Native Problog

.gitignore

@@ -1,3 +1,4 @@
 node_modules/
 dist/
 coverage/
+.tmp/

.ok/problog-bun-test-parity.ok.md

@@ -0,0 +1,37 @@
+# ProbLog Bun Test Parity Ideal Target State
+
+This file captures the evergreen ideal state for Bun-based tests that demonstrate ProbLog behavior in `just-prolog` with parity against `.tmp/problog`.
+
+## Bun Test Execution Path
+
+- `PBT-PATH-001`: The parity suite is invokable from repository root with `bun test` and no required `cd` into a subdirectory.
+- `PBT-PATH-002`: The parity suite execution path is deterministic: the same checked-out commit and fixture corpus yield the same pass/fail result under `bun test`.
+- `PBT-PATH-003`: Drift detection is fail-fast: any unmapped or extra fixture in the declared scope causes `bun test` to fail in the parity suite run.
+
+## Referenced Source Set
+
+- `PBT-SOURCE-001`: The authoritative fixture corpus root is `.tmp/problog/test`.
+- `PBT-SOURCE-002`: The current top-level fixture corpus contains exactly `107` `.pl` files directly under `.tmp/problog/test`.
+- `PBT-SOURCE-003`: The current immediate subdirectories under `.tmp/problog/test` are exactly `bn`, `constraints`, `dtproblog`, `lfi`, `lficont`, `parser`, `sample`, `specific`, and `tasks`.
+- `PBT-SOURCE-004`: The parity scope for this target state is top-level `.pl` fixtures only; fixtures located inside immediate subdirectories are out of scope.
+- `PBT-SOURCE-005`: The referenced source set is treated as read-only input for parity validation.
+
+## Fixture Mapping Parity
+
+- `PBT-MAP-001`: Mapping cardinality is bijective across the declared scope: `107` in-scope source fixtures map to `107` parity test cases.
+- `PBT-MAP-002`: Each in-scope source fixture maps to exactly one Bun parity test case.
+- `PBT-MAP-003`: Each Bun parity test case maps to exactly one in-scope source fixture.
+- `PBT-MAP-004`: A fixture registry drift check fails the suite when any in-scope fixture is unmapped or when any mapped fixture path does not exist.
+- `PBT-MAP-005`: Concrete fixtures `01_inconsistent.pl`, `ad_fact.pl`, `4_bayesian_net.pl`, and `non_ground_query.pl` are in-scope and each has exactly one mapped Bun parity test case.
+
+## Behavior Demonstration Parity
+
+- `PBT-BEH-001`: Each mapped Bun test demonstrates the same behavioral claim as its corresponding source fixture.
+- `PBT-BEH-002`: Numeric probabilistic expectations in mapped Bun tests match the corresponding source fixture expectations.
+- `PBT-BEH-003`: Error-mode expectations in mapped Bun tests match the corresponding source fixture expectations.
+
+## Coverage Criterion
+
+- `PBT-COV-001`: Fixture-level parity coverage across the declared scope is exactly `100%` (`107/107` mapped).
+- `PBT-COV-002`: Any coverage value below `100%` for the declared scope fails the parity suite.
+- `PBT-COV-003`: The parity suite reports failure when corpus drift changes the in-scope fixture set and mapping is not updated to restore `107/107` coverage.

