104

Current locked corpus: closing / terminal function. Role label ender-biased.

Why it is on the table: 17 tokens, start-bias 0.26, end-bias 3.39.

What would weaken it: Bias is corpus-relative. Direction errors and sign-splitting move signs between roles.

Induscript analysis engine on the locked compilation

ICIT 104 is drawn in the stroke / numeral class. Class is a graphemic convenience used so relatives can be compared — not a word, not a translation.

Why it is on the table: Family membership is the cheapest hypothesis: it only claims that similar drawings should be tested together.

What would weaken it: Wells over-splits; Parpola and Mahadevan merge. The family of this page is a working cluster.

Wells/ICIT graphemic sets; contrast Mahadevan 417-sign list

Stroke groups that scale by count are the strongest numeral candidates in the signary. Fairservis and Mahadevan both treat some of this series as quantitative.

Why it is on the table: Shape is systematic (1–4+ bars) and slots are constrained. Indus weights exist independently.

What would weaken it: Strokes can be dividers, check-marks, or phonetics. No object is bilingually labelled by these bars.

Fairservis; Mahadevan 1977 numeral discussion; Kenoyer metrology

Not a one-to-one dump of Mahadevan’s 417-sign list. ICIT 104 is a Wells/Fuls number; merge/split decisions differ.

Why it is on the table: Mahadevan remains the field’s default concordance even when sign counts disagree.

What would weaken it: Crosswalks between M-numbers and ICIT G-numbers are incomplete. Do not treat them as identical signs.

Mahadevan, The Indus Script: Texts, Concordance and Tables (1977)

Hunter produced the first systematic post-Marshall inventory. Early pictographic identifications (fish, jar, man) survive as graphic labels, not as readings.

Why it is on the table: Historiography matters: many popular ‘translations’ still quote Hunter-era picture-names as if they were words.

What would weaken it: A 1934 pictographic guess is not evidence. It is the starting point later concordances refined.

G. R. Hunter, The Script of Harappa and Mohenjodaro (1934)

ICIT encodes this as a three-digit sign in a 700+ inventory (Fuls 2022–23: 709–713 signs in 71 graphic sets). Rare or unlisted in the excerpted frequency tables.

Why it is on the table: Finest published graphemic split. Useful for not collapsing allographs too early.

What would weaken it: Parpola argues Wells over-splits variants. Inventory size is itself a hypothesis.

Wells 2011/2015; Fuls, A Catalog of Indus Signs (2023)

Parpola’s logo-syllabic Dravidian model would assign this stroke / numeral class sign a rebus only after a pictographic identification. None is established here.

Why it is on the table: Most fully worked phonetic hypothesis in print. Internally disciplined, still unproven.

What would weaken it: Requires language identification first. No bilingual. Rebus steps multiply uncertainty.

Parpola, Deciphering the Indus Script (1994); 2010 lecture

Computer-aided counts in the 1960s–80s already treated short texts as structured and leaned Dravidian. This sign is a data point in that tradition, not a Soviet ‘translation’.

Why it is on the table: Independent of Parpola in method; similar language lean. Shows the Dravidian hypothesis is not one author’s idiosyncrasy.

What would weaken it: Early computing on small samples overfits. Language lean was prior, not a result.

Knorozov and collaborators, Soviet studies of the Indus script

Fairservis treated scaled stroke groups as quantities and some implement signs as phonetic. This sign sits in that testable subset.

Why it is on the table: Useful because it names functions (numeral, implement) that can be tested against weights and object type.

What would weaken it: Many Fairservis values are under-constrained. Kept as historiography plus a few live functional tests.

Walter Fairservis, The Harappan Civilization and Its Writing (1992) and earlier papers

Short, non-random emblem sequences on elite objects can be non-writing: lineage marks, totemic formulas, or ritual identifiers.

Why it is on the table: Explains brevity, lack of bilinguals, and seal function without a lost language.

What would weaken it: Rao et al. 2009 argue conditional entropy looks more like a script than a random emblem set. Neither side has a bilingual.

Farmer, Sproat and Witzel (2004); contrast Rao, Yadav, Vahia et al. (2009)

Block-level entropy and bigram directionality are treated as evidence that the system is structured like a script, not as a translation. This sign’s locked-corpus role is ender-biased.

Why it is on the table: Statistics constrain function, not phonetics.

What would weaken it: Structured non-linguistic systems also have bigrams. Entropy is not a language ID (Sproat 2014).

Rao et al., Science (2009); Yadav–Vahia positional papers; Sproat 2014 critique

Whatever the signs ‘say’, seals and sealings are administrative and craft devices. Meaning, if any, is tied to identity, ownership, or office — not narrative literature.

Why it is on the table: Archaeology is the only bilingual we actually have: context, not captions.

What would weaken it: Context can suggest function of the object without reading the signs.

Kenoyer; Vidale; Frenez; Possehl syntheses

Proposed Sanskrit or Vedic values for Indus signs. No bilingual, no accepted crosswalk, and chronology is contested. Recorded so the model is visible — not so it can be quoted as a reading.

Why it is on the table: A research platform that hides a popular school is not evidence-led.

What would weaken it: Failed the bilingual test, the positional test, and the ‘does not overfit’ test as currently published.

Subhash Kak and related Indo-Aryan proposals

A single securely read bilingual (Indus vs a known script) would re-rank every card on this page. Until then, depth means listing tests, not printing a word.

Why it is on the table: Egyptian, cuneiform and Linear B moved when bilinguals or internal grids arrived. Indus has neither.

What would weaken it: Waiting for a bilingual is not an excuse to ignore structure. It is an excuse to refuse fake certainty.

Robinson surveys; platform method

If reversing analytical order destroys this sign’s start/end bias, the role label is direction-fragile and must not be promoted.

Many seals can be rotated. Role labels that survive a flip are stronger.

Platform method; Mahadevan direction discussion

If merging this sign with its stroke / numeral class relatives (or splitting its variants) moves it between opener/ender/core, the role is an artefact of the sign list.

Wells vs Mahadevan vs Parpola disagree on inventory size. That disagreement is a test, not a nuisance.

Wells 2015 vs Mahadevan 1977 vs Parpola CISI

If the sign’s neighbours change by object type (seal vs tablet vs pottery), it may mark register, not a word that is constant across media.

Kenoyer/Vidale: different objects do different jobs.

Kenoyer; Vidale; mixed-register model

A bilingual in which this sign is clearly medial, or a large tablet series where it never ends a line, would weaken the terminal-function claim.

End-bias is the current structural fact. Facts can move.

Platform falsification rule

A weight, vessel, or tally stick whose independent measure matches the stroke-count of this sign would promote the numeral reading one step — still not a language ID.

Indus weights are a real, separate dataset (Kenoyer metrology).

Kenoyer / Hendrickson metrology

Only a securely provenanced bilingual, or an internal grid of the Linear-B type, can promote any card on this page from candidate to established meaning.

That is how every accepted script decipherment actually happened.

Robinson; platform method