Customer rollout · IT + Security audience

The runbook your IT/Security team will hand back to you.

The companion to /embedded/integrate/ (5-min dev walkthrough). This page is written for the IT operations, security engineering, vendor risk, and DPO seats at a SaaS customer who is being asked to approve and operate Kinetic Gain Embedded inside their stack. The format is what those teams will actually want from you: pre-approval checklist, change-management ticket template, phased rollout plan, day-2 operations expectations, audit cadence, and a documented off-boarding plan. Lift wholesale into your ticketing system; everything below is intended as a starting artifact, not a sales surface.

1. Pre-approval checklist

Before the change-management ticket goes in, gather this. Most of it the vendor publishes; the rest you decide. Each line maps to a specific buyer question.

Read the threat model end-to-end

Lives at /embedded/threat-model/. Reviewer notes which trust boundaries the SDK assumes, which it explicitly does not, and what failure modes apply. Time: ~15 min.

Verify the npm package is provenance-stamped

npm view kinetic-gain-embedded should show an npm provenance badge tying the published bundle to a specific GitHub Actions build. If not, the supply chain isn't what the vendor claims it is. npmjs.com/package/kinetic-gain-embedded.

Pull + verify the SBOM for the version you'll deploy

SBOM is attached to each GitHub release as a CycloneDX JSON. Diff against your existing approved-dependency list. Block any package you don't recognize; the SDK ships zero runtime deps so this list should be effectively empty.

Confirm Apache-2.0 license terms with your legal team

SDK is Apache-2.0 (permissive, patent grant). Hosted tier terms are separate; that's where indemnification + DPA + SCC language lives. Get both reviewed; they cover different scopes.

Map data flow against your data-classification register

What data does the SDK see? The Decision Card body, the customer record at tokenization time, and the audit event metadata. Confirm each is classified under your existing register (Public / Internal / Confidential / Restricted) and that the SDK's handling matches your policy for that class.

Confirm signing keypair + public key location

If you're using ed25519 signing on audit events, you (the customer-of-the-embedder) decide where the public key lives. Recommended: your own .well-known/ at your apex domain. The embedder doesn't need your private key — only the public key for verification.

Capture the buyer-side Decision Card

You sign the Decision Card once. It locks the policy bundle the SDK enforces at runtime. Save the canonical JSON in your evidence locker. Reference spec.

Decide audit-stream destination

Three options: (a) self-hosted endpoint, (b) embedder's hosted tier, (c) drop-in to your existing SIEM (Splunk, Sentinel, Datadog). For regulated workloads pick (a) or (c); (b) is fine for non-regulated.

2. Change-management ticket template

The ticket your CAB needs. Each field aligns to a question they will ask. Paste, fill in your specifics.

Title:        Approve kinetic-gain-embedded SDK v<X.Y.Z> rollout to <product>
Risk class:   Low / Medium / High  (default: Medium, drop to Low after one quarter clean)
Type:         Third-party dependency · AI governance instrumentation
Requestor:    <eng lead>
Owner:        <eng lead> (build) · <security eng> (verify) · <DPO> (sign-off)

WHAT
  Drop in kinetic-gain-embedded SDK at version <X.Y.Z> into <service>.
  Emits hash-chained, ed25519-signed audit events on AI-touching code paths.

WHY
  Buyer-side AI governance compliance. Audit-stream creates the
  verifiable record our customers' compliance teams ask for.

DATA SCOPE
  - Reads:    Decision Card (we control), customer record at tokenization point
  - Writes:   Audit events to <destination — SIEM endpoint URL or hosted tier>
  - Retains:  Audit events for <Y days> per data-retention schedule
  - Does NOT: Send raw customer records off-prem; the SDK tokenizes
              before any cross-boundary call

DEPENDENCIES
  Runtime deps:    0
  Build deps:      see SBOM attached
  Network egress:  <list endpoints + ports — or "none, SDK is offline">

VERIFICATION CHECKLIST (pre-prod gate)
  [ ] SBOM diffed vs approved-dependency register
  [ ] npm provenance badge verified at install time
  [ ] Threat model read end-to-end by reviewer
  [ ] Audit-event roundtrip tested in staging
  [ ] ed25519 signature verification confirmed end-to-end
  [ ] Decision Card canonical JSON archived in evidence locker
  [ ] Rollback procedure tested (see Section 6)
  [ ] On-call playbook updated (see Section 4)

ROLLOUT
  Phase 1 — Internal staging (1 week)
  Phase 2 — 5% production (2 weeks)
  Phase 3 — General availability after both phases clean

ROLLBACK
  npm install kinetic-gain-embedded@<previous>
  redeploy <service>; audit-stream pause is graceful
  (the embedder side does NOT block on audit emission)

CONTACTS
  Vendor:          security@kineticgain.com  (RFC 9116 security.txt verified)
  Internal owner:  <eng lead>
  Escalation:      <DPO>

3. Phased rollout plan

Standard SaaS-vendor rollout shape: stage → canary → GA. Each phase has explicit exit criteria; don't progress until every criterion is met.

