acdl/docs/presentations/nova-autonomous-cloud-delivery.md

# Nova — The Autonomous Cloud Delivery Platform

> **Source of truth** (Step 1 of the 4-step deck process).
> Unified narrative deck. 4-beat arc: Problem → Solution → Proof →
> Roadmap + Ask. x3 structure at deck level (opening = the problem + the
> arc, body = tell them, closing = recap + ask) AND per slide (opens with
> what it covers, delivers, closes with a benefit callout written for a
> tech-leadership audience).
>
> **Honesty model:** every metric cited is grounded (cites a source),
> derived (documented formula), or deferred (cites the blocking work).
> No fabricated numbers. Internal provenance (decision IDs, requirement
> IDs, internal file paths) is kept out of the audience-facing slides —
> those live in the appendix and the `.ciagent/` files only.
>
> v1.21 — Deck Refinement & Pipeline Hardening

---

## Slide 1 — The Problem

**Product teams now own their cloud infrastructure — but ownership without
discipline is destroying value.**

The broad shift to "you build it, you run it" put Terraform into the hands
of product teams. The intention was right: teams that own their stack ship
faster. The reality is that infrastructure-as-code is a different craft
from software development, and the engineering standards that teams apply
to application code are rarely applied to the infrastructure that carries
it.

- **No lifecycle planning.** Resources are authored for creation, not for
  patching, decommissioning, or rollback. When a change is needed, the
  change is destructive — because no one planned the lifecycle.
- **Proactive scanning is not part of authoring.** In a year where
  AI-frontier models discover and exploit zero-day vulnerabilities at a
  rapid pace, teams cannot keep up by reacting. Infrastructure modules
  must be scanned as code and at runtime, post-deployment — and remediated
  at the pace the threat moves, not the pace a sprint allows.
- **Bandwidth gaps in infrastructure operations.** An unusual amount of
  time is spent on remediation, the push for innovation does not pause,
  and the result is that operational work is chronically under-resourced.
  Gaps open. Detections are missed. Incidents grow.
- **Tribal knowledge and the rockstar-operator problem.** Operations
  depend on a handful of administrators who hold the infrastructure in
  their heads. When they leave, the knowledge leaves with them. The
  platform should encode the discipline, not the person.

Every hour a developer spends writing, deploying, fixing, or remediating
infrastructure is an hour not spent releasing features to production and
generating value.

> **Benefit:** the rest of this deck shows the answer — an autonomous
> cloud delivery platform that encodes infrastructure discipline as
> policy, scans proactively, remediates rapidly, and makes operations
> visible to leadership rather than hidden in tribal knowledge.

> **Speaker notes:** Do not frame this as "humans are the problem." The
> problem is that ownership was granted without the discipline, tooling,
> and lifecycle planning that infrastructure requires. The operator is
> not the bottleneck because operators exist — the bottleneck is that
> operations depend on a few individuals instead of an encoded system.

> **Transition:** "Here is the destination Nova is building toward."

---

## Slide 2 — Nova's Vision

**Infrastructure operations become visible. Every environment provisioned,
every incident healed, every risk remediated — by an autonomous system
whose trustworthiness is provable, not promised. Human attestation remains
required at stage gates; the operator is never in the loop of normal
operations.**

- **Visibility is the recurring theme.** Security posture, remediation
  velocity, reliability, and lead time are surfaced as queryable signals —
  not hidden in a person's head or a Slack thread.
- **Provable, not promised.** Trust is established by deterministic
  scripts that calculate a score and gate the action. The platform
  functions without AI. "AI decisions" are really automated decisions.
- **Autonomy in operations, human at stage gates.** QA signs off for
  production; SRE greenlights based on operational readiness. The
  absence of an operator in the loop is never the absence of a record.

> **Benefit:** the destination is autonomous operations with provable
> trust — security, remediation velocity, reliability, and lead time made
> visible to leadership, not promised to them.

