AncientOS Governed Capability Portfolio
Status: canonical operator-facing roadmap view.
The adoption authority and detailed capability standard are the Governed Capability Portfolio roadmap. The canonical lifecycle and posture vocabulary is defined by the Architecture and Capability Posture Taxonomy. This page renders those authorities as one practical view for operators. It is not a second roadmap authority.
Current roadmap position
AncientOS completed its previously adopted numbered-era sequence through Era 8. Era 7 delivered Oracle Operational Ergonomics. Era 8 delivered the first Governed Capability Portfolio. Those names remain historical implementation programs and completion evidence.
There is no active numbered era. Era 9 is not adopted. AncientOS no longer advances by automatically opening another numbered era, stage, or phase. Future work enters the active portfolio only through an explicit adoption decision for a bounded capability.
Current portfolio state:
| Portfolio view | Current truth |
|---|---|
| Adopted work awaiting implementation | Keeper exact-task completion lifecycle consolidation |
| Work in progress | None; Media objective milestone completed |
| Blocked adopted work | None |
| Proposed work awaiting an adoption decision | None |
| Candidate work | Present; non-authoritative options are listed below |
| Deferred work | Present; reconsideration conditions are explicit |
| Active numbered era | None |
| Adopted Era 9 | No |
The adopted Keeper exact-task slice remains outstanding. Media's governed
objective milestone is complete at commits 62fb131, 019d900, and d914f2a;
remaining device and delivery limitations are bounded follow-ups.
The current program sequence and recovered work register separate NOW, NEXT, product, migration, research and completed work. The highest leverage next objective is Keeper exact-task completion lifecycle consolidation, which retains its existing adoption. Broad capability-backed read-only cognition is implemented with bounded source coverage through the shared intent boundary; see its architecture and limitations. Recommendations remain advisory and do not authorize execution.
What AncientOS is
AncientOS is a transport-neutral governed operational cognition platform for controlled, inspectable, single-operator use. Luna is an interactive application and interface over AncientOS; it is not the kernel itself.
AncientOS Platform v1 Alpha supplies reusable runtime and governance primitives:
- the shared Runtime Kernel and Runtime Composition;
- durable Kernel Records and restart-safe explainability;
- deterministic Governed Proposals;
- durable, exact-action Lich approval;
- separate bounded execution;
- Zeus evidence and execution or rollback receipts;
- rollback or explicit compensation boundaries;
- transport-neutral routing across Discord, Terminal TUI, and Web TUI;
- Rubick capability and dependency posture;
- Oracle inspection, synthesis, recommendation, queue, validation, and audit;
- LifeVault memory authority and provenance;
- Meepo transition revalidation;
- read-only Spectre terminal observation.
The platform preserves governed execution boundaries. Typed mutation lanes and Clockwerk/Io observation exist; they grant no general shell, Docker, Git-update, package-manager, arbitrary network-write, broad-delete, or autonomous authority.
Portfolio lifecycle
Portfolio lifecycle is a decision record, not a readiness score:
| State | Meaning |
|---|---|
candidate |
A bounded option identified for analysis. It is not authorized work. |
proposed |
A documented adoption packet awaits an explicit operator decision. |
adopted |
The operator approved the capability for implementation, but implementation has not started. |
in_progress |
Work has begun under the adopted scope and acceptance contract. |
blocked |
Adopted work cannot progress until named evidence, dependency, policy, or product decisions are resolved. |
complete |
The adopted acceptance criteria and required validation are satisfied. |
deferred |
The operator intentionally removed the item from active consideration until stated conditions change. |
rejected |
The operator explicitly declined the proposal; reconsideration requires a new proposal. |
historical |
Retained provenance from a retired roadmap vocabulary or completed planning program. |
Only an explicit operator decision can move a capability into adopted.
Oracle may report facts, risks, gaps, and recommendations. Rubick may report
capability posture and dependency state. Neither may silently adopt, reorder,
or activate portfolio work.
Roadmap adoption and runtime mutation approval are separate decisions. Adopting a capability authorizes bounded implementation work; it does not approve any live mutation. Every meaningful mutation remains separately Lich-gated at the exact-action boundary.
Independent capability posture
Portfolio state never substitutes for technical posture. Every capability is described independently along these dimensions:
- implementation: whether bounded code exists;
- integration: whether required services and consumers are wired;
- operator reachability: whether a documented operator route reaches the contract;
- authority: advisory, read-only, proposal, approval, execution, evidence, memory, transport, or application boundaries;
- persistence: durable, process-bounded, ephemeral, external, or not applicable;
- evidence strength: source-only, unit, integration, runtime smoke, or live external validation;
- operational health: current configured, injected, observed, degraded, disabled, unavailable, or unknown state.
