# 🃏 MTG Python Deckbuilder

[![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://opensource.org/licenses/MIT)
[![Python 3.11+](https://img.shields.io/badge/python-3.11+-blue.svg)](https://www.python.org/downloads/)
[![Docker](https://img.shields.io/badge/docker-supported-blue.svg)](https://www.docker.com/)

A fast, menu-driven deckbuilder for MTG Commander/EDH with both CLI and Web UI. Supports interactive and headless runs, CSV/TXT exports, owned-card libraries, and Docker.

## 🔹 Features
- Web UI and CLI, powered by the same builder
- Commander search with smart theme tagging (primary/secondary/tertiary)
- Power bracket selection and ideal count prompts
- **Include/Exclude Card Lists**: Must-include and must-exclude functionality with fuzzy matching, visual validation, and strict enforcement (set `ALLOW_MUST_HAVES=true`)
- CSV and TXT exports with stable filenames
- Headless mode (non-interactive) and a headless submenu in the main menu
- Config precedence: CLI > env > JSON > defaults
 - Visual summaries in Web UI: Mana curve, Color pips and Sources with hover-to-highlight and copyable tooltips
- Web UI: Locks, Replace flow, Compare builds, and shareable permalinks
 - Unified “New Deck” modal (steps 1–3): commander search, themes, bracket, options, and optional Name (used in export filenames)
 - Multi-copy archetypes (opt-in modal): choose quantity for supported archetypes and optionally add Thrumming Stone; added early with target adjustments and a hard 100-card clamp
 - Combos & Synergies: detect curated pairs, show chip-style lists with badges, and dual-card hover previews; optional auto-complete adds missing partners based on your plan.

## ✨ Highlights
- Smart tagging and suggestions for commander + themes, with AND/OR combine modes
- Theme governance: generated theme catalog with whitelist normalization, enforced synergies, and capped synergy lists (top 5) for concise UI; roadmap includes YAML authoring export.
- Exports CSV and TXT decklists to `deck_files/`
- Random Full Build (alpha) now outputs a single authoritative artifact set (CSV, TXT, compliance JSON, summary JSON) without duplicate suffixed files; API returns paths for integration. Summary sidecars carry multi-theme metadata (`primary_theme`, `secondary_theme`, `tertiary_theme`, `resolved_themes`, fallback flags) plus legacy aliases for incremental consumer upgrades.
- Random Mode (alpha) multi-theme groundwork: curated pools now support manual exclusions via `config/random_theme_exclusions.yml` (documented in `docs/random_theme_exclusions.md` and exportable with `python code/scripts/report_random_theme_pool.py --write-exclusions`); UI/diagnostics surface fallback telemetry while the upcoming picker adds Primary/Secondary/Tertiary inputs with deterministic fallback (P+S+T → P+S → P+T → P → synergy overlap → full pool) and an inline “Clear themes” control quickly resets stored inputs.
- Random Mode instrumentation: `python code/scripts/profile_multi_theme_filter.py` captures mean/p95 filter timings, while `python code/scripts/check_random_theme_perf.py` guards against regressions relative to `config/random_theme_perf_baseline.json` (`--update-baseline` refreshes).
- Owned-cards mode: point to one or more `.txt`/`.csv` files in `owned_cards/` and choose "Use only owned cards?"
- If you opt out, final CSV marks owned cards with an `Owned` column
- Build options include Prefer-owned: bias picks toward owned cards while allowing unowned fallback
- Interactive menu with a headless submenu
- Works locally or in Docker (Windows PowerShell friendly)
- Card images and data via Scryfall (attribution shown in the Web UI)
 - Web visual summaries: cross-highlight cards from charts; sources include non-lands and colorless 'C' with an on/off toggle
 - New Deck modal keyboard UX: Enter selects the first suggestion; arrow navigation removed; browser autofill disabled
 - Export naming: optional Name in the modal becomes the filename stem for CSV/TXT/JSON; decks include a sidecar `.summary.json` with `meta.name`
 - Finished Decks and deck banners prefer your custom Name when present
 - Compare page includes a Copy summary button to quickly share diffs
 - New Deck modal shows your selected themes and their order (1 → 3) while picking; Bracket options are labeled (e.g., "Bracket 3: Upgraded"). Default bracket is 3.
 - Bracket policy enforcement: disallowed categories (e.g., Game Changers in Brackets 1–2) are pruned from the pool up-front; UI blocks progression until violations are resolved, with inline Replace to pick compliant alternatives.
 - Bracket compliance UX: the compliance panel auto-opens when non-compliant and shows a colored overall chip (green/WARN amber/red). WARN thresholds (e.g., tutors/extra turns) are advisory—tiles show with a subtle style and no gating; FAIL continues to gate and enables enforcement.

