Enter the access password to view the site.
A worked example traced end-to-end through five phases with RACI applied per phase. The same feature, the same trio, the same gates — narrated so the workflow's shape is visible at a feature scale, not just at the framework level. Rollout-phase feature; SDD level: Spec-First.
Usage — Settings area subnav item. Audience: org admin with billing entitlement. A net-new analytics surface where the tenant-selector pattern has no prior domain work and the interaction model has no established precedent in the problem space.
Two steps inside Phase 1. Step one converts continuous signal into a G0 problem statement. Step two opens a discovery thread that surfaces what is and isn't known before Spec begins.
Signal accumulation is not triggered by this feature — it has been building continuously. When the admin visibility opportunity surfaces as a graduation candidate, the relevant signal is already organized.
The inbound signal agent has clustered a recurring theme: admins cannot see platform consumption before it becomes a billing event. The cluster carries meaningful ARR weight — concentrated in mid-market and enterprise accounts on metered plans with multi-tenant setups. The research synthesis agent has tagged qualitative evidence from four admin interview studies. The recurring theme:
PM reviews the ranked cluster and approves the opportunity as a graduation candidate. The CS signal brief is authored: five qualifying escalations from CSM and SE, ARR-weighted, tagged to the platform retention initiative.
PM authors the G0 problem statement — specific enough that two engineers reading it independently would scope the same system boundary: billing data exposure in Settings for org admins, per-tenant and per-capability, with overage signaling. CS signal brief disposition documented: acted on — five escalations directly inform v1 scope; multi-tenant demand concentration shapes the tenant-selector requirement. Downstream costs documented: anomaly detection does not exist as a service — this dependency is known at G0 and enters Phase 1 as a flagged item, not a surprise at G2.
| Activity | PM | Des | Eng | Ag |
|---|---|---|---|---|
| Inbound signal cluster review + CS signal brief production5 qualifying escalations · $2.1M ARR weighted | AR | C | — | Ag |
| Research synthesis review4 studies · "invoice surprise" theme | AR | AR | — | Ag |
| G0 problem statement authoredAnomaly detection dependency flagged at G0 | AR | C | C | Ag |
| CS signal brief reviewed · disposition documentedActed on · 5 escalations inform v1 scope | AR | I | I | — |
| Downstream costs documentedAnomaly detection service: new contract required | AR | I | C | — |
| G0 artifact signed | A | — | — | — |
Design's accountability: understand the problem space and surface what is and isn't known before Spec opens.
What is known: admin users monitor billing reactively; no current usage surface in Settings; Textura has gauge and chart primitives; the platform admin domain companion file covers Settings layout conventions and form patterns. What is not known: how admins mentally model capability categories relative to their plan structure; whether per-tenant granularity is the primary interaction context or a secondary drill-down. CS signal is clear that per-tenant breakdown is wanted — it is not clear whether admins want to start at tenant or org level.
Research debt declared: the tenant-selector context-switch pattern has no precedent in the platform admin domain. No prior feature in this domain has implemented a scope-switching control that affects an entire surface. Design surfaces this to PM as a generation-time risk — the interaction model for the tenant-selector will require a conservative approach in v1. PM acknowledges — does not block graduation; research debt logged and carried forward.
Feature tier: New capability. No prior analytics surface in Settings; interaction model is net-new; multi-tenant scoping introduces a pattern the domain has not solved. Capacity confirmed by Engineering Lead against the throughput baseline: one new capability feature is within scope for the current sprint window. Architecture early entry: not triggered. No new service, external dependency, or public API change identifiable at G1.
| Activity | PM | Des | Eng | Arch | Ag |
|---|---|---|---|---|---|
| Problem space orientation record authoredKnown/unknown · prior DS coverage · CS signal relevance | C | AR | I | — | Ag |
| Research debt declared + surfaced to PMTenant-selector: no prior domain precedent | C | AR | I | — | — |
| Feature tier assignedNew capability — analytics surface is net-new | AR | C | C | — | — |
| Team composition confirmed · capacity confirmedOne new capability feature within sprint window | AR | I | AR | — | — |
| Architecture early entryNo trigger identifiable at G1 | I | C | R | AR | — |
| G1 sign-offPM: tier + team confirmed · Des: problem space confirmed, research debt surfaced · Eng: capacity confirmed | A | R | R | — | — |
The trio convenes against spec.md. Each role owns its sections. Design leads because variant scope and intent decisions gate everything downstream.
