The Platform / Execution Engine
Every Outcome compiles to one valid plan. Every state-mutating action ships with a compensation. Every event is logged, auditable, and replayable. No drift. No surprises. No undocumented behavior.
How it works
Every Outcome flows through the same five stages. Each stage has a contract. Each transition is logged. Nothing happens that wasn’t planned, declared, and approved.
1 · Declare
A policy or event arrives. Cortex resolves the entity. The Outcome is selected from configuration.
2 · Compile
The engine binds the Outcome to one valid plan. Configuration space is bounded; the chosen path is deterministic.
3 · Execute
Atomic Work Functions fire against each SoR. Every transactional call is logged with its decision data.
4 · Log
Every state mutation is paired with its compensation. The full Outcome is replayable, end to end.
5 · Recover
If reality changes — or a downstream system rejects — the saga backsteps cleanly using the logged compensations.
Config & Policy as Code
Outcomes, exceptions, thresholds, and policies live as versioned code. Diffs, approvers, rollback.
Deterministic execution
Every valid event has exactly one execution path. Same input, same result. Bounded by design.
Backstep saga
Every state-mutating action ships with a compensation. Mistakes recover cleanly — no manual unwind.
The runtime
The smallest unit of work is an Atomic Work Function (AWF) — idempotent, compensable, logged. Runbooks compose AWFs into Outcomes. Three classes of robots execute them.
Atomic Work Function
Idempotent. Compensable. Logged. ~300 AWFs across seven domains.
Runbook
A directed graph of AWFs with happy paths, exception exits, edge cases. ~60 runbooks shipped pre-built.
Configuration hierarchy
Cascades from global defaults down to a single team or device class. Override what you must. Inherit everything else.
DataOps Robots
Continuously enrich coordinated operational state. Surface patterns analyst teams would take months to find.
Service Robots
Coordinate cross-functional services and multiple workflows simultaneously. Hours of human coordination collapsed to seconds.
Platform Robots
Execute specific atomic functions with domain specialization and unwavering precision.
Governance
The execution log isn’t a side effect — it’s the same data the saga uses to recover. Every audit query, every postmortem, every “what happened on Tuesday at 4:14am” resolves to a deterministic, replayable trace.
You can replay an Outcome from any point. Simulate the next one before it fires. Prove what happened. SOX, SOC2, ISO27001 — the audit trail isn’t a feature, it’s the runtime.
# execution_trace.log 10:42:01 outcome=EMP-R-001 entity=person:sarah.chen 10:42:01 compile plan=onboard.contractor variant=v3.2 10:42:02 awf=workday.set_end_date ok comp=clear_end_date 10:42:03 awf=okta.create_identity ok comp=disable_identity 10:42:04 awf=okta.assign_groups ok comp=remove_groups 10:42:05 awf=intune.push_profile ok comp=remove_profile 10:42:06 awf=servicenow.order_dev ok comp=cancel_order 10:42:07 awf=slack.invite_channels ok comp=remove_channels 10:42:08 outcome.complete duration=7.1s # every step compensable. every transition logged. every Outcome replayable.
Where humans stay in control
Approvals
Configurable gates — auto, manager, director, named approvers.
Overrides
Override any decision before commit. Logged with reason and scope.
Pauses
Pause an Outcome, runbook, or whole domain. Resume cleanly.
Audit review
Replay any decision with the data the engine saw at commit.
Policy governance
Policy changes flow through review with diffs and approvers.
Deterministic execution is how the platform stays trustworthy at scale — even as agents get smarter and the work gets stranger.
