# How The Platform Works > **Subtitle:** Agentic Cloud Delivery Platform > **Audience:** Senior Leadership, CTO, Head of Cloud, Head of Infrastructure, Head of DevOps > **Length:** ~15 minutes · 10 main + 6 appendix = 16 slides > **Purpose:** Sell the platform's value to tech leadership — zero-trust, security, observability, auditability, and the shift from "operators guess" to "the platform computes safety." > **Maturity framing:** "Testing" = works internally, dev pilot-ready. "Planned" = on the roadmap, not yet implemented. "Agentic" = involves AI agents or autonomous decision-making. > **Re-verification (2026-07-27):** Every "Testing" claim in this deck was re-verified in v1.10 Phase 54 (D-093). The headline E2E (contract → resolver → adapter → terraform init/validate/plan) passes against the live AWS account; the local emulating tier (Phase 53) runs the full E2E with no cloud credentials. 16/16 auto-verifiable capabilities Verified; 6 IAM-gated cloud resources are escalated (require an admin principal the spike-runner lacks). See `.ciagent/CAPABILITY_INVENTORY.md`. --- ## Slide 1 — Title # How The Platform Works ### Agentic Cloud Delivery Platform > **Speaker notes:** Brief introduction — this deck explains *how* the platform works internally, not what the developer experience is (that's the companion deck). Set the frame: the platform is not a CI/CD tool — it's the organizational lever for shipping safely at the pace the business demands. --- ## Slide 2 — The Problem & The North Star Here's the problem we're solving and where we're going. Software delivery scales with the **coordination surface around it**, not the engineering inside it. Most teams can write code; far fewer get the infrastructure right. Four frictions slow every team: - **Cognitive load** — authoring infrastructure correctly; the long tail of services inconsistent in security and observability. - **Operational work** — promoting a change from "merged" to "running in production." Manual work that **scales with the system, not the change.** - **Red tape** — tickets, approvals, and handoffs that scale with the organization. A merged change waits in a queue. - **Scalability without increasing headcount** — throughput scales without linearly scaling platform engineers. > Consumers **declare intent**; the platform delivers **safe production deployment** — automatically, safely, with a complete audit trail. - A merged change progresses **without a platform engineer joining a thread or approving a ticket.** - A **non-technical consumer** ships by declaring intent — no workflow, no config file, no infrastructure module. - Every production change is **traceable to a human attestation and an immutable evidence stream.** > **Speaker notes:** Open with the cost of the status quo. Every team that stands up its own pipeline, its own Terraform, its own review checklist is paying a tax that doesn't differentiate the business. The platform absorbs all four frictions — that is the value proposition in one sentence. Land the North Star quote: "declare intent → safe production deployment." The litmus test: if a platform engineer still has to touch a ticket for a dev→qa promotion, we haven't delivered the vision. --- ## Slide 3 — Where ACDL Sits in Your World Now that we know the problem, here's where ACDL fits — and where it doesn't. Now that we know the problem, here's where ACDL fits — and where it doesn't. - **Upstream is anything** — your IDE, an agentic SDLC, or a citizen developer vibe coding on a laptop. ACDL doesn't care how the contract was produced. - **ACDL is infrastructure only** — it provisions and governs AWS resources. It does not build, test, or deploy your application code. That's upstream. - **Not a general-purpose AI** — autonomy is narrow, scoped to delivery, bounded by strict policy envelopes. - **Not a permissive delivery highway** — no escape hatches to bypass the confidence framework or human attestation requirements. > **Speaker notes:** This slide gives leadership the framing they need. The platform is deliberately scoped — it is not trying to be everything. The sovereign boundary means the platform team owns delivery and infrastructure, not the upstream development process. The anti-goals are as important as the goals: they tell leadership what not to expect. --- ## Slide 4 — The Contract-Driven Model The contract is the boundary between upstream and ACDL. It's all a consumer writes. The contract is the boundary between upstream and ACDL. It's all a consumer writes. A single YAML contract — **module, environment, inputs**. The platform owns everything else. ```mermaid flowchart LR A["Consumer
writes a contract"] --> B["Platform resolves,
compiles, checks,
deploys, records"] B --> C["Resources running in AWS