The trio convenes for a role-led session against spec.md. Each role owns its sections non-delegably. Design leads because the variant scope and design intent decisions gate everything downstream.
JTBD from G1 carried forward verbatim. Success metrics: reduction in overage-related support tickets at 60 days; dashboard engagement rate at 30 days; anomaly callout interaction rate when the detection service is available. Problem space maturity assessed: Low — new interaction space, limited DS coverage for this use case, no prior work in this problem space. This assessment determines which P7 sufficiency row applies at G2.
UsageSummary:
org_id, tenant_set_id, billing_period_start, billing_period_end,
plan_tier, entitlement_limit {per_capability}, current_usage {per_capability},
usage_percent, overage_status [normal | at_risk | over_limit | data_lag],
last_updated_at
CapabilityMetric:
capability_type [workflows | agents | apps | documents | tables | ai | solutions],
tenant_scope, period_type, usage_value, entitlement_limit, usage_percent,
overage_status, trend_direction, forecast_at_period_end
AnomalyEvent:
event_id, capability_type, tenant_scope, detected_at, severity,
usage_delta_percent, description
⚠️ detection_service_contract — new service required; contract pending
AnomalyEvent.detection_service_contract flagged: service shape can be specified now, but the hook is a hard dependency on a service that does not yet exist. Both detection_unavailable graceful degradation state and the mechanism for when the service is fully absent are logged as open questions Engineering must resolve before G3.
Design intent (§1.7): surface entitlement health at a glance for admins who check reactively, with clear escalation paths to per-capability and per-tenant detail when something looks wrong. The interaction model should reduce cognitive load at org level before inviting drill-down. Constraints (§1.8): entitlement gauge must use Textura gauge component — no custom visualization; chart primitives may be used for trend lines; tenant-selector must be conservative in v1 (scope-switch only, loading state only, no mid-transition states); anomaly callout placement must not occlude the entitlement summary at any viewport on initial load; detection_unavailable must hide the anomaly section entirely, not show an error state.
Variant scope declared at G2: New capability, High interaction complexity — governing table yields 7–10 range. Designer declares 7 variants: 4 option-exploration + 3 edge-case. Confirmed in spec.md variant scope section before G2 closes.
One flag: §2.4 state list does not include detection_unavailable explicitly — implied by §1.8 constraint but not listed as a state. Resolved in session: state added to §2.4 before the gate closes. No other flags. Architecture not triggered: no new service, external dependency, or public API change in scope at G2.
| Activity | PM | Des | Eng | Arch | Ag |
|---|---|---|---|---|---|
| PM-owned spec sectionsJTBD · success metrics · ACs · maturity assessment | AR | C | C | — | Ag |
| Design-owned spec sectionsDesign intent · §1.8 constraints · 7-variant scope declaration | I | AR | C | — | Ag |
| Engineering-owned spec sectionsData shape · state list · service hooks · detection flag | C | C | AR | — | Ag |
| State list co-reviewdetection_unavailable added to §2.4 in session | I | R | AR | — | — |
| Spec quality gap detection (pre-G2)1 flag: detection_unavailable missing from §2.4 → resolved | R | R | R | — | Ag |
| Architecture engagementNot triggered — no new service / API change at G2 | C | C | R | AR | — |
| G2 sign-offPM: P7 sufficiency confirmed · Des: intent + ACs + 7-variant scope · Eng: build approach + data contracts | A | R | R | — | — |
Design's execution thread. Engineering does not begin implementation until G3 clears — the G3-approved spec and prototype reference are required inputs for context loading.
Design's execution thread. Engineering does not begin implementation until G3 clears — the G3-approved spec and prototype reference are required inputs for context loading.
Loading sequence confirmed: textura-design-constitution.md v1.0 → textura-platform-constitution.md v1.0 → platform-admin-design-patterns.md. G2-approved spec.md loaded. Textura MCP v0.9 connected — all seven capability-type components confirmed governed. Context loading gate passes.
Option-exploration (4):
A — Summary-first: org-level entitlement gauge row across top (seven capabilities, overage status + micro-trend sparkline per gauge); anomaly callout strip below; inline drill-down panel. Tenant-selector in header as secondary scope control.
B — Capability-grid: seven-cell card layout; each card with gauge, usage percentage, trend direction; per-card time comparison; full-screen drill-down.
