Nova Atlas — a governed orchestration operating system.
This page is the canonical architecture transparency surface. Every section is generated from the live introspection endpoints — there is no marketing copy, no invented capabilities, no autonomous-execution claim. Nova Atlas is operator-supervised, advisory-only, lifecycle-aware, lineage-aware, and provider-truthful. Reviewers and enterprise teams can inspect every governance contract here without an account.
operator-supervised
advisory-only
orchestration-aware
lifecycle-driven
explainable by design
provider-truthful
Live architecture snapshot
/api/architecture/summary
Rendered directly from the live canonical introspection endpoint. If this section is empty, the underlying service is unavailable — Nova does not fabricate a fallback architecture.
Advisory recommendations only. Nova does not automatically change budgets, pause campaigns, or scale spend. All actions require operator approval.
Explainability by design
Example
Every advisory in Nova Atlas exposes a canonical [Why?] panel that lists rationale, contributing sources, lifecycle stage, freshness, orchestration context, and the split between estimated and confirmed inputs. The panel below is the actual production component — not a mockup.
Example advisory · compression
Example: Prime Day overlaps Spring Launch — operator review suggested.
Severity: warning · Confidence: medium
Orchestration memory
example chain
Every entity created in Nova Atlas writes a deterministic lineage edge into the orchestration memory SSOT. Operators can trace adaptation across channels — Meta → TikTok → Email → Landing Page → Video — without graph engines. The example below is static and clearly labeled; live data is brand-scoped and never exposed publicly.
origin · adapted_from
Meta Ad — Spring Launch Hero
Meta · Performance Data Ready · Meta Ads Builder
adaptation · iteration_of
TikTok Variation — Spring Hook
TikTok · iteration_of Meta hero · TikTok Ads Builder
adaptation · adapted_from
Email Sequence — Spring Welcome
Email · adapted_from winning Meta creative · Email Editor
Nova does not infer ancestry. When an entity has no recorded upstream, the operator surface honestly displays “No upstream derivation recorded.”
Operational intelligence
Example
Two observational surfaces operators read every day: provider freshness (the Spend Trust Meter) and weekly orchestration congestion. Both are advisory — Nova does not auto-shift budgets, launches, or campaigns. The bands and chips below are static examples.
Provider freshness · canonical taxonomy
Meta · healthyGoogle Ads · agingTikTok · stale
Four canonical states: healthy / aging / stale / never. Identical taxonomy across every provider; no provider-specific freshness wording.
Orchestration congestion · 6-week example
Wk 1
Low
Wk 2
Medium
Wk 3
High
Wk 4
Critical
Wk 5
Medium
Wk 6
Calm
Five canonical congestion states: calm / low / medium / high / critical. Cards and chips only — no calendar engines, no Gantt, no graph spaghetti.
Governance and safety contracts
The contracts below are enforced by the canonical lifecycle module (backend/lifecycle/canonical.py) and a permanent pytest lock suite. Every advisory, every builder, every studio, and every public response is validated against these contracts on every commit.
Operator-supervised
Every action that materially affects spend, deployment, or audience requires explicit operator approval. No autonomous execution.
Advisory-only
Every recommendation is flagged as advisory. Nova does not auto-adjust budgets, pause campaigns, scale spend, or shift launches.
Forbidden autonomous wording
Autonomous-execution language is blocked at the pytest contract level — phrases that imply unsupervised operation cannot ship in operator-facing copy.
Provider-truthful
Provider status uses canonical values (provider_confirmed / provider_error / not_submitted). UI truth badges never claim "Active" without a confirmed provider campaign id.
Lifecycle continuity
Every entity participates in an 8-stage canonical lifecycle (intent → orchestrate → produce → channels → approvals → performance → audit → governance).
Deterministic validation
The canonical strict validator (validate_recommendation_strict) is invoked on every advisory before it is surfaced.
Cross-channel lineage
Every produced asset writes a creative_lineage edge. The orchestration memory layer is brand-scoped and read-only on the operator side.
Public-safe
This /architecture page never displays brand identifiers, customer data, provider account ids, secrets, or mutation paths.
Advisory recommendations only. Nova does not automatically change budgets, pause campaigns, or scale spend. All actions require operator approval.