News
v1.0.0March 1, 2026

TraceCore 1.0 ships

v1.0 is the first stable Deterministic Episode Runtime release. The spec is frozen, the tracecorecommand is now the default entry point, and CI teams get wall-clock telemetry, batch execution, and live metrics out of the box.


Highlights

  • Spec v1.0: every artifact declares spec_version, runtime_identity, task hashes, and wall_clock_elapsed_s.
  • tracecore CLI: first-class entry point alongside the legacy agent-bench alias.
  • Batch execution: run suites concurrently with spawn-isolated workers and enforced timeouts.
  • Metrics surface: CLI, REST, and dashboard views for success rate, budget P50/P95, and MTTR.

Spec contracts & artifacts

The canonical spec bundle (spec/tracecore-spec-v1.0.md) promotes all provisional language from v0.1 to normative MUSTs. Section 6 defines batch execution invariants, while the artifact schema now requires wall_clock_elapsed_s. Strict-spec validation refuses to report success unless that metadata is present.

artifact excerpt
{"spec_version": "tracecore-spec-v1.0",
"runtime_identity": {
  "name": "tracecore",
  "version": "1.0.0",
  "git_sha": "..."
},
"wall_clock_elapsed_s": 4.217,
"task_hash": "sha256:..."}

Batch execution & isolation

tracecore run batch now orchestrates parallel jobs using the multiprocessing spawn context. Each worker inherits a clean environment and gets a per-job wall-clock timeout. When the timeout hits, the parent terminates the worker and records a structured failure_type=timeout artifact instead of hanging the suite.

terminal
tracecore run batch --workers 4 --strict-spec   --batch-file suites/ops.json   --timeout 120

Batch summaries report pass/fail counts plus P50/P95 wall-clock stats so operators can spot regressions without spelunking logs.

Metrics everywhere

Reproducibility, budget utilization, and MTTR finally have first-class endpoints. Whether you prefer CLI tables, JSON for Grafana, or the dashboard /metrics view, all three tap the same data model.

terminal
tracecore runs metrics --format table
tracecore runs metrics --task log_stream_monitor@1 --format json
GET /api/metrics?task=log_stream_monitor@1&agent=agents/toy_agent.py

Upgrade guidance

Upgrade with pip install --upgrade tracecore, then verify the spec pairing via tracecore version. There are no breaking CLI flags, but CI assertions referencing tracecore-spec-v0.1 should be updated. Artifacts emitted by v0.9.x remain readable; the runner falls back to the older schema when replaying historical runs.

What's next

With the spec frozen and telemetry in place, the roadmap shifts to trace diff tooling, evidence bundle signing, and richer taxonomy UX. Expect tracecore diff, Cosign-backed trust bundles, and dashboard views that combine termination reason and failure type by default.