### Commander eligibility rules (clarified)
The builder now applies stricter Commander legality filtering when generating `commander_cards.csv`:
1. Automatically eligible:
  - Legendary Creatures (includes Artifact Creature / Enchantment Creature lines)
  - Legendary Artifact Vehicles or Spacecraft that have printed power and toughness
  - Any card whose rules text explicitly contains "can be your commander"
2. Not auto‑eligible (unless they have the explicit text above):
  - Plain Legendary Planeswalkers
  - Plain Legendary Enchantments without the Creature type
  - Generic Legendary Artifacts (non‑Vehicle/Spacecraft or without P/T)
This prevents over-broad inclusion of non-creature legendary permanents.

## 🚀 Quick start

### Docker Hub (PowerShell)
```powershell
docker run -it --rm `
  -v "${PWD}/deck_files:/app/deck_files" `
  -v "${PWD}/logs:/app/logs" `
  -v "${PWD}/csv_files:/app/csv_files" `
  -v "${PWD}/owned_cards:/app/owned_cards" `
  -v "${PWD}/config:/app/config" `
  mwisnowski/mtg-python-deckbuilder:latest
```

### From source
```powershell
pip install -r requirements.txt
python code/main.py
```

### From this repo with Docker Compose (PowerShell)
```powershell
docker compose build
docker compose run --rm mtg-deckbuilder
```

## 🌐 Web UI

Run the browser-based UI backed by the same deckbuilding engine.

### From source
```powershell
pip install -r requirements.txt
uvicorn code.web.app:app --host 127.0.0.1 --port 8080
```

Open http://127.0.0.1:8080

### Docker Compose (PowerShell)
```powershell
docker compose build web
docker compose up --no-deps web
```

Then open http://localhost:8080

Notes:
- Exports/logs/configs use the same folders and will appear under your mounted volumes (see compose file).
- The footer includes Scryfall attribution per their guidelines.
 - Favicon is bundled; browsers load `/favicon.ico` (auto-falls back to PNG).
 - Bracket enforcement is applied in both web and headless runs; see "Bracket policy & enforcement" below.

### Run with Docker Compose (Docker Hub image)

Use the prebuilt image with the provided `dockerhub-docker-compose.yml`. The image defaults to the Web UI.

Web (default):
```powershell
docker compose -f dockerhub-docker-compose.yml up -d
```

Pin to a version (edit the compose file image tag to `:X.Y.Z`) to avoid `latest` drift.

CLI mode (override default):
```powershell
docker compose -f dockerhub-docker-compose.yml run --rm `
  -e APP_MODE=cli `
  web
```

Notes:
- Volumes under `${PWD}` persist your exports/logs/configs locally next to the compose file.
- Health checks are built into the image for the Web UI. In CLI, `APP_MODE=cli` bypasses health.

#### Theme defaults and flags
- Enable the Theme selector with `ENABLE_THEMES=1` (on by default in `web` in compose).
- Set the initial default with `THEME=system|light|dark` (applies only when the browser has no prior preference saved):
  - `THEME=system` (recommended): follows the OS theme; Light uses the consolidated Light (Blend) palette.
  - `THEME=light`: maps to the consolidated Light (Blend) theme in the UI.
  - `THEME=dark`: uses the dark theme.
- You can also force a one-off override via URL: `?theme=system|light|dark|high-contrast|cb-friendly`.
  - Example: `http://localhost:8080/?theme=dark`
  - The choice is saved in localStorage; the URL parameter is removed after applying.
  - Reset: Use the “Reset theme” control in the header to clear your preference and re‑apply the server default (or system).
 - YAML export strategy: By default the app now always regenerates per-theme YAML files even on fast refreshes to keep editorial artifacts synchronized with `theme_list.json`. To opt-out on incremental refreshes (rare), set `THEME_YAML_FAST_SKIP=1` (compose variable). This only skips export during fast-path (no retag) theme refreshes and never during full builds.