.ok/problog-examples.ok.md

@@ -0,0 +1,39 @@
+# ProbLog Examples Ideal Target State
+
+This file captures the evergreen ideal state for a reusable set of ProbLog examples in `just-prolog`.
+
+## Scope and Curation
+
+- `P2EX-SCOPE-001`: A dedicated examples set exists for ProbLog programs.
+- `P2EX-SCOPE-002`: Each example has a stable identifier that is unique within the examples set.
+- `P2EX-SCOPE-003`: Each example states its intent in one concise sentence.
+- `P2EX-SCOPE-004`: Example coverage spans deterministic baseline, probabilistic facts, annotated disjunctions, and evidence-conditioned queries.
+
+## Example Structure
+
+- `P2EX-STRUCT-001`: Each example includes a ProbLog source program.
+- `P2EX-STRUCT-002`: Each example declares at least one `query/1` directive.
+- `P2EX-STRUCT-003`: Examples that model observations include `evidence/1` or `evidence/2` directives.
+- `P2EX-STRUCT-004`: Each example defines expected numeric probabilities for its query atoms.
+- `P2EX-STRUCT-005`: Each example defines a numeric tolerance used for probability comparison.
+
+## Behavioral Guarantees
+
+- `P2EX-BEH-001`: Running an example produces query marginals in the same order as the declared expected outcomes.
+- `P2EX-BEH-002`: Every expected probability assertion evaluates to pass or fail without ambiguous states.
+- `P2EX-BEH-003`: Invalid example definitions produce explicit validation errors.
+- `P2EX-BEH-004`: Inconsistent evidence examples produce explicit inconsistent-evidence errors.
+
+## Portfolio Coverage
+
+- `P2EX-PORT-001`: The examples set includes a two-coin model with a derived query.
+- `P2EX-PORT-002`: The examples set includes at least one annotated-disjunction model.
+- `P2EX-PORT-003`: The examples set includes at least one Bayesian-network style conditional model.
+- `P2EX-PORT-004`: The examples set includes at least one model with contradictory evidence.
+- `P2EX-PORT-005`: The examples set includes at least one model that demonstrates non-ground query usage.
+
+## Evolution Rules
+
+- `P2EX-EVO-001`: New examples are additive and do not silently change existing example identifiers.
+- `P2EX-EVO-002`: Changes to expected probabilities require an explicit rationale captured next to the example metadata.
+- `P2EX-EVO-003`: The examples set remains runnable in automated test environments without network access.

.ok/problog2.ok.md

@@ -0,0 +1,50 @@
+# ProbLog2 Ideal Target State
+
+This file captures the evergreen ideal state for ProbLog2 support in `just-prolog`.
+
+## Boundary and API
+
+- `P2-API-001`: `just-prolog` exposes a dedicated probabilistic runtime API that is distinct from the deterministic `Prolog` runtime.
+- `P2-API-002`: The existing deterministic `Prolog` API and behavior remain unchanged when ProbLog features are not used.
+- `P2-API-003`: Probabilistic query results include numeric probabilities per requested query atom.
+- `P2-API-004`: Probabilistic query results expose explicit error states for inconsistent evidence and invalid probabilistic models.
+
+## Language Surface
+
+- `P2-LANG-001`: The ProbLog parser accepts probabilistic annotations with `::` for facts and clauses.
+- `P2-LANG-002`: The ProbLog parser accepts annotated disjunction heads with `;` for both fact and rule forms.
+- `P2-LANG-003`: The ProbLog parser accepts `query/1` directives.
+- `P2-LANG-004`: The ProbLog parser accepts `evidence/1` and `evidence/2` directives.
+- `P2-LANG-005`: The ProbLog parser accepts negation as failure with `\+` in ProbLog mode.
+- `P2-LANG-006`: The ProbLog parser accepts both `:-` and `<-` as clause separators in ProbLog mode.
+
+## Probabilistic Semantics
+
+- `P2-SEM-001`: Ground probabilistic facts represent independent binary random choices unless linked by an annotated disjunction group.
+- `P2-SEM-002`: Annotated disjunction alternatives are mutually exclusive within a group.
+- `P2-SEM-003`: Annotated disjunction semantics include an implicit null outcome with probability `1 - sum(head_probabilities)` when `sum(head_probabilities) < 1`.
+- `P2-SEM-004`: Probabilistic clauses are lowered to an equivalent internal choice representation that preserves query marginals.
+- `P2-SEM-005`: Probabilities are computed as conditional marginals `P(Query | Evidence)` when evidence is present.
+- `P2-SEM-006`: Contradictory evidence produces an explicit inconsistent-evidence error.
+- `P2-SEM-007`: Models that yield non-ground probabilistic choices after grounding produce an explicit non-ground-probabilistic-clause error.
+- `P2-SEM-008`: Non-ground `query(...)` directives are supported when probabilistic parts are groundable.
+
+## Inference Backends
+
+- `P2-INF-001`: An exact inference backend exists for finite grounded models.
+- `P2-INF-002`: An approximate sampling backend exists for large models where exact inference is impractical.
+- `P2-INF-003`: Backend selection is explicit and observable through runtime configuration.
+- `P2-INF-004`: For finite benchmark models, exact and sampling backends agree within configured numeric tolerance.
+
+## Conformance Suite
+
+- `P2-CONF-001`: The conformance suite includes a two-coin disjunction case equivalent to `.tmp/problog/test/sample/some_heads.pl` with expected probability `0.8` for `someHeads`.
+- `P2-CONF-002`: The conformance suite includes evidence conditioning equivalent to `.tmp/problog/test/sample/some_heads_evidence.pl` with expected probability `0.6` for `someHeads` given `heads1=false`.
+- `P2-CONF-003`: The conformance suite includes annotated-disjunction marginals equivalent to `.tmp/problog/test/ad_fact.pl`.
+- `P2-CONF-004`: The conformance suite includes conditional Bayesian-network cases equivalent to `.tmp/problog/test/4_bayesian_net.pl` and `.tmp/problog/test/5_bayesian_net.pl`.
+- `P2-CONF-005`: The conformance suite includes inconsistent evidence failure equivalent to `.tmp/problog/test/01_inconsistent.pl`.
+- `P2-CONF-006`: The conformance suite includes non-ground probabilistic clause failure equivalent to `.tmp/problog/test/nonground.pl`.
+
+## Compatibility Guardrail
+
+- `P2-COMP-001`: Existing deterministic test suites for `just-prolog` and `just-prolog-tool` remain green after ProbLog support is added.