C — Alerts-first: anomaly and at-risk summary panel above the fold; dense capability table with inline gauges and period-over-period deltas below; slide-out drawer for drill-downs.
D — Tenant-led: tenant-selector as primary entry context; all metrics scoped to selected tenant from load; org aggregate accessible via "All tenants."
Edge-case (3):
E — Fresh billing period: zero usage, forecast unavailable, all gauges at 0%, empty-state treatment per component.
F — Data lag: entire surface in degraded state per data_lag treatment without hiding the surface.
G — Single-tenant org: tenant-selector hidden entirely; simplified header.
DS compliance: all seven variants use Textura primitives correctly. The micro-trend sparkline below the gauge arc in Variant A is not a governed Textura pattern — flagged as a recipe candidate for domain promotion, not a violation. Accessibility: Variant C's color-only overage indicator — critical violation; resolved by adding icon + text label. Variant B's full-screen drill-down missing focus trap on open — major violation; resolved by adding focus management annotation to §1.8 for engineering. State completeness: Variant G does not implement detection_unavailable state for the anomaly section — blocking; designer adds graceful hide to Variant G and re-confirms before advancing.
At Spec-First, designer runs the evaluation and reviews all output. LLM-as-judge evaluates all seven variants against design ACs and design intent. Summary: Variants A and D are the strongest against the "entitlement health at a glance" intent — A because org-level summary is immediately visible, D because it scopes to the user's most relevant context. Variants B and C require more cognitive load before health signal is legible. All three edge-case variants pass their state coverage checks.
Roundtrip criteria applied: state complexity (4+ states × 7 capabilities, plus anomaly and drill-down states — well above the 5-state threshold ✓); stakeholder review required (CPO has asked to see the design before build begins ✓). Two criteria met — roundtrip warranted. Design runs the roundtrip on Variant A: gauge cluster in all four entitlement states, anomaly callout in active and detection_unavailable, drill-down panel. Spacing and typographic refinements applied in Figma, pushed back to code prototype.
Variant checkpoint fires: New capability, High complexity — trigger conditions met. 20-minute trio session. PM: summary-first aligns with how admins monitor plan health — overall first, specifics when something is wrong. Engineering: Variant A's inline drill-down avoids a navigation transition, simplifying routing. Design selects Variant A. Rationale recorded in spec-decisions.md.
UX hard gate cleared: all 7 variants deliverable (not placeholder); all critical and major accessibility violations resolved; all states in §2.4 represented. G3 completeness checklist (§4.2) passes. Engineering Lead has received the Tier 1 prototype and reviewed accessibility annotations before the G3 session is scheduled.
| Activity | PM | Des | Eng | Ag |
|---|---|---|---|---|
| Constitution reference loadedDesign const · Platform const · Domain companion | — | AR | — | — |
| Variant generation — 7 variants4 option-exploration + 3 edge-case | — | AR | I | Ag |
| DS compliance validationNo violations · sparkline flagged as recipe candidate | — | AR | — | Ag |
| Accessibility validation2 violations found and resolved (C + B) | — | AR | — | Ag |
| State completeness validationVariant G: detection_unavailable missing → resolved | — | AR | — | Ag |
| LLM-as-judge evaluationSpec-First: designer runs + reviews all output | I | AR | — | Ag |
| Figma roundtripState complexity + stakeholder review: 2 criteria met | — | AR | — | Ag |
| Variant checkpointRequired: New capability / High complexity · 20-min trio session | C | A | C | — |
| Variant selectionVariant A — summary-first | I | A | C | — |
| UX hard gate clearedAll 7 deliverable · all violations resolved · §4.2 checklist passes | I | AR | — | — |
| G3 handoff package assembledTier 1 code prototype · variant inventory · a11y annotations · spec version pointer | I | AR | C | — |
| G3 sign-offDes: deliverable complete, UX hard gate cleared, §4.2 confirmed · Eng: build-readiness + approach confirmed · PM: informed | I | AR | R | — |
Engineering's AI-assisted implementation thread. Two steps: AI-assisted build with per-unit constitution checks, then a UX Review Gate that fires before CI/CD runs.
Engineering's AI-assisted implementation thread. Begins after G3 — the G3-approved spec and Tier 1 prototype are required context loading inputs. This is a hard sequencing dependency.
Loading sequence confirmed: (1) platform-admin-constitution.md (domain constitution companion file) — current via CLAUDE.md; (2) spec.md — G3-approved version confirmed; (3) Tier 1 Textura code prototype from the G3 handoff artifact — accessible. No open Architecture review items in the G3 decision log. Context loading gate passes.
