Reconcile
Reconcile compares what your systems actually do with what your policies say they
should do. It also grades how mature each business domain is. Open it with
Reconcile, the first entry in the bottom navigation (/ai_center). The page is
titled Business domain value reconciler, under the line CODE VS POLICY
RECONCILIATION.
Two views of one estate
A switch under the header chooses who the page is laid out for:
- Executive briefing covers the maturity verdict, the relationships between domains, provenance, and the rules you bundle for simulation. This is the default.
- Operator deep dive covers the measured gauges, the live agent activity, and the detailed rules list.
Old links still work. /rules and /ai_center?tab=rules open Operator deep dive.
Maturity & rules reconciliation
The command bar at the top of Executive briefing gives two readouts side by side:
- AI maturity: a ring showing the mean level across the domains, as L{level} / {max} on a 1–5 scale. It shows Not assessed yet before any assessment.
- Rules state: a bar that splits the rules by verdict, with a legend giving each count and state.
The {count} invariants badge counts the rules. The domain pills under the bar, starting with All domains ({count}), narrow the rules list further down the page. They don’t change the two readouts. Tap the selected pill again to clear it.
Rule verdicts
Every rule carries one verdict. Each verdict has its own glyph and word as well as its own colour, so it never depends on colour alone.
| Verdict | Glyph | Meaning |
|---|---|---|
| Aligned | circled check | Aligned with policy. |
| Drift | octagon | Drifted: needs reconciliation. |
| Shadow | circled exclamation mark | Undocumented. |
| Missing | warning triangle | Missing: needs reconciliation. |
| Unrated | question mark | The agent returned no verdict. The rule is not yet reconciled. |
Executive briefing
The maturity verdict
The banner states the result in one line:
- Analysing your data domains… while the assessment runs.
- No data domains have been analysed yet. when the assessment answered with nothing.
- {n} domains analyzed · {n} findings across all domains · {n} measures to take on success, followed by Next: {measure} and a button for it. Plan remediation opens Compliance Remediation. Review trends opens a short explanation.
- Posture is unavailable — the reported reason is below. when the assessment failed. The server’s reason follows (Reported reason, in English:), with Retry.
- Could not refresh — this is the last result and may be out of date. when a refresh failed but an earlier result is still shown.
The banner carries an AI-Generated tag when a measure was written by the model.
If no data source is attached, the banner is replaced by Nothing is connected yet, the instance’s own reply, and Connect a data source, which opens Connections.
The findings and measures themselves are listed on Twins.
Sections
| Section | What it holds |
|---|---|
| Cross-domain relational matrix | Starts collapsed; tap to expand. A table of each domain’s Grade, Quality, PII risk, Lineage, Freshness, Policy and Cost, then the relationships the agent reported, word for word. A chord ring draws the domains and their links. Tap a domain on the ring to scope the sections below it to that domain, and tap it again to clear. |
| Object provenance | The quality and risk checks recorded for each domain. |
| Data communities | The communities found in your data beside your source inventory, grouped by kind of source. The notice THESE TWO LISTS ARE NOT JOINED is literal: the two lists are separate readings. |
| Captured media | What this cockpit captured (photos, clips, voice memos) and what the platform said about each. |
| Rules reconciliation engine | The rules, with the checkboxes that bundle them for Simulate. See below. |
| Readiness indicators | What the agent rated on each domain, on its own 1–5 scale, category by category. |
A cell with no measurement says not measured. It never shows zero. Several sections also tell a failed request apart from an empty answer. For example, THE SOURCE LIST FAILED TO LOAD is a failed request, while NO SOURCES ARE CONFIGURED means the list was read and is empty. NONE REPORTED means the agent sent no indicators for a category. NOT MEASURED HERE means FACE does not compute that category.
Set a rule’s enforcement and send rules to Simulate
The drawer also holds the Reconcile inspector. It is a read-only comparison of Actual — implementation against Intended — policy, with where the rule is implemented and what it is grounded in. Nothing to compare yet means FACE captured neither an implementation nor a policy for the rule.
Operator deep dive
- Domain posture console shows six gauges: Quality, PII risk, Lineage, Freshness, Policy and Cost. Each is the mean of the domains that measured it. A gauge with no measurements reads not measured.
- Live agent activity shows the steps the agents emitted, newest first. You can pause, resume or clear the buffer. LIVE or PAUSED shows its state. The activity stream is not connected. means there is no stream. No agent steps yet on this connection. means the stream is connected and quiet.
- Rules Recon lists every rule:
- Per-domain cards with TOTAL RULES and a bar by verdict. Tap a card to filter by that domain.
- The Logic Discovery Stream chips (All, Aligned, Drift, Shadow, Missing, Unrated) filter by verdict.
- The verdict banner. When rules have drifted or are missing, Review what drifted jumps to them. Review the undocumented does the same for shadow rules.
- Tick rules and select Send to Simulate to ground our hypothesis to use them in a simulation.
While the agent reads, the list shows AGENT ORCHESTRATION IN PROGRESS. No business rules have been discovered yet. means the extraction answered with no rules. Failed to extract rules: {error} means it could not run, and Retry asks again.
RPCs behind this page
PostureService.GetHealthScore.RulesService.AnalyzeCode,UpdateRulePolicy.LineageService.GetSourceActivity.ConfigService.ListConnections.ActivityService.SubscribeAgentActivity(server stream).
See the API reference.