The Compounding Model 1.4.0 — the agentic platform

The platforms shipped their agent control planes; the invariants did not move.

• baseline/agentic-platform.md: agent identity, fleet registry, evaluations, guardrails, observability, knowledge — invariants first, Microsoft / AWS / GCP mappings after, including the one-socket dividend: if your estate speaks MCP, a control plane can govern it by registration, not re-architecture.
• ADR-017: adopt native where it meets the invariant, keep custom where it exceeds it — every call a decision, never drift. A2A: documented, not adopted, while orchestration stays on audited rails.
• Two identity classes, one attribution rule: interactive agents act as the person driving them; standing agents are registered workload identities.
• Evaluations: continuous at a declared sampling rate, acceptance thresholds gate releases, native evaluator suites welcome alongside an honest judge.

CHANGELOG.md has the full inventory.

The Compounding Model 1.3.0 — VMs, data & analytics, agent-paced roadmap

Closing the open jobs an architect will not leave behind.

• Roadmap recalibrated to agent execution: hours of build; the calendar belongs to decisions, lead times and evidence.
• VMs covered where unavoidable: golden images, in-guest config as code, patch rings, JIT-only access.
• Data & analytics: notebooks are code through the gate; mandatory catalog; data tests gate promotion; freshness SLOs render 'stale since'. ADR-016 + full cloud mapping.

CHANGELOG.md has the full inventory.

The Compounding Model 1.2.0 — knowledge & work integration

The agent knows the project, and the ticket is the spine.

• The wiki of record and the work tracker become live tools on the one MCP socket: federated search, attributed comments, read-before-acting as contract.
• An action without a ticket never happened — enforced at the gate, reported by the orchestrator, joined into cost and telemetry by ticket id.
• Agents never open tickets unasked. ADR-015 seeds the decision; mappings for Atlassian / Microsoft / GitHub stacks.

CHANGELOG.md has the full inventory.

The Compounding Model 1.1.0 — the agent-first release

Your agent assesses you, stands up day one, and builds the factory itself.

• AI assessment: one prompt your agent runs against your real repos and pipelines, grading on cited evidence — https://savvaniss.github.io/compounding-model/
• Agent-first operating philosophy throughout: humans decide, agents execute, every execution leaves evidence.
• Cloud-agnostic: invariants as the standard, with Azure / AWS / GCP mappings for the foundation and Azure DevOps / GitHub / GitLab / AWS / GCP mappings for delivery.
• Secrets: the vault is the only home; everything else is a transport.

Full details in CHANGELOG.md.