Kaiku

Are scope, spend, and authority enforced while the agent runs?

An agent is a role of the workspace rather than an account, with no password, token or access of its own. It runs as the person who called it, with the permissions they hold at that moment, and is clamped to the issue's project, so a question about one project cannot ship the other nine to a model. A read-only token cannot reach the MCP endpoint at all, which is enforcement at the token layer rather than a policy statement. Not 3: every one of these limits is self-attested with no independent verification, and no spend or budget ceiling is published anywhere -- cost is recorded after the fact, not capped.

Counts Human oversight, Runtime governance and Permission scopes, each cited below.

Is there a tamper-evident record of what it actually did?

Every agent run is recorded against the issue it worked on, itemised by tokens by kind, machine time and who asked, and every answer carries the issues and pages its tools actually returned so the reasoning can be checked against its sources. That is a customer-reviewable record of both action and basis, and it is a core product feature rather than a by-product. Scored above the ai-applyd precedent of 1 (2026-10-12) because that record was self-reported and unitemised, where this one is per-run, itemised and attributable. Not 3: audit_trail_immutable is no_public_information, nothing claims tamper-evidence or hash-chaining, so a customer cannot show the log was not altered.

Our automated recount on Oct 13, 2026 counted enough evidenced fields for 1/3. It never reads what those fields say, so it does not overturn the score above, judged on Oct 13, 2026 — but where the two disagree it is a reason to read this dimension again. Why there are two scores

Counts Audit trail and Explainability, each cited below. Also named Tamper-evident log, which our record now shows as undisclosed; that part of the support no longer holds. What a score rests on

Are compliance obligations attached per engagement?

No public evidence found. compliance_certs and regulatory_alignment are both no_public_information against a capture that was read in full -- the page publishes no certification, no audit, no regulatory alignment and no per-engagement schedule. That is a finding about a page we read, not an unchecked gap.

Rests on Compliance certifications and Regulatory alignment, recorded as undisclosed and cited below. For a 0 that absence is the finding. What a score rests on

Does payment depend on a verified result?

Rubric 0 as written. settlement_mechanism is no_public_information, and the evidenced pricing model is a per-seat subscription -- payment is released on subscription, with no link to any verified outcome. Nothing on the page conditions payment on a result.

Rests on Pricing model, each cited below. Also named Settlement mechanism, which our record now shows as undisclosed; that part of the support no longer holds. What a score rests on

Can the deployment be insured, and is the customer indemnified?

No public evidence found. insurance_available, insurance_carriers, indemnification and coverage_limits are all no_public_information against a fully read capture. No carrier is named, no limit published and no customer indemnity offered.

Rests on Insurance available, Insurance carriers, Coverage limits and Indemnification, recorded as undisclosed and cited below. For a 0 that absence is the finding. What a score rests on