Changelog

All notable changes to VIMMaster are documented here.
Format: Keep a Changelog. Versioning: semver, starting with the first tagged release.

[Unreleased]

Nothing yet.

First tagged release. Everything below shipped together as the Community Alpha; nothing earlier was ever tagged.

Cheat Mode practice buttons work again. The content extraction turned vimLessons into a Map, but the lookup kept using bracket access — property access on a Map reads nothing, so every Practice button silently did nothing. Fixed with Map.get via a named getPracticeLesson(); regression-tested end-to-end (every catalog entry must resolve, and starting practice must load the lesson buffer).
Zero console noise in production (TD-11): ~110 debug console.logs deleted (several dumped full buffer state per keystroke); meaningful warnings/errors now flow through the structured logger (js/logger.js) with categories (lesson, progress, storage, ui) and debug filtering via localStorage.vimMasterDebug.
Lint is clean and stays clean: 195 warnings burned down to 0; ~170 unused imports/variables removed; no-console, no-unused-vars, no-empty, and no-debugger are now ESLint errors (with js/logger.js, scripts/, and tests/ exempted from no-console), enforced by CI on every PR.

Progress is saved the moment a level is won (TD-3). The intended on-completion save was a dead window.checkWinCondition wrapper that never ran; wins relied on the 30-second interval save and could be lost by closing the tab. The save now lives inside the win path itself. Regression-tested.
Challenge task scoring has a single source of truth (TD-6): dead endChallenge/checkChallengeTask exports (carrying a divergent 100-points formula players never experienced) are deleted; the live formula (10 + 1 point per 10s remaining) lives only in challenges.js#calculateTaskPoints, unit-tested, with dead imports cleaned from three modules.

Numeric counts now work as promised. 2x deleted a single character (counted edits looped over a stale buffer snapshot — TD-4b), and 2dd/2dw never worked at all (the operator's first key wiped the pending count). Counts survive multi-key commands and each loop iteration reads fresh state. 8 new regression tests.
0 no longer silently fails after j onto a shorter line (TD-4): the end-of-handler bounds clamp evaluated the cursor captured at handler entry and overwrote valid motions; it now clamps fresh state, and j onto a shorter line clamps the column into the line.
The Golden solution for lesson-practice-delete-character-practice no longer works around these engine bugs (uses counted x and natural j-then-0 order).

Every lesson is now provably solvable (PR24): all 35 lessons carry a solution key sequence; the Golden Suite auto-verifies every lesson with a solution (35/35 passing), and contract rule L007 is promoted from warning to error — a lesson without a working solution cannot be merged.
Lesson start positions restored as declarative initialCursor in every lesson JSON (they were lost in the content extraction; the engine ignored the field entirely, so all lessons started at {0,0} — caught by the ADR-0005 consumption invariant).
The lesson initializer now consumes the full content schema (initialCursor applied; id/version/metadata/focusCommand/solution acknowledged), eliminating per-load invariant warnings.

lesson-practice-insert-mode-practice was unwinnable (target text omitted the line's prefix — inherited from the original cheat-mode data); target, start cursor, and solution corrected.
lesson-practice-delete-character-practice had an unachievable target buffer; replaced with an x-achievable target.
lesson-practice-backward-search-practice target was one column past the search match; aligned to the match start.
Two engine defects discovered by the Golden Suite are documented as TD-4 (confirmed: 0 silently fails when the cursor is past EOL after j) and TD-4b (2x deletes one char — counted edits loop over a stale buffer copy) in docs/TechnicalDebt.md.

JSON lesson architecture: all levels are now loaded from external JSON files in content/lessons/.
Contract Suite: ensures valid structure for all community lessons.
Golden Suite: automated regression testing against actual engine logic.
Regression Suite: automated testing for previously resolved issues.
Content Provider: decoupling engine from lesson generation logic.
Community lesson pipeline: isolated workflow allowing non-programmers to contribute to lessons.
XP system: dynamically generated based on level difficulty metadata (beginner, intermediate, advanced).
Combo system: tracks back-to-back correct executions.
Progressive onboarding: hides advanced UI until appropriate levels are reached.
PROJECT_PRINCIPLES.md principle #8: User data is never destroyed automatically (absolute).
docs/architecture/constants.md: derivable constants must be derived — policy + review checklist.
docs/postmortems/ with PM-0001.
Governance & contributor infrastructure: VISION, PROJECT_PRINCIPLES, NON_GOALS, DECISIONS, PROJECT_STATUS, ROADMAP, CONTRIBUTING, CODE_OF_CONDUCT, SECURITY, GitHub issue/PR/RFC templates, ADR system, full /docs architecture suite, README banner.

UI simplification: stripped non-essential visual elements, heavily increasing focus on the editor (min-h-[320px]).
Focus command visualization: the "Level Complete" modal has been replaced with a dopamine-inducing card presenting the specific focused command.
Lesson metadata: added metadata schema for descriptions, authors, tags, and focus commands.
Progress persistence: saves XP, Combos, and Challenge Points in a backwards-compatible manner.

Level start positions now apply. loadLevel() discarded the result of each level's setup(), so every level started at {0,0} despite its configuration. Lessons (game levels and cheat-mode practice) now initialize through a single pipeline (docs/architecture/level-lifecycle.md) with a consumption invariant: every lesson property must be consumed exactly once during initialization — unconsumed or malformed configuration is reported immediately instead of silently ignored (ADR-0005).
Saves are no longer destroyed after finishing the game (PM-0001). The progress validator hardcoded a 15-level range while the game has 16 levels; reaching the final level made the save "invalid", and the loader deleted it on the next visit. All level counts are now derived from levels.length.
Invalid saves are repaired, never deleted. The load path is now: load → validate → repair → otherwise keep the original, write a backup (vimMasterProgressBackup), and notify the player. The age-based save rejection (saves older than 1 year were deleted) is removed.
Saved level is actually resumed. Loading previously applied the saved level to state but then always started the session at level 1.
"Clear Progress" (user-initiated) now backs up the save before clearing.
The "16 Progressive Levels" text on the page is now computed from the level data.