> **Speaker notes:** "Visible" is the operative word. The vision is not
> just that operations run without an operator — it is that operations
> become observable, queryable, and accountable. That is what makes the
> trust defensible.

> **Transition:** "The vision is ambitious — here are the strategic
> objectives that make it concrete, and the anti-goals that keep it
> focused."

---

## Slide 3 — Strategic Objectives + Anti-Goals

**Four objectives Nova is building toward; four anti-goals that keep it
focused.**

**4 Strategic Objectives:**
1. **Demonstrate production-grade zero-touch operations** — autonomy as
   the default, not the demo. Stage-gate attestation (QA, SRE) remains
   human by design.
2. **Establish provable trust in automated decisions** — deterministic
   scripts calculate a score; a band outcome gates the action. The
   platform functions without AI. The Decision Ledger, confidence
   scoring, circuit breakers, and blast-radius controls make
   "autonomous" a defensible claim, not a marketing one.
3. **Deliver compounding, quantifiable ROI** — measured on four CTO-grade
   metrics, all flowing into PowerBI:
   - **Lead Time** (PR → Production) — downward trend.
   - **Infrastructure Vulnerability Count** — downward trend
     (proactive scanning keeps up with the AI-era 0-day pace).
   - **MTTR** — for platform-detected and platform-remediated incidents.
   - **Cloud Spend Reduction** — on pilot estates vs. the pre-Nova
     baseline.
4. **Integrate with externally owned development platforms — regardless
   of source.** Nova integrates with externally owned PDLC, SDLC,
   Agentic, and Citizen Developer platforms. Nova provides skills and
   MCP endpoints that help the developer or AI agent make their
   application production-grade. Regardless of the source, all intents
   to deploy to production go through the same rigorous controls,
   quality gates, attestation, and evidence stream.

**4 Anti-Goals (what Nova is NOT):**
1. Not a general-purpose AI agent platform.
2. Not a system that removes humans from accountability — only from
   normal operations.
3. Not an upstream development platform (no product backlogs, IDE, code
   authorship).
4. Not a replacement for the Product Development Lifecycle (PDLC).

> **Benefit:** the scope is explicit — Nova governs infrastructure and
> delivery, integrates with any upstream source through one validated
> contract, and measures success on four metrics a CTO can repeat back.

> **Speaker notes:** Objective #2 is the one to land carefully: trust is
> established by deterministic scoring, not by an LLM. The platform
> functions without AI. Anti-goals #3 and #4 protect the scope boundary —
> Nova will not become an IDE or a product-planning tool.

> **Transition:** "The scope boundary is explicit — here is exactly
> where Nova sits relative to the product development lifecycle."

---

## Slide 4 — Scope: Downstream of PDLC

**Nova governs infrastructure and delivery. The PDLC is upstream — Nova
never penetrates it. Integration is through one validated contract.**

- **The PDLC is upstream:** product backlog, code authorship (AI agent,
  IDE, agentic SDLC), sprint planning, application business logic.
- **Nova is downstream:** contract ingestion → submission-readiness gate
  → policy enforcement → cloud resource lifecycle → environment
  progression (dev → qa → prod → dr) → immutable audit + attestation.
- **The integration point is one contract.** The citizen developer's AI
  coding agent, an upstream agentic SDLC platform, or any development
  platform may all produce submissions — the source does not matter
  because all are subject to the same compliance standards.
- **Nova validates the submission, not the author.** The audit trail is
  the same; the policy envelope is the same; the evidence stream is the
  same.

> **Benefit:** a clean scope boundary — Nova is purpose-built for
> infrastructure operations and integrates with any upstream source
> through one validated contract, so the platform team's surface area
> stays bounded.

> **Speaker notes:** This slide protects the scope. The moment Nova
> starts owning the PDLC, it loses focus. The contract boundary is what
> keeps Nova deep on infrastructure and delivery rather than shallow on
> everything.

> **Transition:** "With the scope clear, here is who owns what across the
> delivery lifecycle."