Implementation does not prove wiring. Injection does not prove route reachability. Fixture validation does not prove a credentialed integration. A provider call does not prove reconciled success. A roadmap decision grants no runtime authority.
Completed foundations
Governed Meeting Intelligence vertical slice
The operator-authorized Meeting Intelligence capability is implemented as a capability-local vertical slice rather than a new platform era. It reuses the shared Runtime Kernel, LifeVault artifact authority, Rubick posture, governed inference, Keeper/Lich proposal path, and transport-neutral rendering. The bounded acceptance surface is transcript import and query; optional Meetily or other workstation capture remains an edge-producer integration, not an AncientOS authority or bundled capture implementation.
These are verified foundations, not active work:
| Foundation | Portfolio state | Verified outcome | Important boundary |
|---|---|---|---|
| AncientOS Platform v1 Alpha | complete |
Transport-neutral governed runtime, durable records, approval, evidence, receipts, explainability, and transport parity | Alpha for controlled use; not production |
| Oracle Operational Ergonomics | complete |
Operator inspection, queue, preflight, proposal intake, explicit apply vocabulary, and audit | Oracle delegates to existing services; no approval or execution authority |
| Governed configuration-value edit | complete |
Registered non-secret scalar inspection, proposal, execution, evidence, exact rollback, and restart continuity | Default disabled; named target and allowed-value configuration required |
| Governed atomic file rename | complete |
Registered no-overwrite regular-file rename with evidence and exact reverse rollback | Default disabled; allowlisted disposable workspace only |
| Governed Home Assistant light state | complete |
Registered light.* on/off proposal, reconciled receipt, and separately approved compensation |
Era 8 used a fake provider; no credentialed live device validation |
| Governed capability discovery | complete |
Oracle lists and inspects the implemented portfolio from Rubick evidence | Discovery does not grant capability authority |
| Governed inference registry and model selection | complete for the current production boundary |
Canonical local/frontier targets, freshness-bound evidence, truthful inventory, request/session selection, restore/reset, provider disambiguation, and transport-neutral natural language | xAI and llama.cpp live-validated; outbound OpenAI unconfigured; persistent defaults and autonomous routing absent; cognition never changes authority |
| Media governed objective lifecycle | complete for the demonstrated vertical slice |
Natural objective through approval, acquisition, completion, Plex verification, durable return, Pixel delivery and named-room playback | Cold/profile-bound device actuation, exact-episode identity and fresh post-repair replay/Push acceptance remain bounded follow-ups |
| Media episode inventory | complete for the current read-only slice |
Canonical TVmaze series identity, scoped aired-episode comparison, bounded Plex missing/extra diagnostics, and transport-neutral routing | No acquisition discovery, downloading, filesystem mutation, Plex refresh, scheduling, or approval authority |
| Media acquisition discovery | complete for the current read-only slice |
Canonical missing-episode target, hostile candidate normalization, hard eligibility, deterministic deduplication/ranking, and bounded recommendation evidence | Original read-only acceptance boundary; later governed multi-provider acquisition and completion are delivered separately |
| Spectre terminal observation | complete for terminal v1 |
Explicit spectre-shell production, redaction, ingestion, binding, and Oracle explanation |
No SSH interception, hidden capture, screen capture, or execution authority |
| Governed experiential-learning foundation | complete for the current advisory contract |
Evidence-backed LifeVault assertions, deterministic scoped snapshots, post-boundary LegionCommander advice, and replay provenance | LegionCommander is the only consumer; no confidence, ranking, automatic promotion, cross-domain transfer, routing, readiness, approval, or authority effect |
Era 7 and Era 8 are historical program labels attached to the relevant completion evidence. They do not create progression gates for future capabilities.
The authoritative completion boundary and classification of remaining inference ideas are recorded in the Governed Capability Portfolio roadmap. Those future and deferred items are not active work and have no implied order.
Currently adopted portfolio
The Keeper exact-task completion adoption contract
is adopted and awaits implementation. It covers one exact task completion
through the canonical durable lifecycle. Adoption authorizes implementation,
not any task mutation; each runtime action remains separately Lich-gated.
The Chen watch-loop plan is historical lineage for the completed Media milestone. See the canonical roadmap for current follow-ups; its original excluded playback/provider scope was extended by later explicit operator work.
An implementation must not begin merely because a candidate appears in this document or because Oracle or Rubick ranks it favorably.
