@unified-product-graph/core 0.9.7 → 0.9.9

package/CHANGELOG.md

@@ -7,9 +7,45 @@ This project uses [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
 ---
 
+## [0.9.9] - 2026-06-10
+
+**Structural-integrity forks: validation, AI-prompt, and journey models corrected; duplicate edges collapsed.** A spec-model cut that fixes several long-standing structural mismodels and retires shadow edges. Every renamed or retired edge ships a dual-read migration (`UPG_EDGE_MIGRATIONS['0.9.9']`), so stored graphs upgrade automatically on load. No new tools (118) and no entity-count change; the within-product edge catalogue settles at 975 after the duplicate collapse.
+
+### Changed
+- **Validation experiment-model reshape.** The validation chain is corrected to the single-parent line `hypothesis → experiment_plan → experiment → experiment_run`. The plan-to-experiment containment inverts: `experiment_has_plan` becomes `experiment_plan_designs_experiment` (the plan designs the experiment it owns), and `experiment` gains its stable hypothesis loop (`hypothesis_tested_by_experiment`, `experiment_validates_hypothesis`).
+- **`test_plan` re-homed validation → QA.** `test_plan` is now a QA artefact rather than a validation planner; its planning role is absorbed by `experiment_plan`. New QA edges `product_plans_qa_via_test_plan`, `test_plan_executed_by_test_suite`, and `test_plan_specifies_test_environment`; the validation-side `hypothesis_planned_via_test_plan` and `test_plan_ran_as_experiment_run` are dropped.
+- **AI prompt-model correction.** The prompt abstraction is inverted to the correct `ai_model → prompt_template → prompt_version` containment (`ai_model_defines_prompt_template`, `prompt_template_contains_prompt_version`, `prompt_version_supersedes_prompt_version`), with the evaluation, tracing, and provenance bridges (`eval_run_evaluates_ai_model`, `eval_run_scores_prompt_version`, `ai_trace_executed_prompt_version`, `hallucination_report_traces_to_ai_trace`, `hallucination_report_caused_by_root_cause`, `ai_dataset_sourced_from_data_source`). The backwards/obsolete prompt edges are dropped.
+- **Strategic-theme containment edge renamed.** `objective_rolls_up_to_strategic_theme` → `strategic_theme_contains_objective` (source-first key, clean `contains` verb; endpoints unchanged).
+- **`touchpoint_channel` constrained to a medium enum** (`in_app` | `email` | `phone` | `chat` | `sms` | `in_person` | `mail`), disambiguating the interaction *medium* from a go-to-market *channel* (`marketing_channel` / `acquisition_channel` / `distribution_channel`).
+- **Journey-model de-duplication** and **duplicate / shadow edge collapse**: byte-identical shadows, tense/suffix twins, and near-synonym edges collapse onto their surviving canonical keys (each dual-read), lowering the edge-duplicate collision bound. `UPG_EDGE_COUNT` 979 → 975.
+
+### Migration
+- **`UPG_EDGE_MIGRATIONS['0.9.9']`** dual-reads every renamed or dropped edge key, so a graph written by an earlier version resolves to the new canonical edges on load. Renames with an inverted direction carry a `flip`; superseded edges with no structural replacement are `drop`ped.
+
+---
+
+## [0.9.8] - 2026-06-10
+
+**Registry lifecycle + portfolio-tier audience and metric edges (batch 5).** Closes the lifecycle gaps found while standing up a real shared-entity registry: canonicals can now be edited, batch-created, and promoted from existing nodes; deliberate name divergences can be sanctioned; and two new portfolio edges connect the org axis to its audience and complete the measurement cascade.
+
+### Added
+- **Three cross-edge types** (`UPG_CROSS_EDGE_TYPES` 9 to 12): `area_serves_persona` and `area_targets_market_segment` (a `product_area` to a canonical registry persona / market_segment, carrying optional primary/secondary `relevance` and an `audience_role`), and `rolls_up_to` (a product metric feeds a company/portfolio metric: the measurement cascade, mirroring `contributes_to` for the OKR cascade).
+- **Optional `UPGCrossEdge` qualifiers**: `alias` (sanction a deliberate instance title divergence), `relevance`, and `audience_role`. The canonical serialiser round-trips all three.
+- **Five MCP tools** (113 to 118): `update_canonical_entity` (edit a canonical without disturbing its instances), `batch_define_canonical_entity` and `batch_register_instance` (atomic registry stand-up), `promote_to_canonical` (lift an existing node into the registry), and `link_area_to_audience` (create the area audience edges with qualifiers).
+- **`get_node` resolves `registry/{id}`**: registry entities are first-class for reads, returning the canonical plus its instances.
+
+### Changed
+- **`register_instance` accepts `alias`**, and `portfolio_validate`'s `registry_drift` honours it: a sanctioned divergence leaves the drift count (a new `sanctioned` tally), so `clean: true` means "no unexpected drift".
+- **`create_edge` / `batch_create_edges` error hints**: a `p_` product-header id used where an in-graph node id is expected returns a targeted identity error, and an unknown explicit edge type returns a `did_you_mean` resolved from the endpoint types.
+- **Generic `create_cross_product_edge`** accepts `rolls_up_to` and rejects the registry/area edge types (directing to `register_instance` / `link_area_to_audience`).
+
+No entity, domain, or within-product edge-catalogue count change. The area-anchored edges carry a portfolio `product_area` id as their source: the only cross-edge source that is not a `{product_id}/{node_id}` pair.
+
+---
+
 ## [0.9.7] - 2026-06-10
 
-**Domain-wiring remediation (UPG-666 epic, passes F1–F8) + saturated-fixture refresh.** A spec-integrity hardening cut: eight remediation passes from the 36-domain wiring audit, each shipping its own machine-checkable gate so the class of drift it fixes cannot silently recur. No new tools and no breaking API surface — additive edges/properties plus integrity gates. Audit ledger: `TPC-context/upg/07-initiatives/domain-wiring-audit/`.
+**Domain-wiring remediation (passes F1–F8) + saturated-fixture refresh.** A spec-integrity hardening cut: eight remediation passes from a 36-domain wiring audit, each shipping its own machine-checkable gate so the class of drift it fixes cannot silently recur. No new tools and no breaking API surface — additive edges/properties plus integrity gates.
 
 ### Added
 - **Region/manifest integrity gate** (F1): `region-integrity.test.ts` proves the 10 canonical regions cover every active entity type exactly once, with the Nucleus tier explicitly exempt via `NUCLEUS_DOMAINS` / `SHARED_TYPES`. Reconciles all 10 region rosters against the domain registry (un-orphaning the entities that had drifted out of their region — the P-H pattern) and corrects 7 phantom intra/boundary edge references. Decisions: `2026-06-10-nucleus-region-exemption.md` plus three model-reconciliation ADRs.