---

## Slide 5 — RACI: Who Owns What

**Four roles, one matrix — the citizen developer owns FRs + UAT, the
platform owns NFRs + infra, quality engineering owns the gate evidence,
and SRE owns operational readiness.**

| Work Category | Citizen Dev | Platform | Quality Eng | SRE |
|---|---|---|---|---|
| Functional Requirements | **R/A** | C | I | I |
| User Acceptance Testing | **R/A** | C | I | I |
| Non-Functional Requirements | I | **R/A** | C | C |
| Infrastructure (cloud, state, IAM) | I | **R/A** | I | C |
| QA (policy, confidence, schema) | C | R | **R/A** | I |
| Production deployment to cloud | I | **R/A** | C | C |
| Quality attestation (QA sign-off) | **A** | R | **R** | I |
| Production readiness (SRE sign-off) | **A** | R | C | **R** |

**R** = Responsible · **A** = Accountable (sign-off) · **C** = Consulted · **I** = Informed.

- **Compliance-standard equivalence:** FRs + UAT may come from any
  upstream source (AI agent, agentic SDLC, dev platform) — all pass the
  same submission-readiness gate.
- **Production readiness is co-owned:** the platform runs the
  attestations agentically; the citizen developer authorizes the
  promotion at the stage gate.

> **Benefit:** every party knows what they bring, what the platform
> provides, what quality engineering guards, and where SRE signs off —
> accountability is explicit, never diffuse.

> **Speaker notes:** Quality attestation is now owned by Quality
> Engineering (not the Platform), and Production readiness is owned by
> SRE. The Platform runs the checks agentically but is never the
> Accountable party for the gate — that separation keeps the platform
> honest.

> **Transition:** "With ownership clear, here is how the pipeline
> enforces it."

---

## Slide 6 — The Platform Pipeline

**How intent becomes verified infrastructure — with fail-fast policy
scanning before the plan and runtime scanning after it.**

```mermaid
graph LR
    A[Contract] --> B[Resolver]
    B --> C[Adapter]
    C --> D["Checkov (static code)"]
    D --> E[Terraform Plan]
    E --> F["Wiz (on plan)"]
    F --> G[Confidence Signal]
    G --> H{Stage Gate}
    H -->|dev: autonomous| I[Apply]
    H -->|qa/prod/dr: attested| I
    I --> J[Evidence + Ledger]
```

- **Contract → resolver → adapter → Checkov on static code (before the
  plan) → terraform plan → Wiz on the plan → confidence signal → stage
  gate → apply → evidence + ledger.**
- **Fail-fast, quick feedback.** Checkov runs on the authored Terraform
  code before `terraform plan` so developers get immediate policy
  feedback, not a delayed plan-stage failure.
- **Wiz on the plan when configured; Checkov as a drop-in otherwise.**
  Wiz scans the terraform plan output. When Wiz credentials are not
  available, Checkov runs against the plan as a drop-in replacement. Wiz
  and Checkov are never both run on the plan.
- **Dev is autonomous** (no stage gate); **qa/prod/dr require human
  attestation** (QA for quality, SRE for production readiness).

> **Benefit:** the pipeline gives developers fast, deterministic feedback
> on policy at authoring time and gives the platform a runtime scan on the
> resolved plan — two layers of scanning, zero operator involvement in
> normal operations.

> **Speaker notes:** The two-stage scan is the key design: static code
> scanning catches policy violations before the cost of a plan; runtime
> plan scanning catches what the static code cannot (resolved values,
  cross-resource issues). The platform picks the runtime scanner based on
  configuration — never both, to avoid duplicate noise.

> **Transition:** "The pipeline produces decisions — here is how every
> decision is captured and made accountable."

