v1.3.0 — Explainable Forecasting Knowledge System

Explainable Forecasting Knowledge System

• 将 Explainable Research Layer 升级为 artifact-bound 预测知识学习系统。
• 新增 knowledge/ 结构化知识库、来源索引、知识注册表、evidence linker 和 lesson engine。
• 每次渲染生成 knowledge_explanations.json、learning_cards.json、formula_manifest.json 和 TeX/MathML/SVG 公式资产。
• 公式状态统一为 not_attempted、compiling、compiled、blocked、failed、not_available，编译失败返回明确 reason code。
• 离线 Dashboard 增加 claim、formula variable、learning card 和 lineage 交互；浏览器端不重新计算金融结果。
• MCP get_formula、explain_metric、get_provenance 和 preview_scenario 可读取完整公式、学习卡和证据链。
• README/SKILL/reference/schema/example/test 与 clean-install 支持同步更新。

Validation: pytest, schema validation, source registry validation, skill quick_validate, release gate and wheel clean-install passed.

v1.2.1 — Interactive Dashboard Packaging Fix

修复 clean-install wheel 未包含 interaction.js 的问题，新增 package-data，确保安装后的 fro-html 与源码一致。保留 v1.2.0 的离线交互 Dashboard、Explainable Research、TeX 公式链和 MCP 情景能力。

v1.2.0 — Interactive Explainable Dashboard

新增离线可互动 Dashboard、Explainable Research 输出、TeX 公式构建链和 MCP 受控情景分析。浏览器交互只筛选/展开已保存 artifact；情景重算回到 Python Runtime，并生成新的 scenario_id、lineage 和结果 hash。pytest、schema validator、source registry、skill validator、release gate 和 wheel clean-install 已通过。

v1.1.0 — Reliable Research Calculations

升级组合约束求解、适用性门控的 DM/WRC/SPA/DSR/PBO、情景压力重算、Forecasting 可信度状态、稳定 Python API、离线 Dashboard 与 Explainable Research 层。Release gate、schema、source registry、skill validator、pytest 和 wheel clean-install 已通过。

v1.0.0 — Stable Auditable Research Runtime

v1.0.0 — Stable Auditable Research Runtime

• Completed default overfitting applicability, portfolio baseline/fallback, and post-selection stress-test nodes.
• Frozen v1 protocol schemas with schema_version=1.0 and v0.9 compatibility.
• SQLite WAL is the default multi-client run store; JSON remains an explicit offline compatibility backend.
• Added run-store migrate/doctor/backup/restore and the fro-release-gate release command.
• Missing capabilities fail closed with reason_code, next_action, and user_action_required.
• Preserved source maturity honestly: 2 live-certified, 3 degraded, 9 manual-review, and 5 blocked capabilities.

v0.9.0 — Online Source Grid & Research-grade Forecasting

v0.9.0

• Added a real National Data v2 live smoke path with raw snapshot, schema fingerprint, freshness, health and revision evidence. Because the public endpoint does not expose historical release/availability timestamps, it remains L3 and is not promoted to L4.
• Added real ECB SDMX live smoke and explicit L4 certification with health, freshness, vintage-aware PIT, revision, snapshot hash and expiry gates.
• Added explicit verified-snapshot degradation records for SSE, PBC and CNINFO; these are reported as degraded and never as live official fetches.
• Research-grade forecasting now compares historical mean, naive persistence and rolling mean under expanding-window evaluation, with residual intervals, selection metric, calibration status and OOD status.
• Added fro-forecast and fro-maturity-report; MCP status now includes stage, progress, source status, freshness, next action and user-action requirement.
• Added tests for the NBS v2 parser, fallback maturity grid and three-baseline forecasting contract.

Validation: pytest, schema validator, source registry validator, skill quick validator and v0.9.0 package build passed.

v0.8.2 — Semantic and Certification Hardening

v0.8.2

• Hardened the local forecasting loop with canonical normalization, date/grain/order/future-leak/PIT audits, availability-time preservation, baseline comparison, and corrected result lineage.
• Made adapter evidence explicitly three-dimensional: source_id + dataset_id + access_method; L4 now requires verified revision evidence, snapshot hash, healthy/fresh status, and an unexpired certification.
• Added certification-cycle state with failure counters and circuit status; expired evidence downgrades to L3. Added an explicit SSE verified-snapshot fallback that is reported as degraded, not live.
• Added HTTPS/SSRF/private-address, cross-origin redirect, download extension, and file-size guards for browser transports.
• Added v0.8.2 CLI metadata and published clean-install artifacts.

Validation: pytest, schema validator, source registry validator, skill quick validator, wheel build and installed CLI checks passed.

v0.8.1 — SEC EDGAR L4 Evidence

v0.8.1

Patch release adding the first real L4-certified adapter evidence after the local CA trust chain was restored.

• SEC EDGAR company_facts live smoke: HTTP 200, 25,135 normalized rows.
• Freshness: ready; health: healthy; PIT/revision: verified.
• Snapshot hash recorded and second-run schema drift comparison passed with no drift.
• Explicit certify_adapter.py --write completed; live_certified_count=1.
• Certification expires 2026-10-06 and must be renewed from reviewed live evidence.

No other source was promoted automatically.

v0.8.0 — Default Forecasting Loop & Clean Install

v0.8.0

This release follows the close-loop-first roadmap.

Highlights

• Default run_research(...) can execute local CSV/JSON forecasting without custom handlers.
• The path now emits canonical data, quality audit, naive/historical-mean baselines, rolling evaluation, forecast contract, lineage, offline HTML, and CSV/Markdown decision tables.
• Added dataset_path, artifact_dir, and explicit require_point_in_time contract controls. Missing availability_time blocks PIT-required forecasting with actionable status fields.
• Added installed CLI entry points for adapter smoke, certification, schemas, data quality, and source health.
• Added Python 3.11/3.12 wheel clean-install CI.
• Added explicit verified-snapshot fallback support; no silent fallback is introduced.
• Fixed SEC adapter cache/snapshot wiring and structured online acquisition failures.

Evidence boundary

A real SEC live smoke was attempted and recorded as blocked because the local Python runtime could not verify the remote CA certificate. No adapter was promoted to L4. The repository continues to require reviewed live smoke evidence, freshness, health, PIT/revision evidence, snapshot hash, and explicit expiry before certification.

v0.7.0 — L4 Evidence & Executable Research Loop

v0.7.0

This release turns adapter maturity and the default research path into executable, auditable contracts.

Highlights

• Composite adapter evidence identity: source_id + dataset_id + access_method.
• Factory is the single execution registry; unknown sources fail closed.
• Adapter smoke now fingerprints schemas and compares against the previous snapshot.
• Added explicit scripts/certify_adapter.py for opt-in L4 live certification. Replay evidence cannot promote an adapter.
• Added a minimum local-data loop: normalize → audit → baseline features → historical-mean forecast → result lineage → offline HTML and decision tables.
• Actionable runtime/MCP status fields: reason_code, next_action, and user_action_required.
• SourceResult contract now requires source_capability and freshness; schema examples and validation are synchronized.

L4 remains evidence-gated: this release does not claim live certification without a reviewed live smoke artifact.