.opencode

@@ -0,0 +1 @@
+/Users/tom/Work/work/.opencode

.tasks/archive/20260227-reconciliation-reset/probe-write.md

@@ -0,0 +1 @@
+probe

.tasks/archive/20260227-reconciliation-reset/task-h-problog-bun-parity-suite-20260227203012.md

@@ -0,0 +1,15 @@
+---
+id: task-h-problog-bun-parity-suite
+level: high
+status: pending
+blocked_by: []
+expires_at: 2026-03-21T00:58:00Z
+---
+
+## Objective
+
+Establish Bun parity suite guarantees for top-level ProbLog fixture mapping, deterministic root execution, and drift-fail behavior.
+
+## Done-when
+
+`PBT-PATH-*`, `PBT-MAP-*`, `PBT-BEH-*`, `PBT-COV-*`, `PBT-SOURCE-004`, and `PBT-SOURCE-005` are demonstrably true in automated runs.

.tasks/archive/20260227-reconciliation-reset/task-h-problog-bun-parity-suite.md

@@ -0,0 +1,14 @@
+---
+id: task-h-problog-bun-parity-suite
+level: high
+status: pending
+blocked_by: []
+expires_at: 2026-03-20T23:17:17Z
+---
+
+## Objective
+Establish a Bun parity suite for ProbLog fixtures that can be run from repository root.
+
+## Done-when
+- Root `bun test` includes the intended ProbLog parity suite path.
+- Parity execution path and drift checks are defined and enforceable.

.tasks/archive/20260227-reconciliation-reset/task-h-problog-compatibility-guardrail-20260227181708.md

@@ -0,0 +1,20 @@
+---
+id: task-h-problog-compatibility-guardrail
+level: high
+status: done
+blocked_by: []
+expires_at: 2026-03-20T22:40:17Z
+---
+
+## Objective
+
+Restore deterministic compatibility across root and tool validation suites.
+
+## Done-when
+
+Root and tool regression paths pass with ProbLog support in place.
+
+## Green Phase Evidence
+
+- `npm run tool:test:run` passes in `packages/just-prolog-tool` after prebuilding root package with `--noCheck` to satisfy package entry resolution during Vitest import analysis.
+- `npm run test:run -- test/prolog.test.ts test/runtime-behavior.test.ts` passes in repository root.