Candidate capabilities
Candidates are non-authoritative options. Their presence records analysis, not commitment, order, readiness, or permission to implement.
| Candidate | Operator outcome | Existing evidence | Adoption gaps |
|---|---|---|---|
| qBittorrent pause/resume | Pause or resume one exact observed torrent | Read-only adapter and Chen provider experience | Exact hash policy, durable portfolio proposal/evidence/receipt lifecycle, reconciliation, provider configuration |
| Read-only service diagnostics | Explain service degradation from bounded evidence | Beastmaster and Oracle inspection surfaces exist | Exact operator outcome, evidence freshness, provider boundaries, no-remediation guarantee |
| Optional inference follow-ons | Benchmarking, provider/role reconciliation, additional providers, multimodal/long-context policy, session cleanup/history, Ollama repair, egress/accounting, or per-model metrics | The governed registry/selection initiative is complete; its authoritative follow-on table records each item separately | A future session must adopt one bounded outcome; the grouped row is not a program or priority queue |
No candidate has an implied priority. Adoption analysis must evaluate operator value, scope, dependency evidence, reversibility or compensation, blast radius, testability, provider configuration, and authority boundaries at the time of decision.
Deferred capabilities
| Deferred capability | Why it is deferred | Reconsider only when |
|---|---|---|
| General allowlisted container restart | Large Docker-socket authority and weak reversal semantics; the adopted typed recover_gluetun_vpn exception grants no generic target selection |
A separate typed target contract, health convergence, uncertainty receipts, and safe fixture acceptance exist for each additional invariant |
| Compose deploy or recreate | Combines build, images, networking, volumes, and service health | A separately governed deployment program is explicitly adopted |
| Fast-forward-only Git update | Remote trust, signatures, dirty-tree, divergence, and rollback policy are unresolved | Repository trust and recovery contracts are adopted and tested |
| Package or system update | Root authority, network trust, reboot impact, and downgrade uncertainty | Isolated host validation and reliable rollback evidence exist |
| Broad delete or data removal | Irreversibility and retention risk | A quarantine-first capability proves bounded identity, retention, capacity, and restoration |
| General scheduler | Creates durable unattended execution and cancellation concerns | One adopted capability demonstrates a narrow, cancelable scheduling need |
| Autonomous remediation | Conflicts with current explicit human governance | A separate operator decision changes doctrine; no such decision exists |
| General shell or command executor | Authority is too broad for typed governed outcomes | Not anticipated; prefer bounded typed capabilities |
| Governed experiential confidence/reinforcement | Evidence-backed ranking semantics, reinforcement identity, governance, and replay are unresolved | Real certified experience has accumulated; missing exact evidence adapters are justified; and a dedicated architecture review defines independent evidence, scope, correlation, contradiction, freshness, deterministic ranking, mutation governance, explainability, and the authority firewall |
| Deferred inference automation and mutation | Persistent defaults, automatic routing/fallback, LegionCommander choice, background monitoring, model lifecycle, and updates require governance not present in the completed registry contract | The exact prerequisite for each item in the authoritative inference completion table is met and the operator adopts that bounded capability |
Deferred items are not a queue. Their order carries no meaning.
The experiential foundation is complete, but confidence and reinforcement are not implemented or adopted. The canonical roadmap authority defines the required planning gate and a non-authoritative reconsideration sequence. No Rubick capability or additional advisory consumer is implied.
Known portfolio gaps
The following gaps are observations, not adopted work:
- capability posture metadata is richer for the Era 8 portfolio than for some legacy Rubick records;
- external-provider freshness and reachability are evaluated at request time rather than universally persisted;
- Home Assistant lacks credentialed live-provider acceptance evidence;
- canonical principals and enrolled device sessions are implemented; a general multi-user or cross-instance cryptographic identity model is not established;
- LifeVault ordinary conversational retrieval is not composed into the default runtime pipeline;
- Spectre conversation bindings are process-bounded and live screen capture is unavailable;
- Zeus durable evidence is complete for Governed Actions but not universal across every subsystem;
- legacy application-specific mutation lanes do not all use the canonical portfolio lifecycle.
Closing a gap requires explicit scope and adoption. A gap report is not an implementation instruction.
Capability adoption process
- Identify one bounded operator outcome as a
candidate. - Document its identity, scope, dependencies, provider configuration, authority requirements, persistence, evidence requirements, rollback or compensation classification, risks, and acceptance criteria.
- Ask Oracle and Rubick for advisory evidence, dependency, conflict, and posture analysis when useful.
- Produce a
proposedadoption packet that names the exact capability and preserves unresolved gaps. - The operator explicitly adopts, defers, or rejects the proposal.
- Adoption places only that capability into the active portfolio. It does not alter Lich, Zeus, Oracle, Rubick, Meepo, Spectre, LifeVault, transport, or provider authority.
