Glossary
Plain-English terms for archaeological, linguistic and statistical ideas used on the platform.
Allograph
A visual variant of the same underlying sign. Whether two Indus forms are allographs or separate graphemes is itself a hypothesis, and merges must remain reversible.
Analytical reading order
The sequence used for statistical work after a reading-direction decision. It may differ from the published visual order of a photograph.
Bilingual
A text that presents the same message in a known and an unknown script. No undisputed Indus bilingual of this kind is known — the main reason the script remains unread.
Bigram
Two consecutive signs. Directed bigrams (A→B versus B→A) are a core structural test on this platform.
Classifier
A sign that may mark a class of names, goods or offices rather than spelling a word. A functional hypothesis, not a reading.
Duplicate reduction
Counting a seal and its clay impressions as one independent object so that n-gram frequencies are not inflated by copies.
Ender-biased
A sign that appears at the analytical end of texts more often than its overall frequency predicts. Not automatically a linguistic suffix.
Evidence level
The platform’s six-way label: observed, structural finding, functional hypothesis, semantic candidate, unverified proposal, rejected or unsupported.
Grapheme
The smallest contrastive unit of a writing system. The Indus grapheme inventory is still contested.
Independent object frequency
How many distinct objects carry a sign or pattern, after grouping likely duplicates. Preferred over raw token counts.
Ligature
Two or more signs written as a fused graphic. Fusion is a visual proposal until positional evidence is shown.
Logo-syllabic
A writing system mixing word-signs and syllabic signs. One working model for Indus, not a demonstrated fact.
Null model
A randomised control that asks whether an observed pattern is surprising given text lengths and sign inventories. A default here shuffles order inside each inscription.
Numeral candidate
A sign, often a stroke group, that may encode quantity. Graphic series plus context can support the hypothesis; they do not prove it.
Opener-biased
A sign that prefers the analytical start of texts. Not automatically a linguistic prefix.
PMI (pointwise mutual information)
log2(P(A,B) / (P(A)P(B))). High PMI flags attraction between signs; it still needs a null model and independent-object counts.
Rebus
Writing a word with a sign for a similar-sounding word (English ‘bee’ + ‘leaf’ for ‘belief’). Central to some Proto-Dravidian proposals, and easy to overfit.
Reading direction
Most scholars treat Indus as running right-to-left on stamp seals, with exceptions. Direction is a labelled decision with a confidence score here.
Rosetta Stone
The Egyptian bilingual/trilingual that anchored phonetic values. The Indus corpus has no equivalent object.
Sealing
A clay impression of a seal, often from goods or doors. Sealings can duplicate a sequence without adding an independent composition.
Semantic candidate
A possible meaning under a named model. Always shown with alternatives, sources, a falsification test and a non-decipherment warning.
Signary / sign inventory
The list of distinct signs. Published counts range roughly from under 400 to over 700 depending on how variants are merged.
Template
A recurring sequence pattern with optional variable slots, such as [O] – 033 – 705 – [V] – [E?].
Visual order
The left-to-right order of signs as they appear on a published photograph, before any reading-direction reversal.
Epistemic valve
The one-way path Z₁ (canonical evidence) → Z₂ (features) → Z₃ (graph/model) → Z₄ (ontology). Analytics and models cannot write back into the ledger, and ontology cannot train a model.
Provenance tier
Four-way quality of findspot: securely excavated, documented legacy, context-weak, or unprovenanced. Tier 4 contributes zero weight.
Q_robust
The product of provenance, context and direction scores used to down-weight weak objects. Missing provenance is a zero, not a guess.
DVM (Domain Vocabulary Model)
A quarantined speculative ontology and, in DVM Lab, a constructed stack language indexed on Mahadevan M_001–M_417. Opcodes are not recovered Indus readings. The runtime cannot train structural models or write the canonical ledger.
ISX (InduScript Codex Native)
A dotted command notation such as ISX::298.177.055.051. It compiles to constructed M-code and executes in the Zone-4 DVM. Founder-designed system notation — not an Indus inscription and not a decipherment.