glyphsift demo

Is this extracted Devanagari text clean? The score, the per-line issues and the report are computed in WebAssembly in this tab. Nothing is uploaded; the only network requests are for this page's own files.

Build …

Sprint 5 (18 Nov 2026): Working: everything from Sprint 4, plus a Gate tab that gives a pasted chunk of Devanagari text one of three answers, pass, hold or not scored (text that is not mostly Devanagari is passed through and not scored), with the score, the Devanagari share and a threshold you can change (975000 ppm by default), as the same canonical JSON the command line and the Python wheel print, computed in WebAssembly in the tab; text with an unpaired surrogate is refused in plain text; 0 requests after the page loads. Not yet: scripts other than Devanagari (Tamil is planned for v0.2 as a design note only, Bengali later), a published package or wheel, 8 GB figures (not measured on 8 GB).

Browser requirements

No storage, no network: this page keeps nothing and sends nothing after it has loaded. Without the two file features you can still paste text.

What this demo does not do yet

Text

Paste Devanagari or Hindi text, drop a .txt file on the box below, choose one with the file picker, or load a synthetic sample. Samples are made up; none is a real document.

Drop a plain-text file here, or choose one

The file is read in this tab only: UTF-8 text up to 1 MiB, never uploaded. A file that is not valid UTF-8, or is over 1 MiB, is refused before anything is validated.

0 of 1,048,576 bytes

Document score

—

Each flagged cluster costs the largest weight among its issues; the score is 1 − penalty × 1000 ÷ clusters, in parts per million, shown with four decimals (truncated, never rounded) by integer arithmetic. 1.0000 is clean; empty text scores 1.0000.

Heatmap

One row per line: a strip for the line, then every grapheme cluster of the line. A cluster is shaded by the sum of the weights of the rules that fired on it; the strip uses the same sum over the whole line. Hover a shaded cluster to see the rules. Invisible, control and private-use code points are shown as ⟨U+XXXX⟩.

  • clean: no rule fired
  • low: weight sum 1 to 399
  • mid: 400 to 799
  • high: 800 and above

The sixteen rules

The stable wire names from issue_kinds(), each with its weight from weights/devanagari.toml and how many times it fired on this text.

    Lines

    One row per line (LF, CRLF or a bare CR ends a line). The score is the same formula applied to that line alone. Invisible or private-use code points are shown as ⟨U+XXXX⟩.

    Line Text Clusters Flagged Score Issues
    Grapheme clusters

    UAX #29 extended grapheme clusters the validator judged, with byte offsets into the UTF-8 input. Flagged clusters name the rules that fired.

    # Cluster Code points Bytes Rules
    Canonical JSON report

    Exactly what validate() returned, which is byte for byte what glyphsift check - prints for the same text.

    
            

    Module

    Report spec
    …
    Module id
    … (first 16 hex of sha256 over glyphsift.js + glyphsift_bg.wasm)
    validate() calls
    0 (a refused input is not counted: the validator is not run for it)