Phase 1 · 1 week

Internal staging

Drop the SDK into one non-production service. Wire a synthetic Decision Card. Verify every audit event roundtrips through your SIEM with intact hash chain + valid ed25519 signature.

  • 1 service, staging env only
  • Synthetic Decision Card
  • Synthetic customer records
  • SIEM ingestion verified
  • Signature verification verified
Exit: 100 events emitted, 100% hash-chain-intact, 100% signatures-verify, zero unexpected egress.
Phase 2 · 2 weeks

5% production canary

Enable the SDK for 5% of production traffic on the same service. Real Decision Card from the buyer. Real records. Production audit-stream destination. Hold here for two full weeks of business-hours operations.

  • 5% production traffic
  • Real Decision Card (buyer-signed)
  • Real customer records
  • Production audit-stream destination
  • Real on-call rotation aware
Exit: 14 days clean. Zero security incidents. Zero SDK-attributable performance regressions in p99 latency. SIEM dashboard shows expected event volume.
Phase 3 · ongoing

General availability

100% production. Move into normal day-2 operations (Section 4). The SDK is now part of your supported stack. Quarterly review begins (Section 5).

  • 100% production traffic
  • All services in scope deployed
  • Day-2 monitoring active
  • Quarterly audit cadence on calendar
Exit: none — this is the steady state. Next milestone is the first quarterly audit (Section 5).
Don't skip Phase 1 even if you're tempted. The hash-chain verification in staging proves your SIEM is correctly parsing the chain link order. If a downstream system silently drops or reorders events, you'd rather find out in staging than during a customer audit.

4. Day-2 operations

The SDK does not require dedicated on-call. It is a synchronous library — when your service is healthy, the SDK is healthy. The only operational surface is the audit-stream destination: if that endpoint stops responding, the SDK degrades gracefully (events buffer in-process up to a configurable cap, then drop with a warning log).

Monitoring

On-call playbook entries

The SDK is offline-safe by design. If your audit-stream destination is unreachable, the embedder service does not fail. It buffers, retries, and ultimately drops events with structured warnings. This is intentional: AI governance instrumentation should never take down the product it's instrumenting.

5. Audit cadence

What to put on the operations calendar. Treat the SDK as a third-party dependency with audit obligations.

CadenceWhatWhy
Weekly SIEM dashboard review: event volume, hash-chain breaks (should be 0), signature failures (should be 0) Catch silent failures within 7 days, not 90.
Monthly SDK version reconciliation: services on stale SDK versions surface as a backlog ticket Prevent CVE exposure window from compounding silently.
Quarterly Decision Card review: re-confirm the buyer-side policy bundle is still the policy you want enforced. Re-sign if anything material changed. Decision Cards are immutable per version; if your policy changed, you need a new version + re-signing.
Quarterly Vendor /.well-known/ freshness check: pull kineticgain.com/.well-known/aeo.json + 10 sibling docs, verify ed25519 signatures still validate If the embedder rotated keys or the signatures expired, your signed audit events depend on a chain you no longer trust.
Annually Threat model re-read: assumptions may have changed (new attack classes, new regulatory frameworks). Update accepted-risks list. Threat models age. Yours and the embedder's.
Annually Access review on the audit-stream destination: who can read the audit log, who can write, who can delete An audit log nobody but the right people can touch is worth more than a perfect signature.
Annually Off-boarding drill (dry-run): confirm the off-boarding plan in Section 6 still executes cleanly against current architecture Untested rollback plans are a fiction your incident response can't rely on.

6. Off-boarding plan

How to safely remove KGE from your stack. Treat as a one-quarter operation, not a one-day operation — the audit-stream history needs to remain available for buyer-side compliance even after you stop emitting new events.

  1. Decision (Week 0): The off-boarding ticket goes through the same CAB that approved the rollout. Document the reason (commercial, technical, strategy, or replacement-with-alternative).
  2. Stop new emission (Week 1): Disable the SDK in code paths via feature flag. Verify event volume drops to zero in SIEM. The buffered events still in flight will flush within a few minutes.
  3. Verify the historical record (Week 2): Run your full hash-chain verification against the entire retained audit log. If anything is broken, fix it now — once the SDK is removed, you've lost the live verifier.
  4. Archive (Week 4-12): Export the audit log to long-term retention storage per your data-retention schedule. Confirm the buyer-side Decision Card + public verification key are archived alongside so the log remains independently verifiable.
  5. Remove (Week 13): npm uninstall kinetic-gain-embedded, redeploy. The audit log is now read-only history. Update your data-retention schedule + vendor register.
  6. Post-mortem (Week 14): If this was a replacement, document why; if commercial, document the cost driver. Inform the embedder for product-feedback purposes (this is courtesy, not a contractual obligation).
Don't delete the historical audit log when you remove the SDK. Your customers' auditors may ask to see records covering periods after you've stopped using KGE. The log is your evidence; treat it as evidence you would in any other vendor-departure scenario.

7. Reference materials

Everything below is intended to be read end-to-end before approval, then referenced as needed during operations.

Vendor-side artifacts

Your-side artifacts (Trust Pack tooling — free)

Standing reference (always-on)