Each unit initialized with a separate agent brief: spec section reference, active ACs, and active constraints. Engineering Lead reviews agent output against spec and confirms constitution compliance before proceeding.
Unit 1 — Entitlement gauge cluster: seven gauge components in a horizontal row; four entitlement states per gauge (normal, at_risk, over_limit, data_lag); micro-trend sparkline below each gauge arc. Agent generates against §2.1–2.4. Constitution compliance confirmed — all token references per domain companion file; no hardcoded values.
Unit 2 — Anomaly callout strip + detection_unavailable state: severity badge, capability label, delta percentage, drill-down affordance. detection_unavailable graceful hide implemented per §1.8. Engineering Lead notes the anomaly-flagging feature flag will control full section visibility — but §1.8 only specifies the graceful hide behavior for when the service returns no events, not for when the service is entirely absent. Spec gap flagged for pre-review.
Unit 3 — Time comparison control + tenant-set selector: five comparison modes; tenant-set selector scoped per §1.8 constraint; loading state for mid-switch. Conservative interaction model confirmed.
Unit 4 — Inline drill-down panel: trend line using Textura chart primitive; period-over-period comparison table; per-tenant breakdown; CSV export trigger. Entity fields confirmed against §2.3 data shape.
Automated scan of the full implementation against spec ACs and domain constitution companion file.
Blocking — anomaly detection service mechanism: the anomaly-flagging feature flag pattern is an AI-generated implementation approach with no prior precedent in the domain constitution companion file. Architecture trigger condition met: new AI-generated implementation pattern. Architecture engaged. Architecture reviews: controlling section visibility via a feature flag is an acceptable implementation pattern; does not constitute a new service dependency or cross-team contract. Architecture exits; decision logged in decision-log.md.
Non-blocking — CSV export progress state: aggregate view CSV export under production data volumes will produce a perceptible delay. Progress indicator state is not implemented. Engineering Lead: state addition only — Textura progress bar component available. Logged as spec gap amendment.
| Activity | PM | Des | Eng | Arch | Ag |
|---|---|---|---|---|---|
| Context loading gate confirmedDomain constitution + G3-approved spec + Tier 1 prototype | — | I | AR | — | — |
| Agent brief initialized per unit · AI-assisted implementationClaude Code · Eng Lead reviews each unit before next | — | — | AR | — | Ag |
| Constitution compliance check — per unitAgent self-reports · Eng Lead confirms before next unit | — | — | AR | — | Ag |
| G4 pre-review agent brief run1 blocking (anomaly flag pattern) · 1 non-blocking (CSV progress) | I | C | AR | C | Ag |
| Architecture review — new AI-generated implementation patternFeature flag gating: no prior domain precedent → confirmed acceptable | I | I | R | AR | — |
| Spec gap amendments — surfaced by pre-reviewAmendment 1: anomaly flag mechanism · Amendment 2: CSV progress state | A | C | R | — | — |
anomaly-flagging flag is off, the anomaly section does not render at all. When the flag is on but detection service returns no events, detection_unavailable graceful hide applies. Distinction between flag-off and detection_unavailable logged. Design confirms consistent with spec intent. Architecture confirmed: acceptable implementation pattern. PM disposition: clarification only.The UX Review Gate runs first. QA testing does not begin until design has reviewed the built UI and approved it. Non-negotiable at Spec-First.
Design reviews the implemented feature against the Tier 1 prototype and the G3 handoff package. The review is structural — interaction model and state coverage, not pixel-perfect comparison. Most of the implementation matches cleanly. Three DS compliance flags found:
Flag 1: anomaly callout severity badge uses hardcoded hex (#E6A200) instead of the status-warning token from the domain's token namespace. Traces to engineering implementation of a correct design artifact — agent output during Unit 2.
Flag 2: drill-down panel header has a 12px gap where the design artifact specifies 16px. Wrong Textura spacing token used — not hardcoded, but incorrect.
Flag 3: anomaly callout hover state not implemented. Specified in the accessibility annotations Engineering Lead received before the G3 session.
All three trace to engineering implementation. Design routes them to Engineering. Engineering resolves all three. Design re-confirms resolution before CI/CD runs.
axe-core: passes — accessibility violations from design-time are fully resolved. DS token linting: passes — all three token violations from the UX Review Gate are resolved. Visual regression: passes against Figma reference captures from the roundtrip.
