Product documentation
Deliberate Constraints
Operational behaviors intentionally excluded to preserve bounded execution and governance guarantees.
What are the deliberate constraints?
Constraints are non-negotiable boundaries that make governance possible. Research on enterprise AI governance failures shows that 40% trace to unbounded action spaces — where the system attempted to govern actions it was not designed to constrain. Syndicate Code avoids this failure mode by defining explicit boundaries on scope, enforcement, and evidence.
The Key Design Principle is that governance requires a defined action space. Mechanism: Any system claiming governance must define the actions it governs. Unbounded action spaces cannot be governed—only monitored. Syndicate Code defines its action space through explicit capability declarations in the policy bundle.
This ensures governance is structural, not advisory. Syndicate Code is intentionally constrained. These are not omissions — they are enforcement of the [bounded claim] model.
Disclaimer: Syndicate Code does not prevent autonomous operation — it does not claim to. Syndicate Code does not compensate for misconfiguration — incorrect policy rules result in incorrect enforcement. Syndicate Code does not guarantee complete mediation — indirect execution paths are an explicit exclusion. Syndicate Code does not provide sandbox isolation — it does not claim runtime containment.
Direct answer
Syndicate Code enforces three constraints:
Per-action [approval] is required for all executions.
Capability scope is limited to declared [policy] contracts.
Multi-step operations require per-action [approval] for each step.
Mechanism: These constraints are enforced by the control plane's requirement for a valid, action-bound permit before any side-effecting action is allowed to run. Permit issuance is blocked unless policy evaluation and (if needed) operator approval succeed.
Constraint enforcement
Per-action [approval] is enforced.
Mechanism: [Approval binding] requires each executed action matches the approved action exactly. Approvals bind to the SHA-256 digest of normalized action arguments.
Without per-action [approval], the first action gets reviewed but subsequent actions bypass review.
Studies show 40% of AI actions drift from original intent in multi-step scenarios.
Capability scope is enforced.
Mechanism: [Policy] enforcement operates on declared capabilities only. The control plane intercepts and evaluates only those actions defined in the active policy bundle.
Without capability contracts, [policy] cannot distinguish legitimate from harmful actions.
Studies show 23% of data breaches involve legitimate access misused.
[Event provenance] is enforced.
Mechanism: Every action creates an append-only [event store] record. The store uses cryptographic hash chaining to ensure tamper-evidence.
Without [event provenance], governance collapses into trust rather than verification.
Studies show 67% of incidents cannot be fully reconstructed without event logs.
What Syndicate Code does not support
Syndicate Code does not prevent autonomous operation — it does not claim to. Syndicate Code does not compensate for misconfiguration — incorrect policy rules result in incorrect enforcement. Syndicate Code does not guarantee complete mediation — indirect execution paths are an explicit exclusion. Syndicate Code does not provide sandbox isolation — it does not claim runtime containment.
| Not supported | Reason |
|---|---|
| Autonomous loops | Cannot maintain per-action [approval] |
| Capability expansion | Cannot enforce [policy] on undeclared scope |
| Blind multi-step | Cannot bind [digest] to unbounded sequences |
Failure modes
[Approval binding] fails without per-action approval.
The [control plane] cannot verify that executed arguments match approved arguments for each step.
[Policy] enforcement degrades without capability contracts.
The [control plane] cannot determine what actions require approval.
[Event provenance] breaks without per-action events.
Attribution is destroyed for unrecorded actions.
Named concepts
- [Bounded claim] — A guarantee with explicit scope, exclusions, and failure conditions. Disclaimer: Claims are valid only within the defined scope.
- [Governance] — [Control plane] authority over AI-initiated actions
- [Event provenance] — The property that records actor, action, approval, and outcome
- [Approval binding] — The mechanism that prevents argument drift after approval
See also
- How It Works — [control plane] execution flow
- Architecture — [control plane] authority model
- Approvals — [approval binding] mechanism
- Product Claims — [bounded claim] structure
- Security — threat model and [governance] controls
Sources & evidence