Operator Workflows¶
A governed Runtime operation starts with one initialized workspace and ends with a persisted job, run, attempt, causal artifact graph, and terminal result. Readiness, execution, result resolution, inspection, replay and recovery are separate decisions.
flowchart TD
init["initialize workspace"] --> ready["check operation + profile"]
ready --> submit["submit typed operation"]
submit --> terminal{"terminal job?"}
terminal -- no --> follow["follow with timeout or cancel"]
follow --> terminal
terminal -- yes --> result["resolve result"]
result --> inspect["inspect run + attempt"]
inspect --> replay["replay and compare"]
inspect --> backup["quiesce and back up"]
Initialize and check the exact capability¶
bijux-canon-runtime init --workspace ./canon-workspace --json
export BIJUX_CANON_RUNTIME_WORKING_ROOT=./canon-workspace
bijux-canon-runtime v2 capabilities
bijux-canon-runtime v2 ready \
--operation run \
--profile offline-lexical
Liveness proves only that a process can answer. Readiness evaluates the named operation and profile against the effective configuration. Lexical readiness does not require a model; dense and hybrid readiness additionally validates the locked model, active generation, backend and vector dimension.
Submit and wait deliberately¶
bijux-canon-runtime v2 run \
"What evidence does this corpus support?" \
--source-directory ./documents \
--profile offline-lexical \
--wait \
--wait-timeout-seconds 30
Submission without --wait returns immediately. --wait and v2 status
--follow use worker notification with an explicit timeout rather than an
unbounded polling loop. A timeout does not relabel durable work as failed; use
the returned job identity to inspect or cancel it.
Resolve and inspect¶
bijux-canon-runtime v2 result JOB_ID
bijux-canon-runtime v2 inspect RUN_ID \
--attempt-id ATTEMPT_ID \
--limit 20
Accept a result only when the expected configuration, source archive, corpus, index, model lock or explicit lexical absence, retrieval, citations, claim graph, agent decisions, budgets, and terminal attempt all resolve through the causal graph. Inspection fails closed when a new-format run has a missing or inconsistent edge, locator, byte range, or payload digest.
Default inspection is intentionally small. Page a collection explicitly and
use v2 artifact-payload for a bounded byte range rather than requesting an
unbounded diagnostic dump.
Replay and compare¶
bijux-canon-runtime v2 replay RUN_ID \
--source-attempt-id ATTEMPT_ID \
--network-policy disabled \
--wait
bijux-canon-runtime v2 compare RUN_ID RUN_ID \
--baseline-attempt-id ATTEMPT_ID \
--candidate-attempt-id REPLAY_ATTEMPT_ID \
--dimension outcome \
--dimension claims \
--dimension citations
Replay is a new attempt under the retained request and declared network policy, not deserialization of the old result. Deterministic components require exact identity; graded components retain their declared comparison disposition. A missing source, model, index or artifact fails rather than being reconstructed.
Cancel safely¶
Cancellation is durable and race-aware. Inspect the terminal job and attempt; do not infer cancellation merely because a client stopped waiting. Retrying an operation requires its idempotency and retained-state rules, not a newly invented successful record.
Back up and restore¶
bijux-canon-runtime v2 backup nightly-2026-08-25
bijux-canon-runtime v2 restore BACKUP_GENERATION ./restored-workspace
Backup requires a quiescent workspace and authenticates the DuckDB authority, CAS inventory, workspace controls, indexes, replay state and workspace-owned model state. Restore verifies every governed byte before activating an absent destination. Use this path for relocation; direct directory copies do not rewrite governed machine-local locations.