#### Theme catalog editorial pipeline
Build script: `python code/scripts/build_theme_catalog.py`

Key flags / env vars:
- `--limit N` (preview subset; guarded from overwriting canonical JSON unless `--allow-limit-write`)
- `--output path` (write catalog to alternate location; suppresses YAML backfill so tests don't mutate source files)
- `--backfill-yaml` or `EDITORIAL_BACKFILL_YAML=1` (write auto `description` + derived/pinned `popularity_bucket` into each YAML if missing)
- `--force-backfill-yaml` (overwrite existing description/popularity_bucket)
- `EDITORIAL_SEED=<int>` (deterministic ordering for inference and any randomized fallbacks)
- `EDITORIAL_AGGRESSIVE_FILL=1` (pad ultra-sparse themes with extra inferred synergies)
- `EDITORIAL_POP_BOUNDARIES="a,b,c,d"` (override popularity bucket thresholds)
- `EDITORIAL_POP_EXPORT=1` (emit `theme_popularity_metrics.json` summary)

Derived fields:
- `popularity_bucket` (Very Common / Common / Uncommon / Niche / Rare) based on summed color frequency; authors may pin a value in YAML.
- `description` auto-generated if absent using heuristics keyed to theme class (tribal, counters, graveyard, tokens, etc.).

Editorial quality & metadata info:
- Optional `editorial_quality: draft|reviewed|final` can be set in per-theme YAML (or force-backfilled) to track curation progress.
- Backfills now stamp a `metadata_info` block (formerly `provenance`) with keys such as `last_backfill`, `script`, and `version`. The legacy key `provenance` is still parsed for a short deprecation window; if both appear a warning is emitted (suppress via `SUPPRESS_PROVENANCE_DEPRECATION=1`). Planned removal of the legacy alias: version 2.4.0.
- A running specialization trend is appended to `config/themes/description_fallback_history.jsonl` whenever `EDITORIAL_INCLUDE_FALLBACK_SUMMARY=1` is set.
- When `EDITORIAL_INCLUDE_FALLBACK_SUMMARY=1`, the catalog additionally embeds a `description_fallback_summary` object exposing coverage KPIs (see below) to power regression tests.

Synergy pairs & clusters:
- `config/themes/synergy_pairs.yml` provides curated fallback synergies (applied only if a theme YAML lacks `curated_synergies`).
- `config/themes/theme_clusters.yml` groups themes (Tokens, Counters, Graveyard, etc.) for upcoming UI filters and analytics.
  - Validator checks structure; UI integration planned in Phase E.

Description mapping externalization:
- `config/themes/description_mapping.yml` fully replaces internal mapping when present.
  - Items: `triggers` (list of substrings, lowercase-insensitive) and `description` string.
  - `{SYNERGIES}` placeholder expands to a short clause with top synergy names or is removed gracefully if none.
  - Mapping validator (`scripts/validate_description_mapping.py`) reports duplicate triggers, suggests `{SYNERGIES}` usage, and validates synergy pairs & clusters.

KPI targets (initial schedule):
- Wave 1 (current): Maintain regression test thresholds (generic_pct < 52%).
- Wave 2: Ratchet to <45% after next mapping expansion.
- Wave 3: <38%; Wave 4: <30% (issue templates will adjust thresholds once prior target is achieved and history confirms downward trend).

Lint (`code/scripts/lint_theme_editorial.py`) enhancements:
- `--require-description` / env `EDITORIAL_REQUIRE_DESCRIPTION=1` enforces descriptions present.
- `--require-popularity` / env `EDITORIAL_REQUIRE_POPULARITY=1` enforces popularity bucket present.
- Strict mode also escalates missing cornerstone examples and missing description/popularity to errors.

