Talk to salesOpen workspace

Changelog

What changed.
What stayed the same.

Release, format and evidence changes, including the status of each entry. A version number is not a production approval, and an unreleased change is labelled as unreleased.

Review the evidenceService status
release posture
FormatSAAR v7 · SAST v3versioned
RecoveryExact source byteschecked
WindowsSerial code pathqualification per evaluation
Public artifactsNot publishedevaluation access

Entries

Dated changes, with scope.

The changelog is a record of product and evidence changes, not a substitute for the public evidence page or the signed commercial terms.

1.3.0Locally verified · unreleased17 October 2026

Engine & native runtime

  • Strand mode 34: sparse-residual context mixing closes the sparse-image virtualization gap. A reversible zero-run fold feeds strong context mixing, with byte-exact cross-decoding between the TypeScript encoder and native decoder.
  • Context-mixing model progressed v2 → v6: bit-history state maps, richer text contexts, long-order and indirect-model contexts behind frame version bits, with strict profile selection.
  • Native fast tier: stream I/O overlaps compute and match-chain walks prefetch candidates. Output stays byte-identical, with higher throughput on the same machine.
  • Reversible domain modes recorded and verified byte-exact: PNG (27), signal (28), ZIP (29), WAV (30), raster (32) and PDF (33); SQLite and archive samples measured.
  • Robustness: bounded-memory hardening rejects pathological PPM inputs instead of exhausting memory.
  • A native Linux x64 runtime is built and tested beside the macOS reference (codec, native and app suites green).
  • Hosted large-file delivery: resumable verified part streams with checkpoint continuation, admitting files up to 5 TiB.
  • Published evaluation suites re-measured on 18 October 2026 with the current revision under a uniform whole-command CLI method; the prior fast chart mixed in-process Saar timings with CLI competitor timings and has been replaced.
  • All 31 workload classes now carry a dated measurement (30 wins, 1 loss shown in place); real public samples replaced generated fixtures for six previously unsampled classes.
  • Point-cloud mode 35: LAS point records are transposed into delta-coded numeric columns before PPM coding; both real LiDAR samples now beat xz -9 (49,566 vs 53,228 and 68,640 vs 89,124), byte-exact across TypeScript and native decoders.
  • Structured-text mode 36: hex runs are packed two characters per byte and a fixed JSON/OCI/HTTP vocabulary is substituted before coding; the cloud manifest now beats brotli -q11 (584 vs 588), byte-exact across TypeScript and native decoders.
Excluded from this entry
  • The earlier lossy JSON prototype figures (−20% to −33% on World Bank JSON) are retracted: they were not byte-reversible. The exact record-engine payoff measured is ~6% on tabular CSV.
  • No performance, security or compliance claim beyond the dated records in evidence/ is made.
1.3.0Unreleased candidate13 October 2026

TypeScript codec & CLI

  • SAST v3 checked streams are the default; SAST v2 stays readable and explicitly writable for compatibility.
  • Native streams support deterministic ordered POSIX worker batches via --threads N, byte-identical across tested thread counts.
  • Byte-exact sparse-byte mode 21 cross-decodes between native and TypeScript and rejects malformed frames.
  • Numeric-plane mode 22 adds reversible fixed-width routing for tensors, telemetry and columnar numeric bytes.
  • The TypeScript SDK adds a transport-neutral binary shard protocol and completion-order-independent SAST3 assembly.
  • The range coder removes per-symbol BigInt work; the PPM selector evaluates order 5 where order 4 justifies the search.
  • Web and portal share one component, data and style workspace and build through Turborepo.
Excluded from this entry
  • No supported GPU, FPGA, quantum-compression, native Zstandard or native LZ4 implementation was added.
  • Portal authentication, compression jobs, usage and billing are not integrated.
  • Universal performance leadership, TB/s, Windows execution and independent security review are not established.
1.2.0Locally tested13 October 2026

Native C SDK & CLI

  • Faster native encode and decode path with the SAAR v7 envelope.
  • Mode 20 ratio candidate and byte-exact sparse-byte mode 21.
  • Native decode optimizations with a recorded round-22 acceptance run.
  • CLI I/O failure handling verified.
1.0.0Private workspace13 October 2026

Web & portal applications

  • ESM-first npm workspaces with Turborepo.
  • Separated portal and marketing applications that share one component, data and style workspace.
  • Recorded-evidence benchmarks, a measured-impact section and the interchangeable-brick architecture.
  • Server-only live comparison in the portal.
Engine formatsActive8–11 October 2026

Engine formats

  • Object format progressed SAAR v3 Essence Lattice → v4 Strand Fold → v5 → v6 stored/general envelope → v7 compact binary strand.
  • Byte-exact recovery is enforced by the SHA-256 envelope and stream transcript checks.