Action Results
See what each governed action produced, whether it stayed resolved, and whether external writes matched their preview
Last updated 11 Oct 2026Review · article 4 of 5
Action Results shows what the agent shipped, whether it stayed resolved, and whether each external write landed as previewed. Governed actions record an outcome when they run, and this page reads those outcomes. Open Action results in the sidebar, under the Governance group (or go to /workspace/action-results).
The page is read-only: apart from the CSV export, nothing here changes an action or its outcome. Anyone with governance read access, viewers included, can open it. Until the first outcome is recorded it shows No outcomes recorded yet and a Connect an ad platform button.
Reconciliation applies only to external writes, and live execution against external ad platforms is connector-gated and not enabled by default.
The Window
The date range selector in the header sets the window: a preset (last 7, 14, 30, 60, or 90 days; the default is 7) or a custom range. The figures and the by-action table follow it.
The Summary
One figure leads the page:
| Figure | What it measures |
|---|---|
| Stayed resolved | The share of outcomes in the window that succeeded |
Below it, Decision mix splits the window's decisions into Auto-resolved, Escalated, and Blocked, and Where they end up counts outcomes as Succeeded, Failed, and Executed with approval drift — a write that completed although the capability it dispatched was not the one the approver bound.
By Action
Under the summary, a search box filters the by-action table: one row per action with its number of runs ("fires"), Stick rate (the same measure as Stayed resolved), and a 7-day trend. A flag marks an action whose stick rate fell by a percentage point or more against the previous window of the same length. Select a row to open the action's console.
An Action's Console
The console (/workspace/action-results/action/[name]) shows one action over the window you came from: its stick rate, a 7-day stick-rate trend, and a watchlist note when it has been flagged. A banner summarizes it as Holding, Near threshold, Below target, or Stick rate falling. Stick rate counts as healthy at 95% or above and near threshold from 88%.
Per-Execution Results
The list of individual results stays closed until you ask for it. It opens when you follow a link to one result, turn on Escalated by verifier in the header to see only outcomes the independent verifier escalated for review, or choose a scope. Each row shows:
- Outcome — a badge for the outcome type
- Reconciliation — for external writes, whether the platform's state matched the preview: Reconciled, Drift detected, Read-back failed, Write failed, or Not applicable
- Action, and Audience, Areas, Population, and Metric where the outcome carries them
- Recap — Pass, Partial, or Fail, comparing the action manifest's expected values with the actual ones
- Actual and Date
Expand a row for the breakdown, or use View Detail. Execution decisions and Approval decision open the matching records in the Audit log. Export CSV downloads the results matching the current filters as outcomes-export.csv.
The Result Page
A result's page (/workspace/action-results/[id]) adds:
- Metric Comparison — expected against actual for each metric the action declared
- Capability drift — when the approved operation and entity kind differ from what was dispatched
- Reconciliation — the Pre-write state, Expected post-state, and Actual post-state side by side, plus each field that drifted with its before and after values. If the platform could not be read back, the actual card says "Read-back unavailable" rather than showing a stand-in
- Independent verdict — Verified, Drifted, Failed, or Inconclusive, with a rationale and reason codes. An escalated verdict links to all escalated outcomes
- Decision Context — the policy decision, the Rules evaluated (each links to its rule), and any Knowledge used nearby (active, unexpired records matched by outcome, decision, rule or reason code; if they cannot be loaded the page says "Related records could not be loaded" instead of showing none)
- Pre-state and Post-state snapshots behind View raw state, and a list of what was measured
Narrowing by Scope
The scope bar narrows the per-execution list, its count, the recaps, and the CSV export by Platform and working scope. The summary figures, the trend, and the by-action table cover every result in the date range, whatever the scope, and the page says so. Advertiser, brand, market, and media channel do not narrow it because Cesteral records none of them.
Results the scope cannot place — no platform recorded, or one that cannot be matched — are left out of the scope view. The bar says how many, and Show them lists those instead; Back to your scope returns.
Related articles
Still stuck? Browse the FAQ or contact us.