Recorded run / Resilience test
degraded replay
An inspectable recording of the evidence-to-scenario workflow. The plan, specialist calls, returned evidence, evaluation, limitations, and linked digests below come from the verified bundle; they were not reconstructed for presentation.
Read this as software evidence, not as live market analysis. The inputs are deterministic fixtures, the planner is rule-based, and the signing key is a disclosed demo key. No production bank data, public inference service, or trained Atticus weights are represented.
This run failed partway through and kept going. demo.force_fail failed: forced specialist outage for degraded replay capture. The workflow completed in a degraded state rather than collapsing, and the trace below records exactly where it broke. This recording is published because a system that only shows its successes is not evidence of anything.
Execution trace
Read from top to bottom. Tool activity is visually separated from state transitions; a failed specialist call is marked in text and color.
- receivedtask_receivedTask accepted.
- planningplan_createdCreated bounded plan with 4 step(s).
- planningpolicy_decisionAllowed by deterministic catalog policy
- planningpolicy_decisionAllowed by deterministic catalog policy
- planningpolicy_decisionAllowed by deterministic catalog policy
- planningpolicy_decisionAllowed by deterministic catalog policy
- executingtool_startedInvoking atlas.research_snapshot.
- executingtool_completedAtlas returned 3 point-in-time observations.
- executingtool_startedInvoking demo.force_fail.
- executingtool_faileddemo.force_fail failed: forced specialist outage for degraded replay capture
- executingtool_startedInvoking fedlens.compare_latest.
- executingtool_completedLanguage became less hawkish in the deterministic lexicon.
- executingtool_startedInvoking balancelab.run_scenario.
- executingtool_completedSynthetic annual NII change: $15.81 million; curve slope change: 50 bps.
- evaluatingevaluation_startedSubmitting observable trace and evidence to EvalForge.
- evaluatingworkflow_linkedLinked Atlas, FedLens, BalanceLab, report, and evaluation digests.
- degradedtask_completedWorkflow finished with EvalForge score 1.000.
Evidence lineage
What each specialist contributed, and the digests that tie the artifacts together.
| artifact | present | evidence ids | digest |
|---|---|---|---|
| atlas | yes | atlas-CPI_YOY-2026-06-30atlas-UST_2Y-2026-07-24atlas-UST_10Y-2026-07-24 | — |
| balancelab | yes | balancelab-bear-steepener | sha256:945d82945e14a… |
| evaluation | yes | — | sha256:a19025c733ae2… |
| fedlens | yes | fedlens-FOMC-2026-06-FIXTURE | — |
| report | yes | — | sha256:31a82e3ebd47f… |
Evidence
Evidence identifiers carried by the run record.
atlas-CPI_YOY-2026-06-30atlas-UST_2Y-2026-07-24atlas-UST_10Y-2026-07-24fedlens-FOMC-2026-06-FIXTUREbalancelab-bear-steepener
Limitations
Recorded by the run itself, not added afterwards.
- Macro, market, and Fed inputs are synthetic fixtures for local development.
- BalanceLab uses a simplified educational repricing model, not production bank data.
- The deterministic planner stands in for Atticus Core until the model bake-off.
- Linked workflow is prototype maturity; signed replay packaging is DRL-019.
- demo.force_fail: forced specialist outage for degraded replay capture
Provenance
Build-time verification binds the display to the recorded artifacts. It does not establish a production signing identity.
| code revision | drl-019-fixture |
|---|---|
| executed at | 2026-07-30T00:00:00+00:00 |
| configuration | sha256:a12309d16239748e828a6b7d2… |
| signature | verified at build time |
The manifest signature and every artifact digest were verified before this page was generated; an unverified bundle is not rendered at all. The signing key is a demo key, not a production signing identity — treat the signature as structural integrity, not as an attestation of trust.