DEVELOPER DOCUMENTATION

Actions

Configure how MANDATE responds to different score ranges: allow, challenge, or block.

Start in Observe on one route. Use the deployment URL and installed package version supplied for your workspace. Historical host examples and source capabilities do not establish a live service or published release.

Route decisions

Hosted validation is the normal authority for protected browser actions. The

server returns a documented decision, and your backend retains application

authentication/authorization. Raw scores and detector reasons remain internal;

human is a legacy success alias, not human identity.

Hosted outcomeEnforced adapter behavior
allowContinue the protected handler
throttle429 with Retry-After
queue202 with opaque queue token and wait interval
fresh_proof401 for fresh background proof
step_up403 with customer-owned step-up attempt
blockGeneric403; protected handler does not run

No CAPTCHA, checkbox, branded puzzle or automatic business replay is introduced.

Keep response codes coarse and apply authenticated credential/companion updates

before returning. A configured customer-owned step-up is separate from invisible

browser verification.

Observe before enforce

Start one route in Observe, inspect grouped Activity/Signals and false positives,

then move that route to enforcement with rollback. Observe records a proposed

outcome while allowing application execution; proceed is not verified browser

trust or an authenticated user. Off and valid unmatched routes proceed. Unavailable

or malformed managed configuration returns an error, not an assumed Off.

Policy Studio

The source candidate includes five templates: Browser verification, **Request

integrity, Trust threshold, Intent review, and MANDATE Limit**. Templates

are editable starting points, restricted by plan and route capabilities. Workflows

support immutable saved versions, compilation/validation, grouped simulation,

evidence-bound review, release pinning, activation and rollback. Template presence

is not permission to enforce or proof that a runtime can apply every outcome.

Native policy remains default Observe. Enforced policy results require an exact

enrolled F2/F3 interpretation profile, independent catalog review, signed request

capture, matching release and explicit runtime enablement. Publication readiness

also requires fresh signed receipts from every expected reporter; empty inventory

or a convergence flag does not establish quorum. Ordinary SDK callers cannot

supply an approval boolean. Throttle/queue proposals remain Observe where the

native response adapter is unavailable; explicit Limit admission is separate.

Native denied/challenge replies are generic403 with a durable decision receipt.

These are unpublished source capabilities with isolated proof, not production

catalog commissioning or customer activation.

MANDATE Limit and Reroute

Limit configures a request budget by browser/session or authenticated Subject.

Subject comes from server authentication, never a client account header. Observe

records would-limit pressure; enforce returns429 before origin. Verify+Limit use

one canonical admission without a second charge. Off/paused budget leaves counters

and evidence intact.

Reroute supports 1–8 explicitly enrolled browser actions. Defaults remain off.

The guarded lane shares Gate Verify/Limit admission and retains application auth,

exact raw-body proof and private origin intent/result reservation. Readiness is an

operator attestation, not runtime liveness. See Edge.

Usage and quota

Monthly request usage counts allowed validation decisions, not every incoming

HTTP request. Rejected proofs, rate-limit denials, native mint/refresh and offline

Emergency issuance do not create billable allowed decisions. Observe proposals are metered only when the canonical check actually allows

the request; missing/failed proof is not billed merely because an Observe

handler ran.

At the plan limit, security interventions clamp to Observe while evidence keeps

recording and in-depth analytics may pause; protocol integrity checks such as

invalid, replayed or mismatched credentials remain fail closed. Usage totals and

plan-limit status remain available. Raising a plan limit does not commission a

profile or activate enforcement.

See Backend integration, High-value routes,

and Errors.