Public promise
ACE presents one coherent Intelligence OS: governed data enters, evidence-grounded intelligence is derived, people and machines consume it downstream, and attributable outcomes return to the system. Reasoning is a governed component of that lifecycle rather than the product category. Atrium is its permission-aware intelligence dashboard and human control plane.
Released inputs
0.8 begins from public ace-core==0.7.0 and its reconciled release evidence. It consumes rather than rebuilds:
- Connect → Map → Watch → Brief → Activate onboarding machinery;
- separate Connection, Ontology, Intelligence, and Briefing Agent services plus governed activation-plan admission;
- generated inert Domain Packs and conformance;
- Agent Composition through AC7; and
- authorized Agent Memory through AM3.
AM4, AM6, and other unmerged candidates are not implicit release dependencies. They require explicit compatibility review before entering 0.8.
Canonical lifecycle
evidence → observation → entity state → signal/shift → case/brief → decision/action → outcome → governed feedback
Core supplies governed mechanics. Intelligence owns the evidence-to-orientation lifecycle. Domain Packs supply declarative meaning and policy. Connectors, adapters, and strategy plugins supply explicitly enabled executable specialization. Applications, agents, workflows, and mature external products consume intelligence downstream.
Bounded delivery sequence
- 0.8A — Architecture and compatibility. Freeze ownership, canonical vocabulary, compatibility seams, the landing/rollback order, and the signature acceptance journey. Candidate: draft PR #127.
- 0.8B — Runtime realignment. Put generic planning, authorization, execution admission, assurance, and receipts behind Core; put Observation-to-Feedback behavior behind Intelligence; move artifact creation and external effects behind explicit adapters or strategies.
- 0.8C — Unified intelligence resources. Publish one product-scoped resource plane for sources, source health, entities, Observations, Signals, Shifts, Cases, Briefs, monitors, subscriptions, decisions, actions, outcomes, feedback, provenance, uncertainty, semantic revisions, Context Manifests, and memory lineage.
- 0.8D — Atrium experience. Rebuild the optional dashboard and human control plane around intelligence review, briefings, investigation, semantic diffs, governed controls, and Ask ACE over those same durable resources.
- 0.8E — Cross-domain proof. Run the public World AI Command Center and the parallel Market Intelligence command center through unchanged Core + Intelligence APIs.
- 0.8F — Release acceptance. Pass compatibility, migration, restart, installed-artifact, accessibility, entitlement, degraded-state, cross-domain, and trusted-publication gates.
Packets land independently in A → B → C → D → E → F order. Domain work may develop after A, but only domain-neutral requirements return to Core.
Public scope
- Realign the code, not only the language; preserve supported contracts through declared adapters, deprecations, and migrations rather than a mass rename.
- Replace MAKE/SHIP and arm-oriented canonical architecture with explicit action, delivery, assurance, authorization, adapter, and strategy boundaries.
- Define frameworks as versioned analytical-method contracts with applicability, required evidence, procedure, structured outputs, assumptions, limitations, evaluation criteria, and material-use receipts. Perspectives remain governed analytical lenses; models remain replaceable inference resources.
- Rebuild Atrium around attention and intelligence state. Retain whiteboard investigation as a Case workspace, not a second source of truth.
- Add an intelligence-review inbox, semantic revision diffs, Context Manifest inspection, memory lineage, source/intelligence libraries, and exact participant/stage attribution.
- Drive Atrium through the same public interfaces available to domain applications and machine consumers. Rendering state grants no authority.
- Replace the octopus, partner-team, product-manager-toolbox, Living Product Graph system category, committee-first, and arm-growth narratives across canonical documentation, diagrams, onboarding, APIs, tests, and product surfaces. Historical evidence names remain immutable.
- Preserve the exact eleven-tool thin MCP boundary unless a separate public-contract decision passes.
Signature demonstration
World Intelligence supplies an independently reproducible AI Command Center using changing authorized sources covering model/capability releases, token economics, cybersecurity, regulation/geopolitics, investment, market response, and attributable executive commitments. A user connects sources, edits the proposed ontology and monitors, receives a cited first Brief, activates the exact plan, observes a later Shift or explicit no-change result, asks a grounded follow-up, records a Decision or bounded Action and later Outcome, submits governed Feedback, restarts, and reproduces the same identities and lineage.
Market Intelligence must reproduce the same lifecycle with materially different sources, ontology, detectors, personas, epistemic policy, and cadence. HPE-specific vocabulary and sources remain a private organization overlay, not public pack or Core code.
Acceptance gate
- World and Market complete the same governed evidence-to-outcome lifecycle without a domain-specific branch in Core or Intelligence.
- Atrium and at least one machine interface consume the same durable intelligence through public contracts.
- The old arm/product metaphors are absent from canonical public and code architecture while supported compatibility remains tested.
- Accessibility, authority, entitlement, degraded-state, restart, failure, and migrations are exercised.
- Atrium remains an optional view and control surface, not a second source of truth or authority.
- Public artifacts, hashes, workflows, limitations, rollback, clean installation, and independent consumer evidence are reconciled before the milestone closes.
Explicit boundary
0.8 does not promise multi-tenant collaboration, arbitrary hostile-code sandboxing, marketplace, managed operation, universal source coverage, unrestricted autonomy, omniscience, or general real-world causal accuracy. Those remain later or explicitly out of scope.
See the 0.8 control packet, public roadmap, and live Project.
Public promise
ACE presents one coherent Intelligence OS: governed data enters, evidence-grounded intelligence is derived, people and machines consume it downstream, and attributable outcomes return to the system. Reasoning is a governed component of that lifecycle rather than the product category. Atrium is its permission-aware intelligence dashboard and human control plane.
Released inputs
0.8 begins from public
ace-core==0.7.0and its reconciled release evidence. It consumes rather than rebuilds:AM4, AM6, and other unmerged candidates are not implicit release dependencies. They require explicit compatibility review before entering 0.8.
Canonical lifecycle
evidence → observation → entity state → signal/shift → case/brief → decision/action → outcome → governed feedbackCore supplies governed mechanics. Intelligence owns the evidence-to-orientation lifecycle. Domain Packs supply declarative meaning and policy. Connectors, adapters, and strategy plugins supply explicitly enabled executable specialization. Applications, agents, workflows, and mature external products consume intelligence downstream.
Bounded delivery sequence
Packets land independently in A → B → C → D → E → F order. Domain work may develop after A, but only domain-neutral requirements return to Core.
Public scope
Signature demonstration
World Intelligence supplies an independently reproducible AI Command Center using changing authorized sources covering model/capability releases, token economics, cybersecurity, regulation/geopolitics, investment, market response, and attributable executive commitments. A user connects sources, edits the proposed ontology and monitors, receives a cited first Brief, activates the exact plan, observes a later Shift or explicit no-change result, asks a grounded follow-up, records a Decision or bounded Action and later Outcome, submits governed Feedback, restarts, and reproduces the same identities and lineage.
Market Intelligence must reproduce the same lifecycle with materially different sources, ontology, detectors, personas, epistemic policy, and cadence. HPE-specific vocabulary and sources remain a private organization overlay, not public pack or Core code.
Acceptance gate
Explicit boundary
0.8 does not promise multi-tenant collaboration, arbitrary hostile-code sandboxing, marketplace, managed operation, universal source coverage, unrestricted autonomy, omniscience, or general real-world causal accuracy. Those remain later or explicitly out of scope.
See the 0.8 control packet, public roadmap, and live Project.