Skip to content

Launch Readiness and Quality Gates

Draft 0.1 — 2 August 2026
Passing a launch gate means a defined review was completed against a defined profile. It is not a warranty, certification, legal opinion, or promise that no defect or incident will occur.

ANUKA products need a common answer to a deceptively simple question:

What must be true before we expose this change to real people, real data, or real money?

Launch readiness turns that question into a versioned evidence record rather than a meeting driven by confidence and calendar pressure.

Characteristics:

  • synthetic, public, or tightly controlled data;
  • no customer money;
  • no external promises;
  • no high-impact decisions;
  • small internal audience;
  • disposable environment.

Characteristics:

  • invited participants;
  • explicit experimental notice;
  • bounded data and transaction limits;
  • active monitoring;
  • direct support;
  • feature flags and rollback;
  • predefined stop criteria.

Characteristics:

  • real users and production data;
  • limited geography, cohort, or capability;
  • support and incident processes active;
  • public terms and notices in effect;
  • measured reliability target;
  • controlled expansion.

Characteristics:

  • broadly available within declared markets;
  • normal commercial and support commitments;
  • stable contracts;
  • production SLOs;
  • recurring review and maintenance obligations.

Applies when failure or misuse can materially affect:

  • money;
  • identity;
  • employment or eligibility;
  • sensitive data;
  • public safety;
  • legal rights;
  • multiple Resident products;
  • network governance.

L4 requires stronger independent review, approval, observability, recovery, and human accountability.

A gate must be:

  • proportional to risk;
  • evidence-based;
  • owned;
  • time-bounded;
  • reproducible;
  • capable of failing;
  • explicit about exceptions;
  • linked to rollback and post-launch review.

A gate that always passes is a ritual, not a control.

Every launch creates a Launch Readiness Record containing:

  • launch ID;
  • product and environment;
  • release or change identifiers;
  • launch type;
  • intended users and jurisdictions;
  • capabilities enabled;
  • data classes;
  • financial exposure;
  • risk classification;
  • evidence package;
  • gate outcomes;
  • exceptions;
  • approvers and authority;
  • rollout plan;
  • stop conditions;
  • rollback plan;
  • launch window;
  • review date.
  • pass — requirements satisfied;
  • pass with conditions — launch allowed with explicit conditions and expiry;
  • defer — evidence incomplete or launch not currently justified;
  • fail — unacceptable unresolved risk;
  • not applicable — requirement does not apply, with explanation;
  • emergency exception — temporary risk acceptance through emergency authority.

Required evidence:

  • problem and intended users;
  • user journeys;
  • product scope and exclusions;
  • launch cohort;
  • pricing and commercial terms where applicable;
  • current limitations;
  • success and guardrail metrics;
  • support expectations;
  • claim inventory.

Questions:

  • Can a user understand what the product does?
  • Are experimental and production states distinguishable?
  • Are outcomes and limitations stated honestly?
  • Are irreversible actions clearly marked?
  • Are cancellation, refund, export, and exit paths visible?

Required evidence:

  • current Product Charter;
  • operator and legal owner;
  • release authority;
  • data-publication authority;
  • payment authority;
  • incident authority;
  • conflict disclosures;
  • founder or steward reserved powers;
  • decision record.

No launch may rely on an approver who lacks actual authority.

Section titled “Gate 3 — Identity, consent, and authorization”

Required evidence:

  • authentication flows;
  • role and capability matrix;
  • consent request schemas;
  • partial approval behavior;
  • revocation behavior;
  • recovery and account-transfer behavior;
  • agent principal binding;
  • sensitive-action step-up;
  • authorization tests.

Critical checks:

  • no full Passport disclosure by default;
  • source access is distinct from one-time presentation;
  • least privilege applies;
  • denied or revoked consent is enforced;
  • authorization is server-side;
  • A4 agent actions require human approval.

Required evidence:

  • current data inventory;
  • purpose and legal/contractual basis analysis where required;
  • classification;
  • retention;
  • processors and recipients;
  • data residency or region assumptions;
  • deletion and correction flows;
  • training-use and model-provider rules;
  • public-data controls;
  • privacy notice consistency.

