docs: changelog for v0.2.0 (2026-07-26)

CHANGELOG.md

@@ -0,0 +1,61 @@
+# Changelog
+
+All notable changes to this project will be documented in this file.
+
+The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/),
+and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
+
+## [Unreleased]
+
+## [0.2.0] - 2026-07-26
+
+First changelog entry covering development on `main` from 2026-07-19 through 2026-07-26.
+
+### Added
+
+- Tiered CA persona extraction and LLM prediction framework with the `ca-personas` CLI (#4)
+- Quarto manuscript website for Posit Connect Cloud deployment (#6)
+- Stage-one Random Forest and k-NN CA prediction baselines (#7)
+- Band accuracy metrics, ground-truth scoring script, and full persona prompt bundles (#10)
+- Distance-from-correct evaluation metrics for CA subscale prediction (#11)
+- Factor analysis and feature-importance notebook (#12)
+- ML vs LLM shared-metric comparison notebook (#13)
+- vLLM digital-twin inference launcher for batch persona generation (#14)
+- Secondary RQ: regular public transit vs cohort CA scores (`ca-personas transit-ca`) (#16)
+- Secondary RQ: Random Forest of lat/long predicting regular transit (`ca-personas geo-transit-rf`) (#17)
+- Secondary RQ5: Random Forest of CA scores predicting regular transit (`ca-personas ca-transit-rf`) (#18)
+- Research memo and CLI for CA scores predicting regular transit use (#20)
+- Comprehensive feature-importance RF suite and research memo for regular transit (#23)
+- SHAP/F1 feature predictive-power suite for ML and LLM agents (`ca-personas shap-eval`) (#26)
+- Geo-memo follow-up RFs: car access, employment, and ride-share predicting regular transit (#27)
+- Manuscript and memo for Q27/Q28 traditional ML predictors of regular transit (#28)
+- Full-cohort Posit stats rerun and APA-style site figures (#29)
+- Wave-2 follow-up experiments, research memos, and Quarto site pages (`ca-personas followup-experiments`) (#30)
+- Student status incorporated into the base demos persona tier end-to-end (#31)
+- JackJBurleson Posit Connect publish skill (full-data gated) (#32)
+- Two sample persona prompts per cumulative context tier (#33)
+- Expanded landing abstract, secondary RQ presentation, and GitHub access guidance (#34)
+- AI Terrarium natural-language digital-twin persona prompt rework (#35)
+- Seven-model CA ML baseline suite: Ridge, Elastic Net, k-NN, RF, HistGradientBoosting, XGBoost, and MLP (#36)
+- AI Terrarium persona examples surfaced on the landing page (#38)
+- `AGENTS.md` agent runbook with Cursor Cloud dev environment setup notes (#42)
+- GitHub PR and issue templates; expanded `AGENTS.md` contributor guidance (#43)
+
+### Changed
+
+- Align data pipeline to File A/B/C sibling exports with RQ-focused EDA (#15)
+- Refresh ML suite baseline figures after full-cohort republish (#37)
+- Clean Quarto site navigation: separate Memos from Secondary RQs (#39)
+- Harden cohort loading and sync secondary Posit full-cohort artifacts (#44)
+- Audit Posit site copy: update stale ML claims, tighten prose, and fix path references (#41)
+
+### Fixed
+
+- File C flat-header detection; enforce 252/21/10 merge coverage audit (#19)
+- Silent failures, path fallbacks, placeholders, and doc/requirements drift (#24)
+- `comprehensive-transit-rf` CLI regression and full-cohort republish (#40)
+
+### Notes
+
+- Posit Connect publishing requires private File A/B/C exports and credentials; see `.cursor/skills/posit-connect-publish/`
+- Manuscript and artifact numbers are tied to `--seed 42` and `--join inner` on the full matched cohort