Testing:
Determinism, boundary overrides, and backfill isolation validated in `code/tests/test_theme_catalog_generation.py` using the new `--output` flag.

### Commander catalog upkeep
- The Commander Browser and auto-select flows read from `csv_files/commander_cards.csv`; keep it version-controlled so web and CLI stay in sync.
- Regenerate the file after MTGJSON updates or commander policy tweaks:
  ```powershell
  python -c "from file_setup.setup import regenerate_csvs_all; regenerate_csvs_all()"
  ```
- For quick fixes, you can refresh just the commander list:
  ```powershell
  python -c "from file_setup.setup import determine_commanders; determine_commanders()"
  ```
- Full onboarding guidance (required columns, caching notes, validation steps) lives in `docs/commander_catalog.md`.
- Staging toggle: set `SHOW_COMMANDERS=0` to hide the sidebar link while testing privately; defaults to on.

### Theme Catalog API
Two new endpoints expose the merged catalog for the upcoming theme picker UI.

List themes:
`GET /themes/api/themes`
Parameters:
- `q` substring across theme name & synergies
- `archetype` filter by `deck_archetype`
- `bucket` popularity bucket (Very Common|Common|Uncommon|Niche|Rare)
- `colors` comma-separated color initials (e.g. `G,W`)
- `limit` (default 50, max 200) & `offset`
- `diagnostics=1` (requires env `WEB_THEME_PICKER_DIAGNOSTICS=1`) to expose `has_fallback_description` + `editorial_quality`

Sample response snippet:
```
{
  "ok": true,
  "count": 314,
  "items": [
    {
      "id": "plus1-plus1-counters",
      "theme": "+1/+1 Counters",
      "primary_color": "Green",
      "secondary_color": "White",
      "popularity_bucket": "Very Common",
      "deck_archetype": "Counters",
      "description": "...",
      "synergies": ["Adapt","Evolve","Proliferate","Counters Matter","Hydra Kindred"],
      "synergy_count": 5
      // diagnostics-only: has_fallback_description, editorial_quality
    }
  ],
  "next_offset": 50,
  "generated_at": "2025-09-19T12:34:56",
  "diagnostics": false
}
```

Theme detail:
`GET /themes/api/theme/{id}`
Parameters:
- `uncapped=1` (diagnostics only) returns `uncapped_synergies` (full curated+enforced+inferred ordered set)
- `diagnostics=1` enables extra fields (fallback + editorial quality)

Detail response excerpt:
```
{
  "ok": true,
  "theme": {
     "id": "plus1-plus1-counters",
     "theme": "+1/+1 Counters",
     "description": "...",
     "synergies": ["Adapt","Evolve","Proliferate","Counters Matter","Hydra Kindred"],
     "curated_synergies": [...],
     "enforced_synergies": [...],
     "inferred_synergies": [...],
     "example_commanders": [...],
     "example_cards": [...],
     "synergy_commanders": [...]
     // diagnostics-only: uncapped_synergies, editorial_quality, has_fallback_description
  }
}
```

Caching & freshness: ETag header reflects catalog size/mtime + newest YAML mtime. Use `/themes/status` for stale detection; POST `/themes/refresh` to rebuild in background.

Environment flag: `WEB_THEME_PICKER_DIAGNOSTICS=1` enables diagnostics extras and uncapped synergy toggle.

Performance & optimization:
- Fast list filtering uses precomputed lowercase search haystacks + memoized filtered slug cache (keyed by catalog ETag and query params) for sub‑50ms typical responses once warm.
- Skeleton loaders (picker shell, table list fragment, preview modal) improve perceived load time; real content swaps in via HTMX fragment requests.
- Metrics endpoint `/themes/metrics` (gated by `WEB_THEME_PICKER_DIAGNOSTICS=1`) exposes:
  - Catalog filter requests, cache hits, cache entry count
  - Preview requests, cache hits, average build time (ms), cache size, TTL
- Response headers: `X-ThemeCatalog-Filter-Duration-ms` and `ETag` let you observe server-side filter cost and enable conditional requests.