A launch fails if actual production behavior materially contradicts the notice or consent interface.

The product selects a versioned security profile.

Default direction:

  • ordinary web application: OWASP ASVS 5.0.0 Level 1 baseline plus risk-selected Level 2 requirements;
  • products handling sensitive or material transactions: stronger Level 2 or tailored Level 3 requirements;
  • shared infrastructure: architecture, operations, supply-chain, and recovery controls in addition to application checks.

Required evidence may include:

  • threat model;
  • ASVS requirement results;
  • authentication and authorization tests;
  • dependency and secret scans;
  • code review;
  • abuse-case tests;
  • rate-limit tests;
  • penetration testing where justified;
  • security findings and disposition;
  • vulnerability disclosure path;
  • incident exercise.

ASVS references must include the exact version and requirement identifier.

Required evidence:

  • protected source branches;
  • required review;
  • lockfiles;
  • dependency policy;
  • secret scanning;
  • build isolation;
  • artifact digest;
  • SBOM profile;
  • provenance record;
  • deployment identity;
  • rollback artifact;
  • release authorization.

Extension releases also verify:

  • bundled executable code;
  • declared permissions;
  • store-listing accuracy;
  • no prohibited remote code;
  • update and rollback process.

Gate 7 — Accessibility and inclusive use

Section titled “Gate 7 — Accessibility and inclusive use”

Default public web target:

WCAG 2.2 Level AA for supported critical journeys, with documented exceptions and remediation.

Evidence includes:

  • automated scan results;
  • keyboard-only testing;
  • focus order and visibility;
  • screen-reader testing;
  • form labels and errors;
  • contrast and non-color cues;
  • zoom and reflow;
  • target size and dragging alternatives;
  • captions or media alternatives;
  • accessible authentication;
  • plain-language and cognitive-load review;
  • known barriers;
  • accessibility contact and remediation owner.

Automation alone cannot pass the accessibility gate.

Gate 8 — Reliability and operational readiness

Section titled “Gate 8 — Reliability and operational readiness”

Required evidence:

  • service catalog entry;
  • accountable service owner;
  • critical user journeys;
  • SLIs and SLOs;
  • initial error-budget policy;
  • dashboards and alerts;
  • logs, metrics, and traces;
  • queue and retry behavior;
  • dependency failure modes;
  • capacity assumptions;
  • backup and restore status;
  • on-call or escalation path;
  • runbooks;
  • status communication.

An SLO must influence launch decisions. It is not merely a dashboard KPI.

Required evidence:

  • release artifact;
  • change summary;
  • migration plan;
  • feature flags;
  • cohort plan;
  • exposure logging;
  • canary or phased rollout;
  • rollback or forward-fix criteria;
  • data migration rollback or restore plan;
  • launch owner;
  • monitoring window;
  • automatic stop conditions.

A product must distinguish deploying code from exposing functionality.

Where relevant:

  • test-mode flows completed;
  • merchant and processor roles confirmed;
  • fee calculation tested;
  • ledger entries reconcile;
  • payout destination protected;
  • refund behavior tested;
  • dispute handling assigned;
  • tax responsibilities recorded;
  • sanctions restrictions implemented;
  • negative balance and reserve behavior understood;
  • sponsored-feature promises match actual terms.

Production money movement requires an exact approver and environment confirmation.

Required evidence:

  • public claims inventory;
  • support for objective claims;
  • source and methodology;
  • metric definitions;
  • verification status;
  • date and scope;
  • issuer or reviewer identity;
  • correction and expiry;
  • distinction between measured, estimated, modeled, and self-reported data.

“Verified” must always answer: verified by whom, for what property, using what evidence, and when?

Required evidence:

  • support channels tested;
  • response expectations published;
  • severity and escalation process;
  • user-reporting path;
  • abuse and fraud intake;
  • refunds and complaints process;
  • accessibility-support path;
  • privacy and security request path;
  • vulnerable-user considerations;
  • incident communication templates.

The launch profile records:

  • jurisdictions;
  • customer and worker relationships;
  • regulated or high-impact uses;
  • payment and tax assumptions;
  • privacy obligations;
  • intellectual-property rights;
  • terms and policies;
  • required specialist review;
  • unresolved legal-watch items.