---

## Slide 7 — The Decision Ledger

**Every automated decision is captured, immutable, queryable — and
accountable.**

- **What is captured:** every action the platform takes — the chosen
  action, the confidence score, the alternatives considered, whether a
  human overrode it, and the outcome (backfilled once the apply
  completes). Every stage-gate attestation (QA sign-off, SRE
  production-readiness sign-off) is captured with approver identity and
  the evidence that was presented.
- **"AI decisions" are really automated decisions.** The decisions are
  made by deterministic scripts that calculate a score and a band; the
  platform functions without AI. The ledger captures the real decision
  path — not a fabricated "AI agent." When an LLM planner is added later,
  it will emit richer alternatives without breaking the schema.
- **The value is accountability, not the storage engine.** The ledger is
  an append-only, tamper-evident record. The point is not which database
  it lives in — the point is that every decision is queryable for
  auditing, traceable to an outcome, and impossible to rewrite after the
  fact.

> **Benefit:** "autonomous" is defensible because every decision the
> platform makes is immutable, queryable, and accountable — and the
> audience knows exactly what "automated" means here: deterministic
> scoring, not a black-box LLM.

> **Speaker notes:** Do not dwell on the storage substrate. The audience
> cares that the ledger is append-only, queryable, and tied to outcomes —
> not that it is a hash-chain in a SQLite file. The D-122 honesty point
> is restated without the decision ID: the platform's decisions are
> deterministic; the ledger captures that real path.

> **Transition:** "Decisions are captured — here is how stage-gate
> attestation keeps humans in accountability."

---

## Slide 8 — The Attestation Matrix

**The designed controls that keep humans at stage gates — structured,
freshness-validated, and separation-of-duties-enforced.**

| Concern | Env | Freshness | Description |
|---------|-----|-----------|-------------|
| Functional correctness | qa | 24h | The application behaves as specified; evidence accepted from the consumer's UAT. |
| Performance baseline | qa | 7d | The deployment meets its performance envelope vs. the agreed baseline. |
| Security posture | qa | 24h | The deployment's security findings have been reviewed and accepted. |
| Operational readiness | prod | 30d | SRE confirms the deployment is operable: runbooks, dashboards, on-call coverage. |
| Incident response | prod | 90d | The on-call path has been exercised; the deployment has a working incident-response plan. |
| Capacity & cost | prod | 30d | Capacity headroom and monthly cost are within the agreed envelope. |
| Resilience: DR drill | prod | 180d | A DR drill has been run and the deployment recovered within the RTO. |
| Resilience: chaos | prod | 90d | A chaos exercise has been run and the deployment absorbed the failure. |
| Resilience: backup | prod | 30d | Backups are restorable and have been tested within the freshness window. |
| DR region deploy | dr | 180d | The DR region can be deployed and the deployment is reachable from it. |

- Each concern has a freshness window — evidence older than the window
  does not satisfy the gate.
- **Separation-of-duties on prod:** the approver cannot be the same
  person who built the deployment.
- Concerns that are offline-testable run for real; concerns that require
  external evidence accept signed artifacts.

> **Benefit:** the gate model is explicit — autonomy in operations,
> human in accountability, by design. The matrix is what makes autonomous
> operations safe enough to trust in production.

> **Speaker notes:** The matrix is not a rubber stamp. Each concern has a
> freshness window, a description, and a separation-of-duties rule. The
> "operator-supplied" label from the prior deck was dropped — every
> concern now has a plain-language description of what is being attested.

> **Transition:** "You've seen how Nova works — the pipeline, the ledger,
> the attestation gates. Here is how Nova instruments itself so that
> every claim in this deck is traceable to a real signal."

---

## Slide 9 — Telemetry & Live Ops

**Every metric in this deck is traceable to a real emitted signal — and
the live-ops dashboard makes operations visible in PowerBI.**

```mermaid
graph TB
    A[Platform components] --> B[CloudEvents envelope]
    B --> C[Event log]
    B --> D[Decision ledger]
    B --> E[Run records]
    C --> F[Collector]
    D --> F
    E --> F
    F --> G[Cold store]
    G --> H[PowerBI views]
    H --> I[Live ops dashboard]
```

