Forty-one hypotheses about the Indus script, tested against the corpus itself. The controls killed most of them; what’s left is small, and it holds.
This is not a decipherment
What survived
Every one of these is a distribution, never a meaning. Position, site and object class stay fixed inside each null, duplicated texts get collapsed before anything is counted, and each headline was re-run 300 times through a noise model built from the disagreement between two independent digitizers.
Seven signs close 1,179 of 2,086 texts where a within-text position shuffle predicts 400. Signs 740 and 520 turn up together six times against 27.7 expected, and that pair is the one exclusion still standing after every control we have.
finality z = +47.5 · pair z = −5.45The construction sits at −2 whenever 400 or 90 follows it, so the paradigm floats rather than being nailed to the last position. Behind it there is nothing: 33 candidate exclusions at −2 and not one survives.
−2 cohort: 0 of 33 exclusions after BHRepetition sits far below what the exact-position, site- and object-matched control predicts. It is the most reliable fact in the corpus, and it explains why positional statistics keep working here while sequence statistics dissolve.
z = −14.85 · noise 95% −14.17 to −10.03Both media draw on one inventory and one layout, then part company at the end. Tablets append 400 where seals mostly don’t, and seals carry numerals nearly twice as often.
p = 1e−14Seal endings at Mohenjo-daro and Harappa can’t be told apart, and the smaller towns follow the same template. Harappa’s tablets are the only local practice in the corpus.
p = 0.98Counted off the rendered glyphs rather than inferred from database ids, the values run 1 to 9 with an ordinary decay and then jump at twelve. Sign 55 carries 37 deduplicated tokens; no decay curve reaches that.
Poisson p = 1.4e−45Pooled across the corpus the order comes out nearly 1:1, which hid the structure completely. Fifteen of 37 eligible signs are at least 90% one-sided against 0.01 expected, and the overdispersion holds through every control and every noise draw.
controlled z = +16.09Against a surrogate that preserves every observed bigram exactly, mutual information stays above null at separations of two, three and four signs. Five is borderline and six is nothing. Most of what the corpus has is local.
excess 0.052 to 0.078 bits, BH correctedRepeat avoidance, the 740/520 exclusion, terminal concentration and numeral-side overdispersion all hold in random halves and in disjoint site halves, where no site appears on both sides. They also sit outside nulls built by destroying the pairing the pipeline looks for.
exploratory paradigm scan FPR = 7.1%515 identifiable types once the twelve database markers for unidentified signs come out. Estimators that key on singletons put the real total near 700, and the accumulation curve hasn’t turned over yet.
Chao1 ≈ 715 · ACE ≈ 695250 sequence types recur, producing 684 attestations beyond the first, and 84.4% of those are another copy where the text already sat. Circulation isn’t nothing though: 35% of repeated types reach two to four sites.
577 local · 107 new-site attestationsHouse rules
Most of these came out of a mistake, and each one has a body count. They sit in the order the mistakes happened.
The source database ships a Sanskrit decipherment that its own field rejects. We took the epigraphic layer and nothing else: which signs, in what order, on which object.
Mass-produced tablets repeat one text dozens of times, and left in they manufacture significance. This rule cost 25 sign–motif findings on the day it went in.
If one sign prefers the start and another prefers the end, ordering comes free. Every null here shuffles inside exact positions.
Two signs from different cities never meet, for reasons that have nothing to do with grammar. In this corpus medium and city are very nearly the same variable.
Sorting signs by the behaviour you’re about to test guarantees a result and teaches you nothing. Round 21 withdrew a shape split for exactly this.
Everything below is the record, including the rounds that knocked over earlier headline results of our own.
The ledger
Each round came out of the one before it. When a round overturns something we had already concluded it gets tagged Corrected and the original stays on the record, instead of being quietly edited away. Open a round for the evidence and its plate.
The wall
The corrected merged corpus records 527 sign IDs, twelve of which are database markers for unidentified signs, leaving 515 identifiable types. Roughly 416 non-numeral signs still have fewer than 20 tokens. Partial pooling puts all of them on one uncertainty scale but it can’t invent observations: three rare signs came out with finality intervals above the base rate, two of those look like post-terminal additions, and the model fails its global posterior-predictive check.
515 isn’t a ceiling either. Estimators sensitive to singletons put the merged total near 695–715 and the accumulation curve hasn’t plateaued, so more data will add types as well as tokens. That makes the wall bigger rather than smaller.
Two independent digitizers agree on 93.2% of aligned signs. Propagate that disagreement and the no-repeat effect, the terminal cohort, the 740/520 anchor and the global numeral-side result all come through intact, while most marginal pairwise edges get erased. The wall is explicit uncertainty now rather than a blanket objection: strong effects clear it, and z around −2 to −3 generally doesn’t.
ICIT (4,660 artefacts, 17,957 signs) is the only corpus big enough to move any of this, and it’s request-gated. Round 41 showed our numbering is already ICIT’s, so integrating it would need no crosswalk. The gain would be stratification rather than raw power; doubling N buys about √2 on a z-score and leaves the median sign just as rare.
| Artefacts in the parsed corpus | 2,543 |
| Sign tokens | 11,135 |
| Records after deduplication | 2,086 |
| Recorded sign IDs (merged) | 527 |
| Identifiable types | 515 |
| Non-numerals under 20 tokens | 416 |
| Estimated true inventory (Chao1) | ~715 |
| Inter-digitizer agreement | 93.2% |
| Tokens perturbed per noise draw | 668 (7.3%) |
| ICIT, if it can be obtained | 4,660 / 17,957 |