+ tamper-evident evidence"] ``` The contract names three things: - **Which module** — a catalog of pre-built, security-reviewed building blocks (a static site, a microservice, a database, and more). - **Which environment** — `dev`, `qa`, `prod`, or `dr`. The platform raises the safety bar automatically as the environment gets more sensitive. - **Which inputs** — infrastructure values that vary per deployment (cpu, memory, port, desired_count). - The consumer provides **no AWS account, no VPC, no state backend** — the platform owns the blast radius. The consumer does **not** write infrastructure modules, workflow logic, or adapter code. They declare intent; the platform reconciles, provisions, and progresses. > **Speaker notes:** Emphasize the asymmetry. The consumer's surface is intentionally tiny — a contract that fits on one screen. The platform's surface is large and opinionated. That asymmetry is what makes "declare intent, not execute operations" concrete. Note that the contract examples now show infrastructure inputs (cpu, memory, desired_count, port) — not a container image. The image is upstream; the platform governs infrastructure. --- ## Slide 5 — The End-to-End Flow Once the contract is written, here's what the platform does with it — every time. Once the contract is written, here's what the platform does with it — every time. Every deployment runs the same stages, in the same order, with the same checks — no team-specific pipelines, no tribal runbooks. ```mermaid flowchart TD A["Consumer contract
(module + environment + inputs)"] --> B["Validate contract
against the schema"] B --> C["Resolve to a target stack
(expand the module's pattern)"] C --> D["Security checks
(before any infra is created)"] D --> E["Infrastructure plan
(platform compiles the stack)"] E --> F["Policy checks
(normalized results)"] F --> G["Confidence signal
(6 inputs → score + band)"] G --> H["Evidence event
(hash-chained, tamper-evident)"] H --> I["Infrastructure apply
(dev only — higher envs hold for attestation)"] ``` Two properties matter to leadership: - **Security and policy checks run *before* any infrastructure is created** — not after the fact, not as a post-deployment audit. - **Every stage produces a record** that feeds the confidence signal and the evidence stream. There is no "unchecked" path. > **Speaker notes:** Walk left to right once. Don't dwell on internals — the point is that the flow is fixed, opinionated, and identical for every consumer. The two leadership-relevant beats are (1) checks before creation, (2) every stage is evidenced. The confidence signal (Slide 7) is where the "safety is computed" story lands. --- ## Slide 6 — Zero-Trust by Default Before any infrastructure is created, here's how access is scoped. Before any infrastructure is created, here's how access is scoped. Consumer repositories hold **no long-lived cloud credentials.** Ever. - **Authentication is OIDC federation** between the platform runners and the cloud provider. Each job mints a short-lived token; no credential is stored in the consumer repo or in a runner secret. *(Testing on GitHub Actions runners; planned for all platform runners.)* - **Authorization is attribute-based (ABAC), not role-based.** Two attribute classes scope every action: - **Repository identity** — the role's trust policy binds to the exact consumer repo + branch that invoked the workflow. - **Resource-creation attributes** — every resource is tagged with `acdl:owner=` and `acdl:contract=`. The session policy grants view/update/delete **only on resources whose tags match the calling repo.** **The effect:** a consumer's pipeline can only touch the resources it created. Blast radius is contained to that consumer's own stack instances. One consumer can never touch another's resources, and the consumer cannot escape its own scope. > **Speaker notes:** This is the slide for the Head of Cloud/Security. The key phrase is "blast radius contained to the consumer's own stack." Contrast with the common failure mode of shared CI roles that can touch any account resource. The static-key override exists for edge cases but is rotated daily on platform runners; it is never the default. --- ## Slide 7 — Safety is Computed, Not Assumed Now let's look at how the platform decides whether a deployment is safe. Now let's look at how the platform decides whether a deployment is safe. Every delivery action produces a **measurable, explainable confidence signal** — a weighted sum of observable facts, not a black box. *(Agentic.)* - **Six weighted inputs** — policy conformance, validation, freshness, source provenance, history, and non-functional requirements (NFRs). The weights are **manually tuned**, the inputs are **observable**, and the breakdown is **auditable** — if a consumer asks "why 0.62?", the platform answers with a per-input breakdown. - **Per-environment thresholds** that rise with sensitivity: | Environment | Threshold | Who must attest | |---|---|---| | dev | ≥ 0.50 | No one — fully autonomous *(Testing)* | | qa | ≥ 0.75 | QA *(Planned)* | | prod | ≥ 0.90 | SRE *(Planned)* | | dr | ≥ 0.95 | SRE + a disaster-recovery drill reference *(Planned)* | - **A single critical policy finding hard-blocks the deployment**, regardless of every other input. Critical findings are not averaged away. - **When the platform halts, it gives a measured