.tasks/archive/20260227-reconciliation-reset/task-h-problog-compatibility-guardrail.md

@@ -0,0 +1,24 @@
+---
+id: task-h-problog-compatibility-guardrail
+level: high
+status: in-progress
+blocked_by: []
+expires_at: 2026-03-20T22:18:23Z
+---
+
+## Objective
+Restore compatibility guardrails so deterministic root and tool suites can run green together.
+
+## Done-when
+- Tool dependency/runtime issue is resolved.
+- Mono regression validation can confirm `P2-COMP-001` readiness.
+
+## Green phase evidence
+- `npm run tool:test:run` initially failed with `Cannot find package 'ai'` from `packages/just-prolog-tool/src/tool.ts`.
+- Added runtime `ai` dependency wiring in `packages/just-prolog-tool/package.json` and refreshed lock/install state.
+- Re-ran `npm run tool:test:run`; missing `ai` error no longer appears.
+- Current blocker: tool suite now fails earlier on `just-prolog` package entry resolution (`Failed to resolve entry for package "just-prolog"`).
+- `npm run test:run -- test/prolog.test.ts test/runtime-behavior.test.ts` passes (`2 passed`, `12 passed`).
+
+## Blockers
+- Tool tests are still red due to `just-prolog` workspace entry resolution in Vite import analysis, outside the requested `ai` missing package fix.

.tasks/archive/20260227-reconciliation-reset/task-h-problog-conformance-and-compat.md

@@ -0,0 +1,15 @@
+---
+id: task-h-problog-conformance-and-compat
+level: high
+status: done
+blocked_by: []
+expires_at: 2026-03-20T21:41:19Z
+---
+Objective
+- Establish ProbLog conformance coverage and deterministic compatibility guardrails for rollout safety.
+
+Done-when
+- Conformance/compat medium tasks are unblocked with fixture scope and regression criteria defined.
+
+Completion evidence
+- Red-phase delegation produced failing tests at `test/problog-conformance-and-compat.red.test.ts`.

.tasks/archive/20260227-reconciliation-reset/task-h-problog-examples-automation.md

@@ -0,0 +1,15 @@
+---
+id: task-h-problog-examples-automation
+level: high
+status: pending
+blocked_by: []
+expires_at: 2026-03-21T00:58:00Z
+---
+
+## Objective
+
+Add durable automation and evolution guardrails for ProbLog examples to prevent silent contract drift.
+
+## Done-when
+
+`P2EX-EVO-*` expectations are validated in repeatable offline automation.

.tasks/archive/20260227-reconciliation-reset/task-h-problog-examples-contract.md

@@ -0,0 +1,15 @@
+---
+id: task-h-problog-examples-contract
+level: high
+status: pending
+blocked_by: []
+expires_at: 2026-03-21T00:58:00Z
+---
+
+## Objective
+
+Close ProbLog examples contract gaps so each curated example is structurally valid and behaviorally verifiable.
+
+## Done-when
+
+`P2EX-STRUCT-004`, `P2EX-BEH-001`, and `P2EX-BEH-003` are enforced and tested as true.

.tasks/archive/20260227-reconciliation-reset/task-h-problog-examples-portfolio.md

@@ -0,0 +1,14 @@
+---
+id: task-h-problog-examples-portfolio
+level: high
+status: pending
+blocked_by: []
+expires_at: 2026-03-20T23:17:17Z
+---
+
+## Objective
+Create a reusable ProbLog examples portfolio with runnable validation.
+
+## Done-when
+- Examples schema and portfolio content exist with expected probability assertions.
+- Automation runs the examples without network dependencies.

.tasks/archive/20260227-reconciliation-reset/task-h-problog-language-and-api.md

@@ -0,0 +1,15 @@
+---
+id: task-h-problog-language-and-api
+level: high
+status: done
+blocked_by: []
+expires_at: 2026-03-20T21:41:19Z
+---
+Objective
+- Deliver the ProbLog language surface and public runtime API boundaries needed for implementation and integration.
+
+Done-when
+- Parser/API medium tasks are unblocked with concrete interfaces, syntax targets, and acceptance constraints.
+
+Completion evidence
+- Red-phase delegation produced failing tests at `test/problog-language-and-api.red.test.ts`.

