ARADR: rework StarIntel inference into a StarLang expert-engine actor #182

New issue

Open

opened 2026-09-02 17:26:15 +00:00 by nsaspy · 0 comments

nsaspy commented

2026-09-02 17:26:15 +00:00

Owner

Operator direction

Rework the stale StarIntel inference-actor direction into the current StarLang expert-system architecture.

This issue is authorized to run research refresh -> design without another operator gate between those phases. It does not authorize implementation. Final implementation slices must be presented for operator approval.

Why this is needed

STAR-RESEARCH-003 Budgeted Supervised Prolog Inference Actors is stale. The operator has already explicitly required it to be reworked into a StarLang expert-engine actor, where Prolog is a selectable logic backend rather than the actor architecture.

Do not extend the old Prolog-only actor design and do not revive the old internal-Prolog-server direction.

The generic StarLang research direction is already approved, including:

STAR-LANG-RESEARCH-021-star-logic-runtime-integration
STAR-LANG-RESEARCH-022-three-backend-logic-runtime-selection
STAR-LANG-RESEARCH-023-expert-system-language-surface
STAR-LANG-RESEARCH-026-explicit-logic-backend-selection
STAR-LANG-RESEARCH-028-logic-runtime-api-contract
the approved StarLang expert/backend designs STAR-LANG-007, 009, 010, 011, 012

Canonical approval recovery/evidence: #158 and the architecture refresh decision in #156.

Current implementation reality to inspect first

Inspect current heads of:

lost-rob0t/star-lang
lost-rob0t/starintel-server
relevant current Auto-Research StarLang / StarIntel research and design nodes

Do not design against the August prototype state.

Current StarLang already has final star-logic-protocol and star-logic-ir, but current final-tree inference execution is incomplete. In particular, verify the live status of the real backend adapters before declaring anything implementable. lost-rob0t/star-lang#45 is the current SWI-Prolog MQI worker-adapter foundation issue and intentionally does not yet provide portable query semantics.

Also inspect current final actor runtime semantics, including the known nested synchronous ask limitation, and determine whether the expert engine needs split-phase/deferred actor interaction rather than recursive synchronous ask.

Research refresh requirements

Refresh/supersede STAR-RESEARCH-003 and STAR-RESEARCH-004 against the approved StarLang architecture.

Research and pin the concrete StarIntel expert-system requirements:

Expert-engine actor boundary
- actor/domain name and ownership;
- local and remote invocation contract;
- lifecycle, restart, readiness, degradation and backend-unavailable states;
- mailbox, cancellation, deadlines and idempotency semantics;
- no backend-specific actor API.
Expert definition / package model
- StarLang expert/rule declaration surface;
- immutable package/version/digest identity;
- semantic profile and required backend capabilities;
- typed input/output/fact schemas;
- no arbitrary raw Prolog/Lisp/native engine escapes.
Fact lifecycle
- asserted vs observed vs inferred facts;
- run/session-scoped fact delta;
- durable vs ephemeral facts;
- retraction, stale inference and dependency invalidation;
- ontology/predicate-version compatibility;
- provenance/evidence links.
Inference lifecycle
- open/close logic session;
- fact-delta application;
- solve/query/exists/solutions-style portable operations where appropriate;
- bounded result iteration;
- cancellation and resource budgets;
- deterministic backend selection evidence;
- explicit backend failure/unavailability with no silent cross-backend semantic fallback.
Explanation / proof contract
- proof/derivation refs;
- rule IDs and package versions;
- input fact dependencies;
- confidence or scoring semantics only where explicitly defined by the expert package;
- explanation suitable for StarIntel graph/UI projection without making the graph stronger than the evidence.
Effects boundary
- expert rule evaluation is effect-closed by default;
- rules may conclude/request evidence-needed, collection strategy, escalation, etc.;
- an owning actor/researcher converts those conclusions into authorized capability calls;
- no HTTP/browser/process/tool call directly from a rule RHS unless StarLang later defines an explicit separately reviewed effect model.
StarIntel integration
- canonical ingest -> typed facts;
- inferred assertion -> canonical ingest with proof/provenance;
- entity/identity correlation ownership;
- graph projection;
- Quasar inspection/explanation surface requirements;
- compatibility with v0.9 document/relation/provenance semantics.
IPX integration proof
- use the new StarLang IPX passive HTTP actor pipeline as one concrete consumer;
- define how an HTTP exchange/evidence projection becomes expert facts;
- define a bounded expert invocation and derived finding/proof result;
- do not make IPX itself own a backend or duplicate the expert runtime.
Source Acquisition / orchestration integration
- reconcile with #171 and #172;
- expert correlation remains in the expert engine;
- playbooks/source acquisition remain effectful collection/execution owners;
- partial/blocked/unavailable acquisition is evidence state, never silently notFound.
Backend matrix
- SWI-Prolog;
- LISA;
- N-Prolog;
- exact current implementation status and missing conformance slices for each;
- required semantic profiles/capabilities before auto selection may choose a backend.