At Spec-First, designer runs the evaluation and reviews all output. No Tier A findings. One Tier B flag: tenant-set selector focus management may not be returning correctly on scope switch. Design reviews the accessibility annotation and the implementation — focus management is correctly implemented per the annotation, which specified focus return to the tenant selector control after switch. Flag disposed as false positive; rationale logged.
PM confirms: all ACs in §1.9 verifiable against the spec as authored at G2. Amendments 1 and 2 are clarifications — neither introduced new scope, neither triggered the Tier 3 protocol. Decision log complete: both amendments logged with spec section reference, what changed, why, and PM sign-off. Engineering confirms: all CI/CD gates cleared, pre-review findings disposed, decision log filed in decision-log.md.
| Activity | PM | Des | Eng | DS | Ag |
|---|---|---|---|---|---|
| UX Review Gate3 DS flags found — all trace to Eng implementation | I | AR | I | — | — |
| UX Review Gate — DS flag resolutionstatus-warning token · 16px spacing · hover state · all resolved before CI/CD | — | AR | R | — | — |
| CI/CD automated gatesaxe-core · DS token linting · visual regression — all pass | I | C | AR | — | Ag |
| LLM-as-judge evaluation0 Tier A findings · 1 Tier B flag disposed (false positive) | I | AR | R | — | Ag |
| ACs verifiable against G2 specAmendments 1 + 2 confirmed as clarifications only | AR | R | R | — | — |
| DS token + accessibility compliance confirmed | — | AR | R | C | — |
| Decision log complete · filed in feature branch | R | R | AR | — | — |
| Architecture regression checkNo constitution-level regression at QA | I | I | R | — | — |
| G4 sign-offPM: ACs verifiable against G2 spec · Des: DS token + accessibility confirmed · Eng: CI/CD cleared, decision log complete | A | R | AR | — | — |
Release confirms the build is deployed and stable. Launch is the flag flip — PM owned, PM timed. Close-out activates for rollout-phase teams and feeds the AI FDLC Review.
Release and Launch are separated. Release complete is Engineering's milestone — it confirms the build is deployed and stable. It does not authorize the flag flip; PM controls Launch timing independently.
Engineering deploys to production under the admin-usage-analytics feature flag (off for all orgs). Smoke test runs against a billing-entitled test org: all seven entitlement states render correctly; anomaly section behavior per Amendment 1 confirmed (flag-off → section does not render); CSV export progress indicator per Amendment 2 confirmed on a large data set simulation. Health monitoring shows stable response times and no error rate elevation on billing data service API calls.
Engineering signals Release complete to PM. PM reviews launch coordination: customer comms drafted (in-app tooltip + CSM enablement brief); CSM team briefed; the billing-entitled org enablement list confirmed with RevOps. PM confirms Launch coordination is ready.
| Activity | PM | Des | Eng | Arch |
|---|---|---|---|---|
| Deployment execution | — | — | AR | — |
| Smoke testAll entitlement states confirmed · Amendments 1 + 2 validated | — | — | AR | — |
| Health monitoring confirmed stable | I | — | AR | — |
| Release complete confirmed · PM notified | R | — | AR | — |
| Launch coordination readiness confirmedCustomer comms · CSM enablement · entitlement list confirmed with RevOps | AR | — | I | — |
| Architecture — deployment-surfaced regressionNo constitution-level regression at deployment | — | — | R | AR |
PM confirms GTM readiness and makes the flag flip decision. Engineering executes on PM authorization — flag flip is not automatic on Release complete.
PM authorizes the flag flip for admin-usage-analytics for billing-entitled orgs. anomaly-flagging remains off — anomaly detection service still pending. Engineering executes the flag flip and monitors first-hour health: no error rate increase, no billing API latency spikes, no unexpected client-side errors in session telemetry.
Feature outcome ownership activated. PM links the feature to the platform retention OKR and the admin-visibility workstream tracker. AI FDLC Review reporting committed at Launch. This commitment activates the post-ship close-out checkpoint — this team has advanced to the rollout phase.
| Activity | PM | Des | Eng |
|---|---|---|---|
| Flag flip decision + timingadmin-usage-analytics on · anomaly-flagging remains off | AR | I | I |
| GTM readiness confirmed | AR | I | I |
| Flag flip execution on PM authorization | — | — | AR |
| First-hour post-launch monitoringNo error rate increase · no latency spikes | I | — | AR |
| Feature outcome ownership activated | AR | I | I |
| AI FDLC Review reporting committedRollout-phase team — activates post-ship close-out | AR | I | I |
This checkpoint activated because this team has advanced to the rollout phase — it is not part of the pilot. PM owns the close-out record. Design and Engineering surface amendment candidates. Architecture engages only if an amendment candidate requires a constitution-level or cross-team contract decision.
