Skip to content

What's New

Curated highlights of notable GCF releases, newest first. For the current version of every package see Implementations; for the full notes on any release see the GitHub releases.

The spec has grown additively since v3.0 with no breaking changes: a v3.0 decoder ignores anything it does not recognize.

Current ecosystem: Spec v3.4.1 · SDKs v2.4.0 (Go v1.5.1) · gcf-proxy v0.11.3 · tree-sitter-gcf v1.3.3 · 204 conformance fixtures.

v3.4.1 — Graph delta verification

2026-07-13

  • A graph delta's ## added lines now carry a trailing distance field, so a consumer can reconstruct the new snapshot and verify new_root end to end (Spec §10.1, §10.4).
  • Shipped across all six SDKs (Go v1.5.0, the other five v2.4.0), tree-sitter-gcf v1.3.3, and gcf-proxy v0.11.3.
  • Release notes

v3.4.0 — Labeled streaming trailer + decoder hardening

2026-07-13

  • Optional labeled counts form for the graph streaming trailer (Spec §8.4.1), a producer-side comprehension aid. The default positional form is unchanged.
  • Decoder hardening and a conformance coverage matrix across all implementations.
  • Release notes

v3.3.0 — Delta encoding for the generic profile

2026-07-12

  • Delta encoding, previously graph-only, now works for the generic profile (Spec §10a): a keyed diff with an @id identity column and ## added / ## changed / ## removed sections, content-addressed by pack_root. Delta is now a both-profile capability. See the Delta guide.
  • Release notes

v3.2.0 — Nested-object flattening

2026-06-23

  • Nested objects flatten into > path columns (Spec §7.4.6), so records with sub-objects stay positional instead of falling back to a less efficient encoding.
  • Release notes

v3.1.0 — Optional graph tool field

2026-06-14

v3.0.0 — Inline schema encoding

2026-06-13

  • The v3 generic profile foundation: inline object schemas encode nested objects positionally, shared array schemas omit repeated field headers.
  • Release notes

This page tracks highlights only, updated at each release. For per-commit history, see the GitHub releases and each package's CHANGELOG.

100% comprehension. 71% fewer tokens. 2,500+ LLM evaluations.