Changed

• All payloads are records, not pre-serialised strings. ClientEvent.payload, ClientIssue.payload, ClientMetaData.payload and ExtensionStat.payload changed from an optional string to an optional map of primitives — Record<string, boolean | string | number> in the generated TypeScript (Avro: ["null", {"type": "map", "values": ["boolean", "string", "double"]}], keeping the null-first optional convention). Producers no longer JSON.stringify every payload before buffering it, and consumers read payload fields directly instead of parsing a nested JSON document.

  Why (measured, Node 22, realistic samples with 35–40 payloads): a pre-stringified payload gets every quote escaped when the whole sample serialises, inflating the payload bytes by ~12% on the wire; serialising the sample is ~18% faster end to end (0.37ms → 0.30ms per sample) and the events portion alone ~54% faster (0.042ms → 0.020ms), because each payload was previously walked twice — once by its own JSON.stringify at buffering time (30–40 calls per sample, now zero, along with their intermediate-string garbage) and once more character-by-character for escaping. Server-side decode saves ~3% (one parse instead of 1 + N) and payload fields become directly queryable without a second parse.

  Codec impact: on the protobuf wire the payload still travels as a JSON string (proto3 cannot express a union-valued map — same convention as attachments); the protobuf codec converts it through its JSON-fields mechanism, keyed on the field name, so all four payload fields are covered and protobuf field numbering is unchanged. The JSON codec needs no change: events, issues, meta items and extension stats are value lists written whole, and object values were already handled generically. The TypeScript generator learned to render union-valued Avro maps for this.

Source: ObserveRTC schema changelog. Date verified from the schema version commit.