Decision Engine — AI Enterprise OS

Executive summary

1 critical decision needs you today. The most urgent is escalate Agent credential rotation — blocked at Build.

1 critical item — Agent credential rotation. Escalate to Rhea Kastellan, who owns the outcome. The block sits in engineering — the decision needed is whether to wait, re-route around the stage, or stand the workflow down.
7 of 8 in-flight workflows are carrying an impediment, 1 of them serious. Worst is Agent credential rotation at 55 — Blocked at Build — Four standing credentials belong to a vendor whose API caps rotation at 10 calls/hour. Rotation cannot complete inside the maintenance window until that cap is lifted..
2 of 8 departments are below target. Research at 80 is the furthest down — Time to result at 74% of target.
Capacity is unevenly held: 12 AI employees are at capacity while 9 have room, at a fleet mean of 60%. Operations runs hottest at 68%.
19 workflow gates and 22 executive approvals are outstanding. These are two separate queues — gates pause execution, desk approvals do not.

Where to start

Escalate to Rhea Kastellan, who owns the outcome. The block sits in engineering — the decision needed is whether to wait, re-route around the stage, or stand the workflow down. This is also raising an active alert. (critical priority, high confidence)

26

Recommendations

1 critical

8

Active alerts

conditions true now

70

Risk index

severe

89

Operating health

2 below target

1

Workflows running

1 blocked

19

Open gates

22 on desks

60%

Fleet load

9 idle · 12 at capacity

99

Graph health

515 entities

Composed from 1 recommendation shown below and the rollups each module publishes. This summary cannot assert anything the evidence on this page does not support — it is generated from the same findings, not written alongside them.

Active alerts8

pressure 90/100

1 critical condition of 8 active. Most urgent: Escalate Agent credential rotation — blocked at Build.

StalledCriticalExecution status: Blocked
Escalate Agent credential rotation — blocked at Build
Agent credential rotation has stopped on something outside the workflow, so no gate or dependency will clear it. Blocked at Build — Four standing credentials belong to a vendor whose API caps rotation at 10 calls/hour. Rotation cannot complete inside the maintenance window until that cap is lifted.. Nothing downstream will move until it clears.
CompanyEnterpriseDepartmentEngineering WorkflowAgent credential rotation ExecutiveRhea Kastellan
Open →
Decision waitingHighPending approvals: 3
3 approvals waiting on Ines Marchetti
Ines Marchetti (CAO) is holding 3 decisions, 2 of them rated high risk. Each one is work somebody else has already stopped doing while they wait.
DepartmentPeople ExecutiveInes Marchetti
Open →
Decision waitingHighPending approvals: 3
3 approvals waiting on Nadia Brecht
Nadia Brecht (CKO) is holding 3 decisions, 1 of them rated high risk. Each one is work somebody else has already stopped doing while they wait.
ExecutiveNadia Brecht
Open →
Decision waitingHighPending approvals: 4
4 approvals waiting on Junia Falkenrath
Junia Falkenrath (CMO) is holding 4 decisions, 2 of them rated high risk. Each one is work somebody else has already stopped doing while they wait.
DepartmentMarketing ExecutiveJunia Falkenrath
Open →

4 more not shown. Open the full feed →

Decision waiting5Stalled3

Recommendations26

See all 26 →

CriticalThroughput

Escalate Agent credential rotation — blocked at Build

High98/100

Multiple independent figures agree. Act on it.

Agent credential rotation has stopped on something outside the workflow, so no gate or dependency will clear it. Blocked at Build — Four standing credentials belong to a vendor whose API caps rotation at 10 calls/hour. Rotation cannot complete inside the maintenance window until that cap is lifted.. Nothing downstream will move until it clears.

Suggested action — advisory only

Escalate to Rhea Kastellan, who owns the outcome. The block sits in engineering — the decision needed is whether to wait, re-route around the stage, or stand the workflow down.

Open workflow →

HighGovernance

3 approvals waiting on Ines Marchetti

High89/100

Multiple independent figures agree. Act on it.

Ines Marchetti (CAO) is holding 3 decisions, 2 of them rated high risk. Each one is work somebody else has already stopped doing while they wait.

Suggested action — advisory only

Take the 2 high-risk decisions with Ines Marchetti directly — those are the ones that stay on a desk. The rest can be batched into their next review.

Open CAO →

HighGovernance

3 approvals waiting on Nadia Brecht

High89/100

Multiple independent figures agree. Act on it.

Nadia Brecht (CKO) is holding 3 decisions, 1 of them rated high risk. Each one is work somebody else has already stopped doing while they wait.

Suggested action — advisory only

Take the 1 high-risk decision with Nadia Brecht directly — those are the ones that stay on a desk. The rest can be batched into their next review.

Open CKO →

HighGovernance

4 approvals waiting on Junia Falkenrath

High89/100

Multiple independent figures agree. Act on it.

Junia Falkenrath (CMO) is holding 4 decisions, 2 of them rated high risk. Each one is work somebody else has already stopped doing while they wait.

Suggested action — advisory only

Take the 2 high-risk decisions with Junia Falkenrath directly — those are the ones that stay on a desk. The rest can be batched into their next review.

Open CMO →

HighGovernance

3 approvals waiting on Vale Ordonez

High89/100

Multiple independent figures agree. Act on it.

Vale Ordonez (COO) is holding 3 decisions, 1 of them rated high risk. Each one is work somebody else has already stopped doing while they wait.