### Governance & Editorial
The theme catalog follows lightweight governance so curation quality scales:

1. Minimum Examples Threshold: Target 2 example cards and 1 example commander for every established theme. Enforcement flips from warning to required once coverage >90% (tracked via metrics). Until then, themes below threshold are flagged but not blocked.
2. Curation Ordering: Preview assembly order is deterministic — examples, curated synergy examples, sampled role cards (payoff/enabler/support/wildcard), then synthetic placeholders only if needed to reach target count.
3. Splash Relax Policy: Four- and five-color commanders may include a single off-color enabler (scored with a -0.3 penalty) to avoid over-pruning multi-color creative lines. This is documented to prevent accidental removal during future sampling refactors.
4. Popularity Buckets Are Non-Scoring: Popularity is for filtering and subtle UI hints only and must never directly influence sampling scores (avoids positive feedback loops that crowd out niche archetypes).
5. Taxonomy Expansion Criteria: Proposed new high-level themes (Combo, Storm, Extra Turns, Group Hug / Politics, Pillowfort, Toolbox / Tutors, Treasure Matters, Monarch / Initiative) must demonstrate: (a) a distinct strategic pattern, (b) at least 8 broadly played representative cards, (c) not a strict subset of another existing theme.
6. Editorial Quality Tracking: Optional `editorial_quality: draft|reviewed|final` in theme YAML guides review pass prioritization; metrics aggregate coverage to spot stagnation.
7. Deterministic Sampling: Seed = `theme|commander` hash; changes to sampling heuristics must preserve deterministic output for identical inputs (regression tests rely on this). Score contributors must append reasons for transparency (`reasons[]`).

For deeper rationale & checklist see `docs/theme_taxonomy_rationale.md`.
- Preview sampling caches results (TTL 600s) keyed by (slug, limit, colors, commander, ETag) ensuring catalog changes invalidate prior samples.

Governance tooling & experiments:
- Taxonomy snapshot: `python -m code.scripts.snapshot_taxonomy` writes `logs/taxonomy_snapshots/taxonomy_<timestamp>.json` with a SHA-256 `hash` for auditability. Identical content is skipped unless forced. Use this before tuning taxonomy-aware sampling heuristics.
- Adaptive splash penalty (optional): set `SPLASH_ADAPTIVE=1` to scale off-color enabler penalties based on commander color count; tune with `SPLASH_ADAPTIVE_SCALE` (e.g., `1:1.0,2:1.0,3:1.0,4:0.6,5:0.35`). Analytics recognize both static and adaptive reasons; experiments can compare counters when enabled vs disabled.

Preview endpoint:
`GET /themes/api/theme/{id}/preview?limit=12&colors=G,W&commander=Name`
Returns a deterministic sample (seeded by theme + optional commander) combining:
- Pinned curated `example_cards` (role `example`)
- Pinned curated `synergy_example_cards` (role `curated_synergy`)
- Heuristically sampled cards bucketed into payoff / enabler / support / wildcard with diversity quotas (~40/40/20 after pinned examples)
- Synthetic `[Synergy]` placeholders only if real sampling underfills the requested limit
Commander bias heuristics (color identity restriction + overlap/theme bonuses) influence scoring & ordering.

Preview sampling now includes curated examples, curated synergy layer, role-based heuristic sampling (payoff / enabler / support / wildcard) with diversity quotas, commander bias heuristics (color identity restriction + diminishing overlap bonuses), rarity diminishing weights, splash leniency (single off-color allowance with small penalty for 4–5 color commanders), role saturation penalty, refined overlap scaling curve, and tooltip-expanded heuristic reasons. Synthetic placeholders are only added if the sample underfills.

Edge considerations (Theme picker & preview):
- Empty dataset: graceful skeleton + 'No themes match your filters' message.
- High latency first load: skeleton shimmer + optional prewarm (when `WEB_THEME_FILTER_PREWARM=1`).
- Catalog reload or tagging completion: filter + preview caches bust; stale indicator triggers background refresh.
- Commander not found: commander bias reasons omitted; sampling still deterministic via seed.
- Color filter reduces pool below quota: synthetic `[Synergy]` placeholders pad to requested limit.
- Duplicate card names across tag shards: final defensive dedupe pass before payload.
- Oversized preview cache: FIFO eviction maintains <= THEME_PREVIEW_CACHE_MAX (default 400).
- Disabled diagnostics: uncapped synergies, editorial quality, fallback description markers, raw YAML & metrics all suppressed.
- Tooltip safety: JS errors are swallowed; preview remains functional without enhancements.

