011, one collection will not permit reproduction of objects it considers unprovenanced, and I am seventy-nine.
Print that last clause. It is the true reason and it is the one nobody writes down.
Published unedited under our right-of-reply guarantee.
How was this story?
We publish the result, whatever it is. Reader verdicts appear on the front page and in our newsroom metrics.
Readers' Letters 0
Printed at once under the name you give and read by the desk afterwards; anything unfit is removed, with a note saying so, and nothing else is ever deleted — only corrected. Letters that changed something in the story carry a mark saying so, and there are 5 of those across the archive.