Required adversarial pass

Explicitly test/reject at least:

Prolog-only public actor API;
arbitrary raw goal execution;
arbitrary Common Lisp callbacks inside expert packages;
rules directly performing network/process effects;
backend-specific fact encodings leaking into StarIntel callers;
silent backend failover with changed semantics;
unbounded recursion/result enumeration;
inferred facts without dependency/proof identity;
deleting/retracting source facts while leaving derived facts current;
graph projection treated as authoritative evidence;
visibility/capability discovery treated as authorization;
nested synchronous actor asks that can deadlock/time out under current deterministic runtime semantics.

Required design output

Create a canonical StarIntel design node replacing the old Prolog-specific actor architecture, tentatively:

STAR-EXPERT-001 StarLang Expert Engine Domain/Actor

Use the final naming conventions discovered during the pass.

The design must include:

exact actor/domain contract;
exact ownership boundary between star-lang and starintel-server;
portable expert package/fact/query/result/proof contracts;
backend adapter boundary;
state/lifecycle/restart semantics;
local/remote semantics;
cancellation/deadline/resource budgets;
provenance and canonical-ingest path;
IPX vertical-slice example;
Quasar/operator observability requirements;
RED-first fixtures;
dependency graph and implementation slices.

Required implementation issue output

After the design is internally consistent, create dependency-ordered implementation issues in the correct repositories. At minimum separate:

missing final StarLang portable expert/query semantics;
concrete backend adapter/conformance work (do not merge all engines into one issue);
StarLang expert actor/domain runtime surface if missing;
starintel-server expert-engine integration;
proof/provenance + canonical inferred-assertion ingest;
IPX expert integration fixture;
Quasar explanation/inspection integration if a UI gap is confirmed.

Do not mark implementation issues executable/approved. They must be AWAITING_OPERATOR_IMPLEMENTATION_APPROVAL until the operator explicitly approves them.

RED-first acceptance targets

At design completion, specify failing tests proving untouched current code cannot yet complete at least:

one backend-neutral expert invocation from typed facts to typed result;
backend selection pinned with immutable selection evidence;
fact retraction causing dependent inference to become stale/retracted;
proof carrying exact package/rule/input identities;
unavailable backend producing typed failure rather than fake success/fallback;
IPX exchange projection -> expert facts -> derived finding with proof;
local and remote expert invocation preserving the same typed contract;
cancellation/deadline fencing late results.

Website/index reconciliation

Update the Auto-Research indexes/export metadata so the public site stops presenting the old Budgeted Supervised Prolog Inference Actors direction as the current expert architecture. Preserve it as superseded history with a link to the replacement.

Completion gate

Done when:

stale Prolog-only research is explicitly superseded;
current StarLang implementation/backend gaps are verified against source;
the canonical StarLang expert-engine actor design exists and is DESIGN_READY_FOR_OPERATOR_REVIEW;
IPX has a concrete expert integration vertical slice in the design;
dependency-ordered implementation issues exist but remain human-gated;
the website/index points at the new design instead of the stale architecture.

Implementation authorization: PENDING / OPERATOR ONLY.