.tasks/archive/20260227-reconciliation-reset/task-h-problog-nonground-query-parity-20260227181708.md

@@ -0,0 +1,31 @@
+---
+id: task-h-problog-nonground-query-parity
+level: high
+status: done
+blocked_by: []
+expires_at: 2026-03-20T22:40:17Z
+---
+
+## Objective
+
+Close the non-ground query parity gap for ProbLog behavior.
+
+## Done-when
+
+Non-ground query parity is implemented and validated against the red parity expectations.
+
+## Red evidence (P2-SEM-008)
+
+- Added/validated executable red spec in `test/problog-nonground-query-parity.red.test.ts` asserting non-ground query expansion returns per-ground-instance marginals for `p(1)` and `p(2)` and omits `p(X)` aggregate key.
+- `npm run test:run -- test/problog-nonground-query-parity.red.test.ts` fails on assertion: expected `result.probabilities["p(1)"]` to be close to `0.3`, received `undefined`.
+- Runtime loads cleanly via async import; failure is assertion-driven (no import crash).
+- Red phase complete; implementation remains pending.
+
+## Green evidence (P2-SEM-008)
+
+- Implemented non-ground `query(...)` expansion in `src/Prolog.ts` by grounding query terms against discovered constants and emitting per-instance query keys.
+- `query(p(X)).` now resolves to grounded keys (for this case `p(1)` and `p(2)`) and no longer emits existential-only `p(X)` probability output.
+- Validation passes:
+  - `npm run test:run -- test/problog-nonground-query-parity.red.test.ts`
+  - `npm run test:run -- test/problog-conformance-and-compat.red.test.ts`
+  - `npm run test:run -- test/prolog.test.ts test/runtime-behavior.test.ts`

.tasks/archive/20260227-reconciliation-reset/task-h-problog-nonground-query-parity.md

@@ -0,0 +1,19 @@
+---
+id: task-h-problog-nonground-query-parity
+level: high
+status: in-progress
+blocked_by: []
+expires_at: 2026-03-20T22:18:23Z
+---
+
+## Objective
+Define and deliver non-ground `query(...)` behavior that reaches ProbLog parity in this codebase.
+
+## Done-when
+- Contract and implementation path for non-ground query handling are accepted.
+- Work items tied to `P2-SEM-008` can be completed without ambiguity.
+
+## Red evidence
+- 2026-02-27: Ran `npm run test:run -- test/problog-nonground-query-parity.red.test.ts`.
+- Failing assertion: `expected undefined to be close to 0.3` at `test/problog-nonground-query-parity.red.test.ts:34`.
+- Failure indicates non-ground `query(p(X))` does not expose per-instance marginals like `p(1)` and `p(2)`.

.tasks/archive/20260227-reconciliation-reset/task-h-problog-sampling-backend-20260227181708.md

@@ -0,0 +1,35 @@
+---
+id: task-h-problog-sampling-backend
+level: high
+status: done
+blocked_by: []
+expires_at: 2026-03-20T22:40:17Z
+---
+
+## Objective
+
+Deliver a real sampling backend path for ProbLog inference.
+
+## Done-when
+
+Sampling mode executes stochastic estimation with seed-sensitive, sample-driven outputs.
+
+## Red evidence (P2-INF-002)
+
+- Command: `npm run test:run -- test/problog-sampling-backend.red.test.ts`
+- Result: 1 file failed, 3 tests failed, 0 runtime test errors.
+- Assertion failures confirm sampling path still reuses exact semantics:
+  - Missing stochastic marker: expected `inference.randomness` to equal `"stochastic"`.
+  - Exact-reuse mismatch: expected sampling marginal to differ from exact marginal, received identical value (`0.37`).
+  - Seed insensitivity: expected at least one marginal to change across seeds, received no changes.
+- Red phase complete: failing assertions capture the true-sampling backend gap.
+
+## Green evidence (P2-INF-002)
+
+- Implemented stochastic sampling backend path in `src/Prolog.ts` using seeded Monte Carlo world selection over grounded probabilistic groups.
+- Sampling results now expose stochastic metadata semantics:
+  - `inference.randomness: "stochastic"`
+  - `meta.executionPath: "sampling"`
+- Validation:
+  - `npm run test:run -- test/problog-sampling-backend.red.test.ts` -> 1 file passed, 3 tests passed.
+  - `npm run test:run -- test/prolog.test.ts test/runtime-behavior.test.ts` -> 2 files passed, 12 tests passed.