#### Editorial coverage metrics / specialization progress
- Set `EDITORIAL_INCLUDE_FALLBACK_SUMMARY=1` when running `build_theme_catalog.py` to embed a `description_fallback_summary` block in the resulting JSON. This includes:
  - `generic_total`, `generic_with_synergies`, `generic_plain`, and `generic_pct`
  - `top_generic_by_frequency` (capped list of the highest-frequency themes still using generic fallback phrasing)
- Used by `test_theme_description_fallback_regression.py` to enforce a shrinking ceiling on generic descriptions after each optimization wave.
 - History persisted to `description_fallback_history.jsonl` for KPI trend visualization and automated threshold ratcheting.

#### External description mapping
- Extend or override auto-description heuristics without code edits via `config/themes/description_mapping.yml`.
  - Each list item: `triggers: ["substring1", ...]`, `description: "Text with optional {SYNERGIES} placeholder"`
  - First match wins; triggers are lowercase substring checks.
  - `{SYNERGIES}` is replaced with a short clause (e.g., `Synergies like Tokens and Sacrifice reinforce the plan.`) when synergy context exists; omitted cleanly otherwise.
- If the file is absent, the internal mapping rules are used as a fallback.

### Quick guide (locks • replace • compare • permalinks)
- New Deck modal (Steps 1–3 unified): search for your commander, pick themes, choose bracket (defaults to 3), and optionally set Name. Submitting starts the build immediately. Name becomes the export filename stem and display name.
- Lock a card: click the card image or the lock control in Step 5. Locked cards are pinned for reruns. A small “Last action” chip confirms lock/unlock.
- Replace a card: toggle Replace in Step 5, then click a card to open Alternatives. Filter (including Owned-only) and pick a swap; the deck updates in place.
  - When bracket violations exist, Step 5 shows them first and disables Continue/Rerun. Pick replacements inline; alternatives exclude commanders, locked, in-deck, and just-added cards and prefer role-consistent options.
- Permalink: use “Copy permalink” from Step 5 or a Finished deck to share/restore a run (commander, themes, bracket, ideals, flags). “Open Permalink…” lets you paste one to restore to review.
- Compare: from Finished Decks, open Compare and pick A and B (quick actions: Latest two, Swap A/B). Use Changed-only to focus diffs. Copy summary or Download .txt.
- Keyboard: In the New Deck modal, Enter selects the first commander result. Browser autofill is disabled for cleaner inputs. In Step 5, use L to lock/unlock the focused card, R to open Alternatives, and C to copy a permalink.

### Setup speed: parallel tagging (Web)
- On first run or when data is stale, the Web UI prepares and tags the card database.
- This tagging runs in parallel by default for faster setup.
- Tuning via env vars (set on the `web` service in `docker-compose.yml`):
  - `WEB_TAG_PARALLEL=1|0` — enable/disable parallel workers (default: 1/enabled)
  - `WEB_TAG_WORKERS=<N>` — number of worker processes (default: 4 in compose; omit to auto-pick)
- The UI shows progress and falls back to sequential tagging automatically if parallel init fails.

Virtualized lists and image polish (opt-in)
- Opt-in with `WEB_VIRTUALIZE=1` to enable virtualization in Step 5 lists/grids and the Owned library for smoother scrolling on large sets.
- Virtualization diagnostics overlay (for debugging): enable `SHOW_DIAGNOSTICS=1`, then press `v` to toggle an overlay per grid and a global summary bubble; shows visible range, total, rows, row height, render time, and counters.
- Thumbnails use lazy-loading with srcset/sizes; LQIP blur/fade-in is applied to Step 5 and Owned thumbnails, and the commander preview image.
- Respects reduced motion: when the OS prefers-reduced-motion, blur/fade transitions and smooth scrolling are disabled for accessibility.