reason** — a policy violation, an insufficient signal, a missing attestation — never an opaque, manual-debugging exercise. > **Speaker notes:** This is the bet that separates this platform from "yet another CI/CD tool." Reliance on operator instinct or tenure is not a substitute. The signal is auditable; the thresholds are tunable by Infra & Ops + SRE jointly, and any override is itself a confidence-event in the audit stream. Leadership cares about this because it makes promotion decisions *reviewable*. The new confidence signal diagram makes the six inputs and the per-input breakdown visible — emphasize that the weights are manually tuned and the breakdown is auditable, not a black box. --- ## Slide 8 — Security by Construction Beyond the confidence signal, security defaults are on by construction — not by opt-in. Beyond the confidence signal, security defaults are on by construction — not by opt-in. Security defaults that **do not require a team to opt in.** Checks run on **every** deployment, normalized to a single schema regardless of which engine produced them. *(Testing.)* - **Infrastructure-as-code policy** (Checkov) — secrets in plaintext, public ingress, IAM wildcards, KMS key references, **required tagging standards** (`acdl:owner`, `acdl:contract`, `acdl:environment`, `acdl:cost-center`). All run *before* infra is created. - **Cloud security posture** (Wiz adapter) — translates cloud security findings into the same normalized record. *(Adapter testing; activates when a Wiz tenant is configured.)* - **Kubernetes-native policy** (Kyverno adapter) — ready for the GitOps reconciler roadmap item. *(Adapter testing; inactive for Terraform-only stacks.)* - **Encryption on every resource** — at-rest encryption is on by default for every primitive (S3, RDS, ECR, ECS, and more). *(Testing.)* - **Per-stack customer-managed keys (CMKs)** — one key per deployment, 90-day rotation at creation, **no shared keys across stacks.** *(Testing.)* - **Managed-key fallback with a loud warning** — standalone primitives fall back to cloud-managed keys only when no CMK is provided, and the platform warns explicitly. *(Testing.)* - **Deletion protection on by default** — every resource has `prevent_destroy` on unless a consumer explicitly disables it via a documented feature flag. *(Testing.)* - **Safe decommission** — a 2-step pipeline (disable protection → zero counts → destroy) with **two SRE human-attestation gates** and a **change-request validated against the platform CMDB** before any destructive action. *(Testing.)* Encryption keys enter a grace window (default 30 days) so encrypted data remains recoverable during decommission. > **Speaker notes:** The phrase to land is "secure by default, not secure by effort." The selling point is *normalization* — we can add a new security tool without changing the confidence model or the evidence stream. For the Head of Security: tagging standards are enforced, not advisory — a missing `acdl:owner` tag fails the check, not a warning. The decommission flow is the counter-argument to "deletion protection makes cleanup impossible" — it's a deliberate, gated, two-approval path, not a lock with no key. --- ## Slide 9 — Accountability & Audit Computed safety handles the gate. But humans still matter — here's how accountability works. Computed safety handles the gate. But humans still matter — here's how accountability works. - **Dev is fully autonomous.** The confidence signal (≥ 0.50) is the only gate. Queue-based handoffs are eliminated from lower environments. *(Testing, Agentic.)* - **qa, prod, and dr require deliberate human attestation** — not rubber stamps, but policy-mandated acts of accountability via protected deployment approvals. The approver reviews the contract, the planned Terraform changes, and the accumulated evidence. *(Planned.)* - **QA attests to infrastructure readiness — the contract, the planned Terraform changes, and the accumulated evidence. QA does not review application code (that's upstream).** - **Separation of duties is enforced** *(Planned)* — the person who approved the qa promotion **cannot** be the person who approves the prod promotion. The platform reads both identities from the outbox and **blocks** on a match, emitting a `SEPARATION_OF_DUTIES_VIOLATION` and routing a halt artifact to SRE on-call. - **Timeout discipline** — 1 business day = warn + escalate; 2 business days = auto-freeze + re-submit. Rejection extends the audit chain; it does not tear it up. Version control is a **coordination tool, not an evidentiary fortress.** True compliance requires an immutable, externally-stored ledger. - **Every deployment writes a hash-chained evidence event** — each event links to the previous via a cryptographic hash. Tampering breaks the chain. *(Testing — the DynamoDB outbox.)* - **Tiered storage design:** cold, tamper-proof source of truth (S3 Object Lock, compliance mode, 7-year retention) + a hot query index for fast lookup. *(Outbox tested; S3 Object Lock + JWS detached signatures are planned regulatory-ledger build-out.)