Agent 14 · Keep the agent team accountable
Steward
Keeps Rever and its connected-system dependencies reliable, explains the financial impact of operational failures, and makes sure health issues are repaired and verified.
Read-only connections · Zero-day scan · 100% coverage · Your team keeps the final say
01When it runs
What starts the work
Eve dispatches Steward when one of these events arrives, checks readiness first, and records why the work was allocated.
- A failed or delayed sync, a stalled agent run, a missing approval progression, a stale source, a changed access or a recurring system warning.
- A specialist reports that required data or functionality is unavailable or incomplete.
02What it checks
The procedures Steward performs
Every procedure has a stable identifier in the approved SOP, an expected result and required evidence. Each check records its result, its supporting evidence and anything it could not assess.
- 01
Inventories available log sources and confirms permissions, coverage periods, timestamps and expected refresh, recording inaccessible logs and the resulting monitoring limitation.
- 02
Inspects Rever and connected-system logs for failed jobs, dropped or repeated events, stale synchronisation, unresolved errors, unusual access changes and unprocessed warnings.
- 03
Reviews deviations in user behaviour where logs support it, and refers suspected financial control abuse to Sentinel rather than treating it as technical noise.
- 04
Detects work stuck in dispatch, waiting on data, on a human action, on an execution receipt or on a rerun, and identifies the responsible dependency and the affected procedure.
- 05
Prioritises by time pending, recurrence, amount at risk, business deadline and downstream impact. A quiet, long-pending failure can outrank a recent noisy warning.
- 06
Explains consequences in finance terms, such as incomplete billing, a stale cash forecast, unverified inventory or a delayed close, with the affected population.
- 07
Creates and maintains the operational item in Health logs, with a linked system finding and any human repair task routed through the Inbox.
- 08
Performs only approved safe retries or maintenance within delegated scope. Changes to credentials, access, accounting mappings or source records need their proper owners and controls.
- 09
After repair, verifies source freshness and completion, then requests the affected agent's rerun. Technical recovery alone does not prove the financial procedure is complete.
- 10
Publishes recurring causes, monitoring blind spots and preventive maintenance recommendations to Eve and the responsible users.
Your agreed scope identifies the procedures available for your connected systems, the evidence they need and any coverage limits. Missing sources and blocked checks stay visible rather than counting as a clean pass.
03Who it works with
One lead per finding. Named support.
Every agent can report a dependency failure. Eve reprioritises financial work, the zero-day scan rechecks landing, and the originating agent rechecks its procedure.
Sentinel receives potential user or control abuse, and Auditor tests whether detection, escalation and recovery worked.
04What it files
Workbooks you can download. Reports you can read.
Each workbook reproduces the run: population, formulas, checks, exceptions and evidence references, linked back to the workpaper in Reconciliations and the originals in Records.
- Log source and monitoring coverage
- Open health incidents and pendency
- Affected agents, ledgers and deadlines
- Repair, retry and verification history
- Recurring causes and monitoring gaps
- Platform and connector health report
- Financially prioritised blocker digest
- Log coverage report
- Incident recovery review
05What it raises, what stays with you
Findings for the agent. Decisions for your team.
A stale source, a stalled duty, an ignored system warning, missing log coverage, an access failure or an incomplete recovery.
Health issues appear as linked system findings, with no monetary recovery attached.
Repair ownership, access requests and consequential operational changes go from the system finding to your Inbox and link back to Health logs. Routine permitted retries proceed under the approved policy.
Staleness, run duration, retry limits, pendency bands, recurring-error frequency, log lookback and escalation priorities.
The repaired service, complete data and a successful rerun by the affected agents are verified before the health log and the linked finding close together. Residual financial exceptions stay with their specialists.
06What good looks like
What a completed result looks like
The observable outcomes Steward has to produce before the duty counts as complete on your books.
- A source without accessible user logs is reported as a monitoring gap, not as a clean user-behaviour review.
- A repaired connector stays financially incomplete until the affected procedures rerun.
- Every health item is discoverable under Settings and shares one linked finding and one human task trail.
Where Steward works
Bank reconciliation
Watches bank-feed freshness and failed syncs, so a stale statement becomes a health item rather than a false clean reconciliation.
Balance reconciliation
Tracks outstanding confirmation requests and stalled reconciliations, and shows what a delay does to the close.
Audit readiness
Records source coverage and monitoring gaps, so the evidence pack states what could not be checked.
Continuous monitoring
Watches failed syncs, stale sources and stalled work, prioritised by the finance work they block.
See Steward on your own books
Book a demo, or get a free Proof of Value on a limited period of your data.