Primary metric — feature throughput: G0 artifact to Launch in 18 working days at Spec-First. Spec drift rate: 2 amendments across the full lifecycle — both confirmed clarifications; 0% attributable to spec authoring gaps. Amendments were surfaced by the pre-review agent brief during Phase 4, not by QA. Rework rate: 3 DS compliance flags at the UX Review Gate — all traced to engineering implementation of a correct design artifact; 0 rework traceable to spec gaps. Token spend actuals: within expected range for New capability / High complexity at Spec-First.
PM authors and files the close-out record in the feature branch decision-log.md. Eight sections: feature summary, G0-to-Launch timeline, amendment log (Amendments 1 and 2), spec drift analysis, rework analysis, token spend actuals, open items carried forward (anomaly-flagging service; feature outcome measurement at 30 days), and FDLC Review reporting data.
Tenant-selector context-switch pattern — this feature is the first implementation of a scope-switching control in the platform admin domain. The interaction model (scope-switch only, loading state, no mid-transition states) should be documented as a recipe in the platform admin domain companion file before the next feature needing this pattern. Design surfaces this to DS Team. Candidate: domain companion file addition.
Entitlement gauge cluster recipe — seven gauges in a horizontal summary row with threshold-aware state transitions and micro-trend sparklines. First documented composition of governed Textura components for quota/entitlement contexts. Candidate: domain companion file addition; experience-type promotion candidate on second use.
Feature flag gating for service-not-ready sections — Architecture confirmed acceptable during Phase 4, but it has no prior domain precedent. Engineering surfaces as a candidate for the domain constitution companion file: a documented pattern for AI-generated implementation of features with deferred service dependencies. Recipe-level addition.
CSV export progress state pattern — likely to recur in other admin-facing data export features. Engineering surfaces as a domain companion file addition candidate.
Architecture not triggered — all four candidates are recipe-level additions to companion files. None require a constitution-level decision or cross-team contract.
| Activity | PM | Des | Eng | DS | Arch |
|---|---|---|---|---|---|
| AI FDLC Review reporting data presentedThroughput · drift rate · rework rate · token spend | AR | I | I | — | — |
| Close-out record authored + filed8 sections · filed in feature branch decision-log.md | AR | C | C | — | — |
| Design constitution amendment candidates surfacedTenant-selector recipe · Entitlement gauge cluster recipe | I | AR | I | C | — |
| Engineering constitution amendment candidates surfacedFeature flag gating pattern · CSV progress state pattern | I | I | AR | — | — |
| Architecture engaged for amendment evaluationAll candidates are recipe-level — no constitution-level decision required | I | C | C | — | AR |
Recipes, amendments, and validated patterns now live in the companion files. The next feature in this domain does not start from a blank generation session — it extends.
Four recipe-level amendment candidates are queued for the next AI FDLC Review: the tenant-selector context-switch pattern, the entitlement gauge cluster composition, the feature flag gating approach for deferred service dependencies, and the CSV export progress state. None required a Textura component change — they are documented compositions and patterns that now have production validation. The next feature in this domain that needs a scope-switching control does not start from a blank generation session. It starts from a tested, reviewed recipe in the domain companion file.
Two amendments were logged during this feature's lifecycle. Both were clarifications. Neither reopened G2, and neither triggered the Tier 3 protocol. The spec quality gap detection at G2 caught the missing detection_unavailable state before it reached Phase 3 generation. The pre-review agent brief during Phase 4 surfaced the anomaly detection service gap and the CSV progress state gap before they reached QA. The UX Review Gate caught three DS compliance violations before CI/CD ran. Each finding was caught at the right layer — not the layer downstream.
The Release/Launch separation meant Engineering could confirm the deployment was stable before PM made the flag flip decision. The post-ship close-out surfaced four recipe candidates that would otherwise have been tacit knowledge. That knowledge is now in the domain companion files that will be loaded at context loading gate for the next generation session in this domain. The next feature does not reconstruct what this feature decided — it extends it.