Detection and conversion

DetectCopy Anchor Link

POST /v1/text/detect returns encoding evidence without changing text.

{ "text": "မြန်မာစာ" }

Possible encodings are unicode, zawgyi, mixed, unknown, and non_myanmar. unknown is a legitimate answer for ambiguous shared sequences. mixed means separate Myanmar runs have opposing evidence; a switch within one uninterrupted run cannot be reliably located. Detection does not identify the language or establish orthographic validity.

ConvertCopy Anchor Link

POST /v1/text/convert requires both from and to to be unicode or zawgyi.

{ "text": "မဂၤလာပါ", "from": "zawgyi", "to": "unicode" }

The output is မင်္ဂလာပါ. Equal directions produce an exact no-op. Do not pass unknown or mixed as a known source encoding. Preserve original text and the chosen direction. Unicode should be your canonical storage format.

Guard against mixed or mislabelled inputCopy Anchor Link

For pasted text, enable validateSource explicitly:

{
  "text": "ေၾကညာစာတမ္း",
  "from": "zawgyi",
  "to": "unicode",
  "validateSource": true
}

Before a direction-changing conversion, this rejects mixed evidence (encoding_mixed), evidence conflicting with from (encoding_mismatch), or any uncertain Myanmar segment (encoding_uncertain). These are HTTP 400 problems, with no partial converted output. Use /v1/text/detect to inspect the original text and segment offsets; do not repeatedly try both directions.

Omitting the flag or setting it to false keeps the original unchecked behavior. Equal directions still return the input unchanged. The flag is a conservative guard, not automatic mixed-text repair: detection can miss switches inside one Myanmar run and cannot guarantee spelling or linguistic correctness. Preserve the original and review uncertain input before deliberately opting out.