## Operator direction Rework the stale StarIntel inference-actor direction into the current StarLang expert-system architecture. This issue is authorized to run **research refresh -> design** without another operator gate between those phases. It does **not** authorize implementation. Final implementation slices must be presented for operator approval. ## Why this is needed `STAR-RESEARCH-003 Budgeted Supervised Prolog Inference Actors` is stale. The operator has already explicitly required it to be reworked into a **StarLang expert-engine actor**, where Prolog is a selectable logic backend rather than the actor architecture. Do not extend the old Prolog-only actor design and do not revive the old internal-Prolog-server direction. The generic StarLang research direction is already approved, including: - `STAR-LANG-RESEARCH-021-star-logic-runtime-integration` - `STAR-LANG-RESEARCH-022-three-backend-logic-runtime-selection` - `STAR-LANG-RESEARCH-023-expert-system-language-surface` - `STAR-LANG-RESEARCH-026-explicit-logic-backend-selection` - `STAR-LANG-RESEARCH-028-logic-runtime-api-contract` - the approved StarLang expert/backend designs `STAR-LANG-007`, `009`, `010`, `011`, `012` Canonical approval recovery/evidence: #158 and the architecture refresh decision in #156. ## Current implementation reality to inspect first Inspect current heads of: - `lost-rob0t/star-lang` - `lost-rob0t/starintel-server` - relevant current Auto-Research StarLang / StarIntel research and design nodes Do not design against the August prototype state. Current StarLang already has final `star-logic-protocol` and `star-logic-ir`, but current final-tree inference execution is incomplete. In particular, verify the live status of the real backend adapters before declaring anything implementable. `lost-rob0t/star-lang#45` is the current SWI-Prolog MQI worker-adapter foundation issue and intentionally does not yet provide portable query semantics. Also inspect current final actor runtime semantics, including the known nested synchronous `ask` limitation, and determine whether the expert engine needs split-phase/deferred actor interaction rather than recursive synchronous ask. ## Research refresh requirements Refresh/supersede `STAR-RESEARCH-003` and `STAR-RESEARCH-004` against the approved StarLang architecture. Research and pin the concrete StarIntel expert-system requirements: 1. **Expert-engine actor boundary** - actor/domain name and ownership; - local and remote invocation contract; - lifecycle, restart, readiness, degradation and backend-unavailable states; - mailbox, cancellation, deadlines and idempotency semantics; - no backend-specific actor API. 2. **Expert definition / package model** - StarLang expert/rule declaration surface; - immutable package/version/digest identity; - semantic profile and required backend capabilities; - typed input/output/fact schemas; - no arbitrary raw Prolog/Lisp/native engine escapes. 3. **Fact lifecycle** - asserted vs observed vs inferred facts; - run/session-scoped fact delta; - durable vs ephemeral facts; - retraction, stale inference and dependency invalidation; - ontology/predicate-version compatibility; - provenance/evidence links. 4. **Inference lifecycle** - open/close logic session; - fact-delta application; - solve/query/exists/solutions-style portable operations where appropriate; - bounded result iteration; - cancellation and resource budgets; - deterministic backend selection evidence; - explicit backend failure/unavailability with no silent cross-backend semantic fallback. 5. **Explanation / proof contract** - proof/derivation refs; - rule IDs and package versions; - input fact dependencies; - confidence or scoring semantics only where explicitly defined by the expert package; - explanation suitable for StarIntel graph/UI projection without making the graph stronger than the evidence. 6. **Effects boundary** - expert rule evaluation is effect-closed by default; - rules may conclude/request `evidence-needed`, collection strategy, escalation, etc.; - an owning actor/researcher converts those conclusions into authorized capability calls; - no HTTP/browser/process/tool call directly from a rule RHS unless StarLang later defines an explicit separately reviewed effect model. 7. **StarIntel integration** - canonical ingest -> typed facts; - inferred assertion -> canonical ingest with proof/provenance; - entity/identity correlation ownership; - graph projection; - Quasar inspection/explanation surface requirements; - compatibility with v0.9 document/relation/provenance semantics. 