- Move adopted work to
in_progressonly when implementation begins. Record named blockers rather than inventing substitute work. - Record source, unit, integration, runtime smoke, and live external evidence honestly. Fixture, local, credentialed, and live-provider validation remain distinguishable.
- Declare
completeonly when the adopted acceptance criteria and required repository-wide validation pass. - Preserve the proposal, adoption decision, evidence, completion record, and any superseded historical material.
An adoption decision should state:
- capability ID and operator outcome;
- decision actor and date;
- exact scope and non-goals;
- dependencies and provider requirements;
- authority and approval boundaries;
- persistence and restart behavior;
- evidence and acceptance requirements;
- rollback, compensation, or non-reversibility;
- security and privacy risks;
- compatibility impact;
- conditions for completion, deferral, rejection, or reconsideration.
Authority boundaries
Oracle
Oracle reports facts, observations, inferences, risks, gaps, and recommendations. It may render the portfolio and an adoption packet. It may not adopt work, mutate roadmap state, approve a capability, execute a capability, or create a separate governed lifecycle.
Rubick
Rubick is the capability and dependency posture authority. It reports evidenced implementation, wiring, reachability metadata, authority, persistence, evidence, configuration, and limitations. Registration is not readiness and Rubick does not grant execution or approval authority.
Lich
Lich approves or rejects exact governed mutations. Portfolio adoption does not pre-approve implementation outputs or live actions.
Zeus
Zeus owns governed execution evidence and receipts and provides bounded supervision evidence. Zeus does not adopt roadmap work, approve actions, or execute providers.
Meepo
Meepo revalidates transition assumptions across approval-scoped state changes. It does not choose work, approve, execute, or create authority.
LifeVault
LifeVault is durable memory authority. Portfolio documents and Oracle reports do not become long-term memory merely by existing; memory promotion retains its governed provenance and review requirements.
Spectre
Spectre observes and normalizes attributable terminal evidence. It cannot control a terminal, intercept SSH, capture hidden screens, approve, execute, or certify success.
Validation language
Use these distinctions:
- fixture validation: deterministic fake or fixture provider;
- local validation: real local process or disposable workspace without a credentialed external service;
- credentialed integration validation: authenticated external API exercised in a bounded environment;
- live-provider validation: the intended external provider and target were exercised and the result reconciled safely.
The strongest evidence recorded for a claim must match what was actually tested. Era 8 Home Assistant acceptance is fixture validation and a shared-runtime smoke test, not credentialed or live-provider validation.
Historical numbered-era record
AncientOS previously organized work using numbered eras and sub-era phase labels. That sequence is closed.
| Historical program | Historical outcome |
|---|---|
| Eras 0–4 | Early runtime, subsystem, artifact, orchestration, and integration foundations |
| Era 5 | Governed live-execution concepts; implementation landed through several later bounded lifecycles rather than one completed scalar program |
| Era 6 and sub-era labels | Memory, interoperability, cognition, and validation foundations; individual surfaces have independent current posture |
| Era 7 | Oracle Operational Ergonomics, complete |
| Era 8 | Governed Capability Portfolio, complete |
| Era 9 and later numbered headings | Never adopted candidates or historical planning vocabulary |
Root ERA*_PLAN.md files and superseded proposal documents remain provenance.
Their numbering, stages, phases, priorities, and “next” language do not govern
current work.
Canonical sources
- Documentation Reality and Status classifies authorities, supporting pages, historical records, and generated output.
- Architecture and Capability Posture Taxonomy defines portfolio lifecycle and independent posture dimensions.
- Governed Capability Portfolio roadmap governs adoption and the capability standard.
- Durable Governed Actions defines the implemented mutation lifecycle.
- Runtime Composition reports current process wiring without unsafe probes.
- Rubick reports capability posture; verified source and tests remain implementation evidence when authored documentation drifts.
No generated page, component knowledge file, historical era plan, model recommendation, or runtime summary is an independent roadmap authority.
Rote design knowledge
The preserved Rote corpus is now part of the
repository's authoritative design documentation. P1 in the canonical portfolio
records reconciliation, decisions and dependencies.
Universal semantics are Core conceptual concerns; Apps may specialize domain
Rotes without owning universal authority. The adopted reconstructed addendum's
provenance remains explicit. The first bounded informational exchange
is implemented in 79dc3628b5e1627c7e9aec29bc6f903c9c1b6ca7. Relationship-scoped
exchange with no public advertisement is the near-term posture; wider discovery
and effectful Rotes remain deferred. Project-corpus reads and narrow source
establishment have deployed evidence; broad planning remains separately incomplete.
The next recommended product slice is governed read-only inspection of one
received informational Rote, including its provenance and restrictions.
Keeper exact-task completion remains the adopted development slice.