Commons Atlas · Timeline Plate · edition 2 · forged 2026-07-23 · supersedes ed.1

Indus script

Four thousand objects. Four hundred signs. The longest text is thirty-four characters. A city of forty thousand people wrote something down, and every attempt to read it has so far read the reader back.

Deep · clamped Human · piecewise linear · densities derived from this ledger Horizon · below · Position, not fact

Learning ◆ Skill ▲ Culture ● Green ✦ · dot = exact · dot + halo + ~ = circa · band = range · ⚑ = contested · dotted wash = presence

Derived segments (width follows beats, not years)
7000 BCE → 5500 BCE	1 beats	3.1% of width	36 px/beat
5500 BCE → 3500 BCE	no beats	4.0% of width	46 px/beat
3500 BCE → 100 BCE	16 beats	49.1% of width	36 px/beat
100 BCE → 1875	no beats	4.0% of width	46 px/beat
1875 → 2026	13 beats	39.9% of width	36 px/beat

True scale — the same beats, proportionally

Computed at forge time, not estimated. Linear from 7000 BCE to 2026 — 9,026 years across the full width. The mature Harappan script (2600–1900 BCE, blue) is 7.76% of it. Everything from Cunningham's first published seal in 1875 to today — every excavation, every decipherment attempt, every rebuttal (dark bar, right edge) — is 1.67%. The spine above stretches that sliver by a factor of about 24×; this strip is what it actually looks like.

Select a beat on the spine — or use ← / → to walk them in time order.

Horizon · Position, not fact — each is somebody's stance, named

“Machine learning will read it.”

Whose position: Computational-linguistics groups applying transformer and computer-vision models to the corpus (TACL 2025; arXiv 2503.21074). · source (en · CC-BY)

“No bilingual, no decipherment — ever.”

Whose position: The sceptical mainstream: corpus too small, inscriptions too short, no confirmed descendant, no bilingual artefact. · source (en · CC-BY-SA)

“There may be nothing to decipher.”

Whose position: Farmer, Sproat & Witzel and successors — the signs may be emblematic rather than linguistic. · source (en · quotable-with-attribution)

What this plate cannot show

◦ A001 — Inscriptions on perishable materials — no papyrus, bark, cloth, palm-leaf or wooden example survives anywhere in the corpus. What we hold may be only the fireproof residue of a much larger writing practice.

◦ A002 — A bilingual inscription — none known. The single most decisive missing object in the whole problem.

◦ A003 — Non-English editions unread (run 1) — superseded by A007, which names the reachability problem precisely. Partially closed: Tamil article titles resolved and one framing divergence recorded as B033.

◦ A004 — Wikipedia oldid permalinks — the probe was run once this forge and FAILED (Special:PermanentLink, the revision API and action=info all returned empty). Documented fallback taken: live article URLs, absence kept open, declared here and in the colophon.

◦ A005 — The Harappan spoken language itself — unidentified. No confirmed descendant, and no full name of any person or place attested from within the civilisation.

◦ A006 — Oral memory of Sindh and Kutch — present in living practice, undatable by this method. Covers presence band B029.

◦ A007 — Full-text access to non-English editions — unreachable from this forge environment; every direct fetch of ta / sd / hi Wikipedia returned empty. Tamil was reached by search snippet only: enough to record a framing divergence, not enough to read a chronology.

◦ A008 — The shipped layout auditor — scripts/layout-audit.mjs, which Phase C.5 makes blocking, is not present in the skill bundle available to this forge, and the Ed.2 reference sections it cites (time-axis §densities-are-DERIVED, §packing; harvest-sources §permalink-fallback) are absent too — both reference files are still at Ed.1 text. The checks enumerated in SKILL.md Phase C.5 were implemented locally and run against this plate instead.

◦ A009 — Sindhi scholarship, still unreached after two runs. The city is in Sindh.

Leap of Faith Labs · Commons Atlas

Languages read: English (full text); Tamil (search snippet only — titles and lead sentences).

Languages attempted and unreachable: Tamil (full text), Sindhi, Hindi, Urdu — every direct fetch returned empty (absence A007).

Archives & sources: English Wikipedia (CC-BY-SA) — Indus script, Mohenjo-daro, Indus Valley Civilisation, Great Bath, R. D. Banerji; Tamil Wikipedia (CC-BY-SA, snippet); harappa.com essays; Farmer, Sproat & Witzel 2004 (safarmer.com); Rao's response page (Univ. of Washington); Mukhopadhyay (SSRN); arXiv 2503.21074 (CC-BY); CNN 2025; Britannica. A list, never a score.

Citation form: live article URLs — the oldid permalink probe was run once this forge and failed (absence A004).

Vendored code: none. The vendor slot is empty on evidence: the three-register frame, the derived piecewise scale, label-aware packing and the honesty strip are all arithmetic over ~30 beats. vis-timeline remains rejected-with-cause (single linear axis, incompatible with piecewise seams); d3-scale was not needed because the piecewise map is nine lines of code.

Ledger: indus-script-ledger.md · plate edition 2 · supersedes ed.1 (2026-07-23) — kept beside this one until M retires it · localStorage key lofl-deeptime-v1 (view state only).

Audit: geometry computed at forge time; verified by layout-audit.mjs (local implementation — see absence A008) with a text metric 10% wider than the renderer's.

CC-BY-SA-4.0 · contains material from Wikipedia (CC-BY-SA), cited per beat · no personal data of any living private person appears in this ledger or plate.