* - **RPO = 0** — the evidence write is synchronous; a deployment is not acknowledged until the evidence event is durably recorded. - **Every production change is traceable to a human attestation** — the QA and prod approver identities are the only durable record outside the VCS's audit log, stored in the outbox keyed by contract. > **Speaker notes:** The "lower environments autonomous, higher environments attested" tenet is the resolution to the classic "move fast vs. be safe" false dichotomy. Be honest: the separation-of-duties *mechanism* (CODEOWNERS routing, identity-distinctness check, the 8-concern attestation matrix) is designed and the dev path is wired; the qa/prod/dr wiring is on the roadmap. The new attestation flow diagram makes the human-in-the-loop path visible. Note the QA clarification: QA attests to infrastructure readiness — the contract, the plan, and the evidence — not application code. The audit trail is a byproduct of deployment, not a project. Note honestly that the full regulatory ledger (S3 Object Lock, JWS signatures, daily checkpoints) is planned; what ships today is the outbox + hash chain that makes every event tamper-evident and queryable. Badge reclassification: separation of duties = Planned (not "design tested"), dev autonomous = Testing, qa/prod/dr attestation = Planned. --- ## Slide 10 — The Vision Realized Here's what success looks like when the North Star is reached. Here's what success looks like when the North Star is reached. - **Velocity without sacrificing safety.** Speed is in the ergonomics (a simple contract, a one-line `uses:`); safety is in the gates the consumer cannot bypass. - **Security, observability, and compliance as platform defaults** — not per-team effort, not post-hoc remediation. - **Auditability as a byproduct, not a project.** Every production change is traceable to a human attestation and a tamper-evident evidence event. - **Blast radius contained by design.** Zero-trust OIDC + ABAC means a consumer can only touch its own tagged resources. - **Infrastructure as a utility, not a craft.** Teams consume infrastructure, they don't maintain it. - **A path to the citizen developer.** The same safety envelope that serves a senior engineer will serve a non-technical consumer. *(Agentic.)* > **Speaker notes:** Close on the strategic frame. The platform is not "a CI/CD tool," it's the organizational lever for shipping safely at the pace the business demands. The investment is in the abstraction, not the tool. Ship safely at the pace the business demands, with the security and audit posture the regulators require. --- ## Appendix — Table of Contents For deep dives — these slides cover details omitted from the main 10. **Contents:** 1. Platform-Managed Environments (detail) 2. Observability Built In (detail) 3. The Road to the North Star (phased roadmap) 4. Testing vs. Planned (full inventory) 5. Glossary > **Speaker notes:** These are deep-dive slides for follow-up questions. Don't walk them in the main 15-minute talk — pull them up when an audience member wants detail on a specific topic. --- ## A1 — Platform-Managed Environments For deep dives — these slides cover details omitted from the main 10. A consumer provides **no AWS account, no VPC, no subnet, no state backend, no runner key.** The platform owns the blast radius. A named environment is a platform-owned bundle of: - An AWS account (or a scoped partition of one). - A network (VPC + subnets). - A state backend (S3 + DynamoDB for infrastructure state + locking). - An IAM role surfaced to the consumer via ABAC, scoped to the consumer's repository identity and resource tags. The consumer selects an environment **by name** in their contract (`environment: dev`). The platform resolves the name to the underlying account/network/state/role at run time. **The consumer never sees the raw credentials.** **Friendly onboarding:** the first run detects no environment and emits a guided prompt (not an opaque failure) telling the consumer what the platform will provision and how to request it. *(Testing.)* **Self-service environment provisioning is planned.** > **Speaker notes:** For the Head of Cloud: this is the governance story. The platform team owns the accounts, the network design, the state hygiene. Consumers can't drift into misconfigured state backends or over-permissioned roles because they never touch them. The onboarding prompt matters — first impressions of a platform are made when it fails for the first time. --- ## A2 — Observability Built In Monitoring is **a platform default, not a per-team project.** *(Testing.)* - **Uptime monitoring deployed automatically with every stack** — a dedicated monitoring instance (Uptime-kuma on ECS Fargate) is provisioned after any module deploy, in a separate state, with a feature flag to disable. - **Monitored endpoints passed from the deployment's own outputs** — the platform constructs a synthetic monitoring contract from what was just deployed. No manual endpoint registration. - **Alert channels:** Microsoft Teams webhook, email, SMS, and GitHub issues. *(Testing.)