Suggested action — advisory only

Take the 1 high-risk decision with Vale Ordonez directly — those are the ones that stay on a desk. The rest can be batched into their next review.

Open COO →

Risk analysis26

Severe

70

exposure index

8 severe · 15 elevated

8 severe exposures against an index of 70. The largest single entry is "Escalate Agent credential rotation — blocked at Build" at 20 of 25. Exposure is spread across categories rather than concentrated.

Exposure by category

Throughput123 · 9
Governance92 · 6
Risk67 · 7
Capacity17 · 2
Knowledge9 · 1
Integrity4 · 1

Impact × likelihood

Impact 51

1

3

6

1

5

7

1

Likelihood 15

Severe ≥ 15Elevated ≥ 9Moderate ≥ 4Contained ≥ 0

Register — highest exposure first

20
of 25
Escalate Agent credential rotation — blocked at Build
Throughputimpact 4 × likelihood 5
Throughput exposure with significant consequences, already happening on the current evidence. Scored 4 × 5.
CompanyEnterpriseDepartmentEngineering WorkflowAgent credential rotation ExecutiveRhea Kastellan
16
of 25
3 approvals waiting on Ines Marchetti
Governanceimpact 4 × likelihood 4
Governance exposure with significant consequences, likely on the current evidence. Scored 4 × 4.
DepartmentPeople ExecutiveInes Marchetti
16
of 25
3 approvals waiting on Nadia Brecht
Governanceimpact 4 × likelihood 4
Governance exposure with significant consequences, likely on the current evidence. Scored 4 × 4.
ExecutiveNadia Brecht
16
of 25
3 approvals waiting on Soren Aldiss
Governanceimpact 4 × likelihood 4
Governance exposure with significant consequences, likely on the current evidence. Scored 4 × 4.
ExecutiveSoren Aldiss
16
of 25
3 approvals waiting on Vale Ordonez
Governanceimpact 4 × likelihood 4
Governance exposure with significant consequences, likely on the current evidence. Scored 4 × 4.
DepartmentOperations ExecutiveVale Ordonez
16
of 25
4 approvals waiting on Junia Falkenrath
Governanceimpact 4 × likelihood 4
Governance exposure with significant consequences, likely on the current evidence. Scored 4 × 4.
DepartmentMarketing ExecutiveJunia Falkenrath

Health

Company health3

2 below target

Agent-weighted mean of department health, read from utils.companyHealth(). A struggling ten-agent department moves it more than a struggling one-agent department.

Department health8

2 below target

2 more, all at or above the ones shown.

Mean attainment across each department's own tracked metrics, read from selectors.departmentHealth(). The drivers are the metrics under target — health is their mean, itemised.

Workflow health8

7 below target

2 more, all at or above the ones shown.

This module's own read, and the only score on this page it does not borrow: 100 less each impediment — 40 blocked, 20 paused at a gate, up to 15 for blocked tasks, 10 starved, 10 for scope overrun. Progress is deliberately not scored, only reported: a workflow opened last week at 15% is young, not unhealthy.

Agent utilization

60% mean load

43

AI employees

deployed workers

60%

Mean load

12

At capacity

≥ 80% load

9

Idle

≤ 20% or idle state

3

Blocked

holding work, stuck

Load by department — hottest firstover 85% · under 45%

These are the 43 AI employees deployed across departments — workers with a load and a state. The Agent Fleet separately registers 127 installed agents, which are capabilities rather than workers: an entry there has tool grants, not utilisation. The two figures are meant to differ.

AI advisors1

0 available

Every recommendation on this page was produced by a deterministic rule in the registry. No model was called, no network request was made, and no API key exists in this application.

No AI provider configuredNot configured
The engine is running on deterministic rules only. Every recommendation on this page was produced by a rule in the registry, with no model involved.

What a provider would be allowed to do

Same shape. An advisor returns the same Recommendation objects the rules do, ranked by the same generator and drawn by the same card. No surface learns that models exist.
Same explanation requirement. Reason, evidence and suggested action are required fields — a model returning a conclusion without the figures behind it produces an object that does not type-check.
Advisory, identically. An advisor is handed a snapshot and returns data. It is given no store, no setter and no callback, so there is nothing for it to write.
Never replaces a rule.Where a model and a rule fire on the same subject, the rule wins — deterministic advice is reproducible and a model's is not. Model findings are capped at moderate confidence and tagged as AI-generated wherever they appear.

The division of labour is thresholds versus patterns. A rule can tell you a department is at 91% load. It cannot tell you that it has been at 91% for three cycles while its objectives slipped and its manager's messages got shorter — that is a pattern across time, modules and prose, and writing a rule for it would mean anticipating it. Rules for anything a threshold catches; a model for what nobody thought to threshold.

How this page is produced

Snapshot. All nine modules are read once, through the selectors each of them publishes — departmentHealth(), workflowStatus(), companyHealth() and six rollups. Nothing is recomputed here, so no figure on this page can disagree with the module it came from.
Rules. 30 pure functions run over that snapshot, each returning findings that carry their own evidence. A rule that throws is contained and traced; the rest still run.
Recommendations. Findings get a stable id, a capped priority, a scored confidence and a rank. One formula for all of them, with every input recorded on the card.
Projections. Alerts, the risk register and this summary are all derived from that single ranked stream — never re-derived from the snapshot, which is what stops an alert and a recommendation contradicting each other.

The whole pipeline is pure. This module imports no store setter and exports no mutation, so “advisory only” is a property of the code rather than a promise about it — every action on this page is a link into the module that owns the change.

See every rule and what it fired on →