This gate must not be represented as automated legal compliance.

Required evidence:

  • key and account recovery;
  • operator succession;
  • domain and repository continuity;
  • provider failure response;
  • user data export;
  • outstanding-payment handling;
  • product transfer process;
  • shutdown notice plan;
  • archival policy;
  • affiliation-removal behavior.

Critical journeys are tested end to end.

Examples:

  • sign in and recover access;
  • review and partially approve a data request;
  • submit a contribution;
  • accept or reject work;
  • fund and refund a sponsored feature;
  • receive or revoke a credential;
  • change payout details;
  • export data;
  • report harm or security issue;
  • cancel or exit.

Testing includes happy path, denial, timeout, duplicate, partial failure, revoked permission, and accessibility variants.

Each gate item can be labeled:

  • E0 — assertion only;
  • E1 — internal artifact;
  • E2 — reproducible test or source-connected evidence;
  • E3 — independent review or controlled experiment;
  • E4 — strong external assurance or repeated production evidence.

Required strength depends on launch risk.

An exception requires:

  • exact unmet requirement;
  • business reason;
  • user and system risk;
  • compensating control;
  • owner;
  • affected cohort;
  • monitoring;
  • expiry;
  • remediation plan;
  • approver;
  • public disclosure decision.

No exception may silently become permanent.

The launch review should be asynchronous-first, using a durable evidence package.

The synchronous meeting focuses on:

  • unresolved risks;
  • exceptions;
  • stop conditions;
  • ownership;
  • disagreement;
  • final decision.

It should not be the first time reviewers see the launch.

Launch typeDefault approval direction
L0 prototypeProduct owner or delegated lead
L1 pilotProduct owner + technical owner
L2 limited productionProduct + security/privacy/operations owners as applicable
L3 general availabilityProduct operator under charter approval matrix
L4 critical/high-impactLegal authority + independent specialist and governance review

Exact authority comes from the Product Charter and governance rules.

Examples:

  • critical security finding;
  • unauthorized disclosure;
  • consent failure;
  • inaccessible critical journey;
  • payment reconciliation failure;
  • elevated fraud;
  • error-budget burn beyond threshold;
  • incorrect public claim;
  • missing support coverage;
  • failed data migration;
  • provider instability;
  • incident-response readiness gap.

Stop conditions must have named decision owners.

A review occurs after the defined observation period.

It records:

  • actual cohort and exposure;
  • product and guardrail metrics;
  • SLO performance;
  • incidents and support load;
  • accessibility reports;
  • security and privacy signals;
  • financial reconciliation;
  • user feedback;
  • unexpected effects;
  • whether to expand, hold, roll back, or change.

The dashboard may summarize gate status, but it must link to evidence and must not produce a misleading single “quality score.”

Recommended views:

  • gate status;
  • unresolved findings;
  • exceptions and expiry;
  • launch owner;
  • current rollout percentage;
  • stop-condition status;
  • SLO and guardrail health;
  • next review.
  • launching because the date was announced;
  • treating all gates as equal checkboxes;
  • approving one environment and launching another;
  • accepting screenshots instead of reproducible tests where tests are possible;
  • claiming WCAG conformance from automated scans alone;
  • using penetration testing as a replacement for secure development;
  • hiding known risks from pilot users;
  • leaving rollback untested;
  • assuming feature flags eliminate migration risk;
  • using a green dashboard to bypass accountable judgment.
  • launch types L0–L4 exist;
  • every production launch has a Launch Readiness Record;
  • critical journeys and stop conditions are declared;
  • security profile references ASVS versioned requirements;
  • accessibility includes human testing;
  • SLOs and error budgets affect rollout decisions;
  • payment flows reconcile before production;
  • public claims link to evidence;
  • exceptions have owners and expiry;
  • rollout uses bounded exposure and rollback;
  • post-launch review is scheduled;
  • no automated score can approve an L4 launch.
  • Sources opened and checked: 2 August 2026
  • WCAG baseline: 2.2 Recommendation dated 12 December 2024
  • OWASP ASVS stable version observed: 5.0.0
  • Automated-only accessibility approval: Prohibited
  • Human launch authority: Required