8. **IPX integration proof** - use the new StarLang IPX passive HTTP actor pipeline as one concrete consumer; - define how an HTTP exchange/evidence projection becomes expert facts; - define a bounded expert invocation and derived finding/proof result; - do not make IPX itself own a backend or duplicate the expert runtime. 9. **Source Acquisition / orchestration integration** - reconcile with #171 and #172; - expert correlation remains in the expert engine; - playbooks/source acquisition remain effectful collection/execution owners; - partial/blocked/unavailable acquisition is evidence state, never silently `notFound`. 10. **Backend matrix** - SWI-Prolog; - LISA; - N-Prolog; - exact current implementation status and missing conformance slices for each; - required semantic profiles/capabilities before `auto` selection may choose a backend. ## Required adversarial pass Explicitly test/reject at least: - Prolog-only public actor API; - arbitrary raw goal execution; - arbitrary Common Lisp callbacks inside expert packages; - rules directly performing network/process effects; - backend-specific fact encodings leaking into StarIntel callers; - silent backend failover with changed semantics; - unbounded recursion/result enumeration; - inferred facts without dependency/proof identity; - deleting/retracting source facts while leaving derived facts current; - graph projection treated as authoritative evidence; - visibility/capability discovery treated as authorization; - nested synchronous actor asks that can deadlock/time out under current deterministic runtime semantics. ## Required design output Create a canonical StarIntel design node replacing the old Prolog-specific actor architecture, tentatively: `STAR-EXPERT-001 StarLang Expert Engine Domain/Actor` Use the final naming conventions discovered during the pass. The design must include: - exact actor/domain contract; - exact ownership boundary between `star-lang` and `starintel-server`; - portable expert package/fact/query/result/proof contracts; - backend adapter boundary; - state/lifecycle/restart semantics; - local/remote semantics; - cancellation/deadline/resource budgets; - provenance and canonical-ingest path; - IPX vertical-slice example; - Quasar/operator observability requirements; - RED-first fixtures; - dependency graph and implementation slices. ## Required implementation issue output After the design is internally consistent, create dependency-ordered implementation issues in the correct repositories. At minimum separate: 1. missing final StarLang portable expert/query semantics; 2. concrete backend adapter/conformance work (do not merge all engines into one issue); 3. StarLang expert actor/domain runtime surface if missing; 4. `starintel-server` expert-engine integration; 5. proof/provenance + canonical inferred-assertion ingest; 6. IPX expert integration fixture; 7. Quasar explanation/inspection integration if a UI gap is confirmed. Do not mark implementation issues executable/approved. They must be `AWAITING_OPERATOR_IMPLEMENTATION_APPROVAL` until the operator explicitly approves them. ## RED-first acceptance targets At design completion, specify failing tests proving untouched current code cannot yet complete at least: - one backend-neutral expert invocation from typed facts to typed result; - backend selection pinned with immutable selection evidence; - fact retraction causing dependent inference to become stale/retracted; - proof carrying exact package/rule/input identities; - unavailable backend producing typed failure rather than fake success/fallback; - IPX exchange projection -> expert facts -> derived finding with proof; - local and remote expert invocation preserving the same typed contract; - cancellation/deadline fencing late results. ## Website/index reconciliation Update the Auto-Research indexes/export metadata so the public site stops presenting the old `Budgeted Supervised Prolog Inference Actors` direction as the current expert architecture. Preserve it as superseded history with a link to the replacement. ## Completion gate Done when: - stale Prolog-only research is explicitly superseded; - current StarLang implementation/backend gaps are verified against source; - the canonical StarLang expert-engine actor design exists and is `DESIGN_READY_FOR_OPERATOR_REVIEW`; - IPX has a concrete expert integration vertical slice in the design; - dependency-ordered implementation issues exist but remain human-gated; - the website/index points at the new design instead of the stale architecture. Implementation authorization: **PENDING / OPERATOR ONLY**.

nsaspy referenced this issue

2026-09-02 17:26:16 +00:00

DESIGN_READY_FOR_OPERATOR_REVIEW: managed runtime components, shutdown deadline, and reconnect fencing #200