- **Platform components emit a CloudEvents envelope** → event log,
  decision ledger, and run records → collector → cold store → PowerBI
  views → **live ops dashboard.**
- **The live ops dashboard (PowerBI)** surfaces the four CTO-grade
  metrics — Lead Time, Infrastructure Vulnerability Count, MTTR, Cloud
  Spend — alongside the trust metrics (Decision Ledger coverage,
  Attestation coverage) and the efficiency metrics (touchless
  resolution, escalation frequency).
- **The architecture is deliberately minimal.** Nova-native envelopes;
  no Kafka, no Prometheus, no ClickHouse. The cold store is sufficient
  for batch and historical analysis; the live-ops surface is built in
  PowerBI on top of the exported views.
- **Every number in the Proof slides is traceable to a signal.** When a
  CFO asks "where does this number come from?", the answer is a query
  against the cold store, not a Slack thread.

> **Benefit:** the architecture is the trust substrate — leadership sees
> the same numbers the platform produces, in PowerBI, with full
> traceability to the emitted signal. Operations become visible.

> **Speaker notes:** The value is not the plumbing — it is that the
> platform's metrics surface in a tool leadership already uses (PowerBI),
> and every number is traceable. The live-ops dashboard is where the
> "infrastructure operations become visible" theme lands concretely.

> **Transition:** "The architecture is sound — here is the measured
> proof."

---

## Slide 10 — Decision Ledger + Attestation Coverage

**By design, no change reaches production without a ledger entry and a
human attestation — both queryable for auditing, with full
traceability.**

- **Decision Ledger coverage: 100%.** Every platform run emits a
  decision record with outcome backfill. No automated decision is ever
  lost.
- **Attestation coverage: 100%.** Every prod/dr promotion is attested by
  a human — QA for quality, SRE for production readiness — recorded with
  approver identity, separation-of-duties check, and the evidence matrix.
- **No change to production without both.** The ledger entry and the
  human attestation are mandatory, not optional. This is enforced by the
  pipeline, not by policy.
- **Easily queried for auditing.** The ledger and the attestation
  records are queryable by run, by environment, by approver, and by
  outcome — the audit trail is a query, not a forensic exercise.
- **Full traceability.** A production change is traceable from the
  contract that declared intent, through the policy scan, the confidence
  score, the attestation, to the applied outcome. Nothing is opaque.

> **Benefit:** trust is provable — not a marketing claim, a queryable
> record. An auditor can answer "who approved this, when, on what
> evidence?" in one query; a CTO can answer "how many of last quarter's
> prod changes were touchless?" in one query.

> **Speaker notes:** The mandatory-by-design point is the one to land.
> The ledger + attestation are not a best-effort feature; they are a
> gate. No change reaches production without both. That is what makes
> the 100% numbers credible — they are enforced, not aspirational.

> **Transition:** "Trust is provable — here is the cost side of the ROI."

---

## Slide 11 — Cost & ROI

**The ROI formula and the cost estimates — grounded, with the production
denominator honestly flagged.**

- **Cost estimates are pre-apply and offline.** The platform reads the
  terraform plan and estimates cost before anything is applied — so a
  regression in cost is caught before the spend happens, not after.
- **The ROI formula:**
  `Platform ROI = (FTE hours saved × blended rate + cloud savings + avoided downtime) ÷ platform op cost`
- **The four CTO-grade metrics (from Slide 3) are the ROI proof:**
  Lead Time (PR → Prod), Infrastructure Vulnerability Count (trend), MTTR,
  Cloud Spend Reduction. All flow into PowerBI.
- **Honest caveat:** the derived metrics are computed on internal runs
  today; the production-denominator activates when a pilot estate runs.
  The formula is grounded; the production numbers are not yet.

> **Benefit:** the ROI is not a black box — the formula is shown, the
> four metrics are committed, and the production-denominator caveat is
> stated up front. The CFO can see exactly what is real today and what
> activates with a pilot.

> **Speaker notes:** The formula is shown inline, not hidden. The
> "no fabrication" constraint in action: show the formula, show the
> caveat, do not pretend the production numbers exist.

> **Transition:** "The proof is grounded — here is what is honestly
> deferred, and why."

