Skip to content

v0.4.2

Choose a tag to compare

View all tags
@odvcencio odvcencio released this 20 Mar 22:55
· 203 commits to main since this release
v0.4.2
09a974f

v0.4.2

Engine

  • per_fact keyword — expert rules marked per_fact fire once per matching fact instead of short-circuiting on the first. Instance-keyed truth maintenance tracks supports per rule + fact key pair.
  • Temporal windows — facts carry __asserted_at and __age_seconds wall-clock metadata. Sessions expose __now and current_round. Injectable clock via expert.Options.Now for deterministic testing.
  • HTTP middleware — arbiter.Middleware(compiled, buildContext, next) evaluates governed rules per HTTP request. DefaultHTTPContext auto-extracts method, path, headers, and query params with type coercion.
  • Decision diff — arbiter diff base.arb candidate.arb --data-file contexts.json compares two rulesets against the same data. Reports added, removed, and changed rule matches.
  • Decision replay — arbiter replay rules.arb --audit decisions.jsonl re-evaluates historical decisions against new rules.
  • Terraform/HCL fact source — parse .tf files via gotreesitter's HCL grammar or read terraform plan -json output. Resources, data sources, modules, variables, outputs, and locals become typed facts.
  • Bidirectional fact sources — Saver interface mirrors Loader. JSON and JSONL writers ship built-in.
  • Multi-arbiter workflow engine — topologically sorted chained arbiters with persistent sessions, compiled outcome filters, and chain:// fact routing.
  • Durable Runner — reliable source polling with retry/backoff/last-known-good, durable sink delivery with JSONL journal and crash recovery. Source/sink health metadata injected into evaluation context.

Platform

  • Arbiter Cloud — managed control plane with bundle lifecycle, HTTP evaluation API, browser dashboard, API key auth, continuous runners, and durable webhook delivery.
  • Dashboard — syntax-highlighted source viewer, trace explorer, decision diff, API key management, kill switch panel.
  • Agent keys — scoped keys for AI agent authentication with rate limits and permissions.

Infrastructure

  • GitHub Actions CI — test, vet, build CLI, build agent, grammar check on every push/PR.
  • Apache 2.0 LICENSE file added.
  • Status section in README documenting project maturity and stability boundaries.
  • Grammar artifacts regenerated — 89 productions covering all v0.4.x language features.

Performance

Profile Throughput p50 Latency
Sidecar (50m CPU, 1 pod) 90 evals/sec 836μs
Production (1 core, 3 pods) 41,088 evals/sec 1.9ms
Expert inference (3 pods) 315 sessions/sec 2.2ms
Assets 2
Loading