### Build options (Web)
- Use only owned cards: builds strictly from your `owned_cards/` lists (commander exempt).
- Prefer owned cards: gently prioritizes owned cards across creatures and spells but allows unowned when they’re better fits.
  - Implemented via a stable owned-first reorder and small weight boosts; preserves existing sort intent.
Notes:
- Steps 1–3 are consolidated into a single New Deck modal. Submitting starts the build immediately, skipping the old review page.
 - The modal includes an optional Name; when set, it is used as the export filename stem and display name in the UI.
 - The modal displays selected themes in order (1, 2, 3). The Bracket selector shows numbers (e.g., "Bracket 3: Upgraded") and defaults to 3 for new decks.

### Staged build visibility
- Step 5 can optionally “Show skipped stages” so you can see phases that added no cards with a clear annotation.

### Multi-copy archetypes (Web)
- When your commander + themes suggest a multi-copy strategy (e.g., petitioners, approach, apostles), the Web UI shows a one-time modal to add a package.
- Choose how many copies (bounded by the printed cap) and optionally include 1× Thrumming Stone when it synergizes.
- The package is applied as the first stage so later phases account for the volume.
- Targets adjust automatically (reduce creatures for creature packages, or spread reductions across spells). The UI shows an “Adjusted targets” note on that stage.
- A final safety clamp trims overflow from this stage to keep the deck at 100. A “Clamped N” chip appears when this happens.
- Suggestions won’t re-prompt repeatedly for the same commander+theme context unless you change selections; you can also dismiss the modal.

### Visual summaries (Web)
- Mana Curve bars with hover-to-highlight of matching cards in both list and thumbnail views
- Color Pips (requirements) and Sources (generation) bars with per-color tooltips listing cards
- Cross-highlighting from charts to the card views; list-mode highlights only the name span
- Sources detection includes non-land producers (artifacts/creatures/etc.) and colorless 'C'
- Fetch lands are not counted as mana sources; basic lands and Wastes are handled reliably
- Optional: Show colorless (C) toggle for Sources; persisted per browser
- Tooltips include a Copy button to copy the card list

### Combos & Synergies (Web)
- Detection: curated two-card combos/synergies are detected from the final deck (commander included) and shown with list version badges.
- UI: chip-style rows with badges (cheap/early, setup). Hover a row to preview both cards side-by-side; hover an individual name for single-card preview.
- Auto-Complete Combos: when enabled in the New Deck modal, the builder completes up to N pairs before theme fill/monolithic spells so partners stick.
  - Settings: Prefer combos (on/off), How many combos (target), Balance (early/late/mix) to bias partner selection.
  - Existing completed pairs are counted first; only missing partners are added.
  - Color identity enforced via the filtered card pool; off-color/unavailable partners are skipped.

### Owned page (Web)
- View and manage your owned lists with:
  - Export TXT/CSV, sort controls, and a live “N shown” counter
  - Color identity dots and exact color-identity combo filters (including 4-color)
  - Viewport-filling list with styled scrollbar
  - Optional virtualization when `WEB_VIRTUALIZE=1` (improves performance on very large libraries)
- Uploading `.txt` or `.csv` lists enriches and deduplicates entries at upload-time and persists them, so page loads are fast.

### Finished Decks (Web)
- Theme filters are a dropdown with shareable state for easier browsing of saved builds.
 - Each finished deck has a permalink you can copy and revisit; Compare mode lets you diff against another run. The Compare page has a Copy summary button to copy a plain-text diff.
 - Locks and Replace flow: lock-in picks you like or replace a card with an alternative during iteration.

### Diagnostics (Web)
- Health endpoint: `GET /healthz` returns `{ status, version, uptime_seconds }`
- Responses include `X-Request-ID` for easier error correlation; unhandled errors return JSON with `request_id`

Logs and system tools:
- Logs page (`/logs`, enable with `SHOW_LOGS=1`):
  - Auto-refresh toggle with adjustable interval
  - Level filter (All/Error/Warning/Info/Debug) and keyword filter
  - Copy button to copy the visible log tail
- System summary (`GET /status/sys`, shown on `/diagnostics` when `SHOW_DIAGNOSTICS=1`):
  - Returns `{ version, uptime_seconds, server_time_utc, flags }`
  - Shows resolved theme and stored preference; includes a Reset preference button.