---

## Slide 12 — What's Deferred — and Why

**Honesty about what is not measured yet — and the blocking work for
each.**

To be clear: these deferrals are measurement infrastructure, not the
autonomy itself. The platform runs without an operator in the loop of
normal operations. What is deferred is the evidence pipeline for certain
metrics — not the autonomy.

| # | Deferred metric | Blocking work |
|---|-----------------|---------------|
| 1 | Live infrastructure health | Live AWS re-provisioning (currently torn down to a zero-cost steady state) |
| 2 | Live outbox write rate | Live AWS re-provisioning |
| 3 | Tamper-evident ledger checkpoints | Audit-ledger build-out (S3 Object Lock + signed checkpoints) |
| 4 | Onboarding funnel (requested → granted) | Auto-grant implementation |
| 5 | Drift auto-reversal | Drift-detection scheduler (not yet built) |
| 6 | Live cost reconciliation | Live AWS re-provisioning + actual-spend feed |
| 7 | SLA / unplanned downtime | Live AWS re-provisioning |
| 8 | Predictive vs reactive ratio | ML anomaly-forecasting service (not yet built) |

> **Benefit:** the boundaries are explicit — what Nova measures today,
> and exactly what blocks the rest. The autonomy is real; the measurement
> gaps are documented with the work that unblocks each one.

> **Speaker notes:** The preempt is critical: these deferrals are
> measurement infrastructure, not autonomy. The platform runs without an
> operator in the loop. What is deferred is the evidence pipeline for
> live-infra health, drift, predictive remediation — not the autonomy
> itself.

> **Transition:** "The proof is honest — here is the roadmap from here to
> the targets."

---

## Slide 13 — Roadmap to the North Star

**The path from the grounded metrics to the 12–18 month targets — each
deferred metric has an unblock path and a candidate milestone.**

| Timeframe | Work | Unblocks |
|-----------|------|----------|
| Near-term | Live AWS re-provisioning | Live infra health, live outbox write rate, live cost reconciliation, SLA |
| Near-term | Auto-grant implementation | Onboarding funnel (requested → granted) |
| Mid-term | Drift-detection scheduler | Drift auto-reversal |
| Mid-term | Audit-ledger build-out (Object Lock + signed checkpoints) | Tamper-evident ledger checkpoints |
| Mid-term | Hot-path activation (live-ops dashboard goes from batch to near-real-time) | Live-ops dashboard freshness |
| Longer-term | ML anomaly-forecasting service | Predictive vs reactive ratio |

- Each deferred metric has a specific unblock requirement and a
  candidate future milestone.
- Re-evaluation triggers: each blocking piece of work lifts on its own
  schedule; the metrics layer evolves as each one lands.

> **Benefit:** every deferred metric has an unblock path — nothing is
> hand-waved; everything has a plan and a timeframe.

> **Speaker notes:** This is the bridge from "honestly deferred" to
> "here is how we get there." The roadmap uses timeframes, not status —
> most of it is not implemented yet, so a status column would be noise.

> **Transition:** "The unblock path is clear — here is the 12-month
> product arc."

---

## Slide 14 — 12-Month Product Roadmap

**The product arc from pilot activation to integration — four quarters,
four outcomes.**

| Quarter | Theme | Board-level outcome |
|---------|-------|---------------------|
| **Q1** | Pilot Activation | Nova runs a real customer estate end-to-end, autonomously, with a measurable zero-touch rate. |
| **Q2** | Provable Trust | Every automated decision lands in a tamper-evident ledger; the CFO sees real cloud-spend reconciliation. |
| **Q3** | Compounding ROI | Quarter-over-quarter cloud spend drops; drift is detected and reversed without a human. |
| **Q4** | Integration & Predictive | AI agents deploy through Nova by default; the ML anomaly-forecasting service goes live. |

Grounded in the four strategic objectives (autonomy, provable trust, ROI,
integration) and the deferred-metric unblock paths.