* - **The uptime URL is published to the developer** via a PR comment — they don't hunt for it. - **Roadmap:** deeper observability bootstrap (dashboards, runbooks, on-call bindings) as first-class contract fields for prod/dr. *(Planned.)* > **Speaker notes:** The Head of DevOps cares about this. The framing: "you don't deploy a service and *then* remember to set up monitoring — the platform does it as part of the deploy." The feature flag means teams with existing monitoring (e.g. Datadog) can opt out cleanly. --- ## A3 — The Road to the North Star *Proposed phasing — not formally planned.* A phased roadmap from the current Testing baseline to the full North Star: - **Phase 1 — Testing baseline (current):** contract-driven deploys, zero-trust OIDC + ABAC on GitHub Actions, confidence signal gating, hash-chained evidence, encryption by default, deletion protection + safe decommission, uptime monitoring, platform-managed environments. - **Phase 2 — Production readiness:** HITL wiring for qa/prod/dr, all-runner OIDC, full regulatory ledger (S3 Object Lock + JWS signatures + daily checkpoints), environment self-service. - **Phase 3 — Compliance & expansion:** compliance milestone (GDPR, SOX, SOC2, DORA extension points), additional engine adapters (OpenTofu, Pulumi, Kubernetes CRDs), deeper observability bootstrap. - **Phase 4 — Agentic frontier:** dynamic module creation from a contract (the agentic citizen-developer composition mechanism), pattern recognition that compounds value over time. > **Speaker notes:** Be clear with leadership: this is a proposed phasing, not a formally committed plan. The phases are sequenced by dependency, not by calendar — each phase's items are gated on the prior phase's maturity. Invite questions on any phase boundary. --- ## A4 — Testing vs. Planned (Full Inventory) **Testing** (works internally, dev pilot-ready) — 11 capabilities: - Contract-driven deploys with a versioned reusable workflow. - Module catalog (primitives + modules) with validated examples. - Zero-trust OIDC + ABAC on GitHub Actions runners. - Security + policy checks before infra creation (Checkov; Wiz + Kyverno adapters ready). - Confidence signal (6 inputs, per-env thresholds) gating promotion. *(Agentic.)* - Hash-chained, tamper-evident evidence outbox (RPO = 0). - Encryption by default + per-stack customer-managed keys. - Deletion protection by default + safe decommission with SRE gates + CMDB validation. - Uptime monitoring deployed automatically with every stack. - Platform-managed environments + friendly onboarding. - Engine-agnostic core (1 adapter: Terraform) + VCS-agnostic ingestion (GitHub + Gitea). **Planned** (on the roadmap, not yet implemented) — 9 capabilities: - Real OIDC federation on all platform runners (Gitea Actions OIDC pending an upstream merge). - HITL wiring for qa / prod / dr environments (design shipped; wiring is next). - Full regulatory ledger: S3 Object Lock (7-yr compliance mode) + JWS detached signatures + daily checkpoints. - Compliance milestone: per-module extension points for GDPR, SOX, SOC2, DORA. - Environment self-service (a consumer-facing flow to request and provision a new environment). - Dynamic module creation from a contract (the agentic "citizen developer" composition mechanism). *(Agentic.)* - Pattern recognition compounds value over time. *(Agentic.)* - Additional engine adapters (OpenTofu, Pulumi, Kubernetes CRDs). - Deeper observability bootstrap (dashboards, runbooks, on-call bindings). > **Speaker notes:** Close on honesty. The platform delivers real, verifiable value today — 11 capabilities that work internally. The roadmap is concrete, not aspirational hand-waving — 9 planned items, each with a defined milestone and a clear reason it isn't shipped yet (usually an upstream dependency, not an engineering gap). Emphasize: 0 consumer adoption today — "Testing" means it works internally and is dev pilot-ready, not that it's released. --- ## A5 — Glossary | Term | Meaning | |---|---| | **OIDC** | OpenID Connect — federation protocol for short-lived tokens, no long-lived credentials | | **ABAC** | Attribute-Based Access Control — access scoped by resource tags + repo identity, not roles | | **CMK** | Customer-Managed Key — per-stack encryption key, 90-day rotation, no shared keys | | **CMDB** | Configuration Management Database — validates change requests for decommission | | **RPO** | Recovery Point Objective — RPO = 0 means evidence is written synchronously, no data loss | | **HITL** | Human-in-the-Loop — deliberate human attestation required for qa/prod/dr environments | | **VCS** | Version Control System — the git hosting platform (GitHub, Gitea, GitLab) | | **NFR** | Non-Functional Requirement — encryption, tagging, observability standards | | **IR** | Intermediate Representation — the engine-agnostic stack definition between contract and Terraform | > **Speaker notes:** Use this slide as a reference when the audience asks for term definitions. Don't read it aloud — point to it as a takeaway reference.