> **Benefit:** the 12-month product arc — each quarter activates a
> strategic objective and its corresponding board-level metric, from
> pilot activation through integration leadership.

> **Speaker notes:** The roadmap is organized by product outcome, not
> by technical milestone. Each quarter activates one strategic
> objective from the North Star.

> **Transition:** "Here is the quarter-by-quarter detail."

---

## Slide 15 — Quarter-by-Quarter Outcomes

| Quarter | Product theme | Key deliverable | Target metric | Grounding |
|---------|---------------|-----------------|---------------|-----------|
| **Q1** | Pilot Activation | Re-provision live AWS; activate first pilot estate; onboarding auto-grant | Touchless ≥ 99% · Escalation < 0.1% · Accuracy ≥ 99.5% | Objective #1 — autonomy as the default |
| **Q2** | Provable Trust | Tamper-evident ledger (Object Lock + signed checkpoints); daily checkpoints; live cost reconciliation | Decision Ledger Coverage 100% · Cost Savings ≥ 25% | Objective #2 — trust is the moat |
| **Q3** | Compounding ROI + Drift | Drift-detection scheduler; auto-reversal; pre-apply → actual-spend reconciliation on the pilot estate | Drift Auto-Reversal ≥ 95% · Spend Reduction ≥ 25% | Objective #3 — CFO-pointable numbers |
| **Q4** | Integration + Predictive | ML anomaly-forecasting; AI-agent intent surface; multi-cloud (Azure/GCP) preview | Predictive:Reactive ≥ 3:1 · AI-Agent Intent Share (first measurement) | Objective #4 — default substrate for agents |

**Month-18 destination:** *"Nova is the layer enterprise leadership
points to when they say 'we don't have an infrastructure ops team
anymore, and the audit trail is stronger than it ever was.'"*

> **Benefit:** each quarter has a concrete deliverable, a target metric
> grounded in a strategic objective, and a path from "honestly deferred"
> to "shipped and measured."

> **Speaker notes:** Q1–Q3 are committed (grounded pipeline + known
> unblock paths). Q4 targets are committed-deliverable,
> aspirational-metric — the ML service ships, the intent-share number is
> a first measurement (we do not control adoption rate).

> **Transition:** "Production-grade guidance is how Nova helps the
> citizen developer's AI agent meet the bar — here is the first half."

---

## Slide 16 — Production-Grade Guidance via Atelier (1/2)

**Nova instructs the citizen developer's AI agent on production-grade
engineering — a set of skills and an MCP server.**

- **Skills** — markdown files keyed to production-grade engineering
  domains (API, security, data, testing, observability, errors, DevOps,
  infrastructure-as-code, compliance). The skills extend the baseline
  catalog with Nova-specific production-grade principles.
- **MCP server** — a plugin-registry, stdio server exposing four tools:
  `lookup_principle`, `list_domains`, `matrix_lookup`, and
  `validate_against_principles`. The developer's AI agent (or any
  agentic SDLC platform) calls these tools to look up the principles
  that apply to its submission.
- **The integration point is the same regardless of source.** Whether
  the submission comes from an AI coding agent, an agentic SDLC
  platform, or a traditional IDE, the same skills and MCP server apply.
  This is how Nova makes the citizen developer production-grade without
  owning the PDLC.

> **Benefit:** the citizen developer's AI agent is not unguided — Nova
> provides production-grade engineering principles as skills and as an
> MCP surface, so submissions arrive at the contract boundary already
> aligned with the platform's standards.

> **Speaker notes:** This is the first half of the Atelier story — the
> surface (skills + MCP). The next slide is what the surface catches
> that deterministic scanners cannot.

> **Transition:** "Here is what that guidance catches that deterministic
> scanners cannot."

---

## Slide 17 — Production-Grade Guidance via Atelier (2/2)

**Agentic validation catches engineering-discipline gaps that deterministic
scanners miss — and the validation is reproducible.**

- **Beyond deterministic scanners.** Wiz, Checkmarx, and Mend check
  policy and secrets — they do not check engineering discipline. The
  Atelier MCP server catches correctness, clarity, and observability gaps
  that deterministic tools cannot: "is this service observable?",
  "is this error path handled?", "is this API contract clear?"
- **Agentic validation, not a second policy engine.** The MCP server
  gives the AI agent the principles to validate against; the agent does
  the validation. This is agentic validation — the agent reasons about
  the submission against the principles, not a second static scan.
- **Vendored for audit reproducibility.** Atelier is vendored at a
  pinned tag. A validation result is replayable against the exact
  principles that produced it — so an audit can reproduce a validation
  months later, not just trust a log line.

> **Benefit:** the citizen developer's submission is checked for
> engineering discipline, not just policy compliance — and the check is
> reproducible for audit. That is what makes the submission
> production-grade, regardless of which upstream platform produced it.

> **Speaker notes:** The value is the gap deterministic scanners leave:
  engineering discipline. Policy scanners catch "is this S3 bucket
  public?"; the MCP server catches "is this service observable if that
  bucket fails?". The vendoring point is audit reproducibility — the
  validation is not a black box.

> **Transition:** "You've seen the problem, the solution, and the proof.
> Here is the recap and the ask."

---

## Slide 18 — Recap + Ask

**The 4-beat recap + the business decision.**

**Recap:**
- **Problem:** product teams own infrastructure without the discipline
  and lifecycle planning it requires; bandwidth gaps and tribal
  knowledge leave operations exposed.
- **Solution:** autonomous cloud delivery — operations become visible,
  trust is provable (deterministic scoring), humans at stage gates.
- **Proof:** 100% ledger coverage, 100% attestation coverage, grounded
  ROI formula, four CTO-grade metrics flowing into PowerBI.
- **Roadmap:** deferred metrics have unblock paths; the 12-month product
  arc activates one strategic objective per quarter.

**The ask:** "Approve a pilot estate to activate the production-denominator
metrics (Lead Time, Vulnerability Count, MTTR, Cloud Spend), and approve
the tamper-evident ledger build-out to move from the local hash-chain to
S3 Object Lock + signed checkpoints. These two decisions move Nova from
'pipeline-ready' to 'production-proven.'"

> **Benefit:** a clear business decision — approve a pilot and the ledger
> build-out — with the confidence that every claim in this deck is
> grounded, derived, or honestly deferred.

> **Speaker notes:** The ask is a business decision, not insider
> language. "Approve a pilot estate" is a C-suite decision. "Approve the
> ledger build-out" is a budget decision. The recap reinforces the 4-beat
> arc — the audience leaves with the structure, not a pile of facts.

---

## Appendix A1 — Metrics Glossary

| KPI | Definition | Status |
|-----|-----------|--------|
| Touchless Resolution Rate | runs without operational stage-gate block ÷ total | partial (Post-Pilot) |
| Human Escalation Frequency | operational stage-gate blocks ÷ total | partial (Post-Pilot) |
| Automated Decision Accuracy | decisions not followed by failure within 5min | partial (Post-Pilot) |
| MTTR (p95) | apply.failed → successful retry | grounded |
| Confidence-Gate Halt Rate | runs with band=block ÷ total | grounded |
| Provisioning Lead Time | run.completed − run.started | grounded |
| Deployment Frequency | count(run.completed) per day | grounded |
| Cost Savings (pre-apply) | sum(delta_usd where delta < 0) | partial (live reconciliation deferred) |
| FTE Hours Saved | run count × manual baseline × rate | derived (N=0 caveat) |
| Platform ROI | (labor + cloud + avoided downtime) ÷ op cost | derived (N=0 caveat) |
| Decision Ledger Coverage | decisions with outcome ÷ total | grounded |
| Attestation Coverage | prod/dr attested ÷ total prod/dr | grounded |
| Policy Compliance Rate | 1 − failed_assets ÷ total | grounded |

> **Benefit:** a reference for every metric mentioned in the deck.

---

> **End of deck.** 18 main slides + 1 appendix slide = 19 total.