Compose: enable diagnostics/logs (optional)

```yaml
services:
  web:
    environment:
      - SHOW_LOGS=1
      - SHOW_DIAGNOSTICS=1
      # Random Modes (optional; alpha)
      - RANDOM_MODES=1
      - RANDOM_UI=1
      - RANDOM_MAX_ATTEMPTS=5
      - RANDOM_TIMEOUT_MS=5000
  # RANDOM_BUILD_SUPPRESS_INITIAL_EXPORT:  # (Default behavior auto-enables suppression; set to 0 to debug legacy double-export)
```

### Budget Mode and price/legal status
- Price and legality snippet integration is deferred to a dedicated Budget Mode initiative. This will centralize price sourcing, caching, legality checks, and UI surfaces. Track progress in `logs/roadmaps/roadmap_9_budget_mode.md`.

## 🤖 Headless mode (no prompts)

- Auto-headless: set `DECK_MODE=headless`
  - Example (PowerShell):
    ```powershell
    docker compose run --rm -e DECK_MODE=headless mtg-deckbuilder
    ```
- Use a JSON config: mount `./config` to `/app/config` and set `DECK_CONFIG=/app/config/deck.json`
  - Example (PowerShell):
    ```powershell
    docker compose run --rm `
      -e DECK_MODE=headless `
      -e DECK_CONFIG=/app/config/deck.json `
      mtg-deckbuilder
    ```
- Override via env vars (subset): `DECK_COMMANDER`, `DECK_PRIMARY_CHOICE`, `DECK_SECONDARY_CHOICE`, `DECK_TERTIARY_CHOICE`, `DECK_ADD_LANDS`, `DECK_FETCH_COUNT`, `DECK_BRACKET_LEVEL`
  - Precedence: CLI > env > JSON > defaults

### Enhanced CLI with Type Safety & Theme Names

The headless runner now features comprehensive CLI arguments with type indicators and intelligent theme selection:

```powershell
# Show all available options with type information
python code/headless_runner.py --help

# Build with theme names instead of index numbers
python code/headless_runner.py `
  --commander "Aang, Airbending Master" `
  --primary-tag "Airbending" `
  --secondary-tag "Exile Matters" `
  --bracket-level 4

# Override ideal deck composition counts
```

## 🧪 Theme Catalog Validation & Schema

The theme system now uses a merged catalog plus per-theme YAML files.

Validator usage:

```powershell
python code/scripts/validate_theme_catalog.py                # Standard validation
python code/scripts/validate_theme_catalog.py --rebuild-pass # Ensures idempotent rebuild
python code/scripts/validate_theme_catalog.py --schema       # Catalog JSON Schema
python code/scripts/validate_theme_catalog.py --yaml-schema  # Per-file YAML JSON Schema
python code/scripts/validate_theme_catalog.py --strict-alias # Fail if alias display_name still present
```

CI runs the non-strict validator every push. A second step runs `--strict-alias` as an allowed failure until legacy alias YAMLs (e.g., `reanimator.yml`) are removed, after which the step will be enforced.

Per-file YAML structure (excerpt):

```yaml
id: plus1-plus1-counters
display_name: +1/+1 Counters
curated_synergies: [Proliferate, Counters Matter]
enforced_synergies: []
inferred_synergies: []
synergies: [Proliferate, Counters Matter, Adapt, Evolve]
primary_color: Green
secondary_color: White
notes: ''
```

Add or edit a theme by creating/updating a file in `config/themes/catalog/` and running the build script:

```powershell
python code/scripts/build_theme_catalog.py
```

Metadata info (formerly provenance) is written to `theme_list.json` and consumed by tests and future UI diagnostics. During the deprecation period the builder accepts either key; prefer `metadata_info` in all new tooling / docs.

### Duplicate suppression & example_cards quality controls

Example card generation (and rebuild / padding via `synergy_promote_fill.py`) includes safeguards to reduce noisy staples and over-represented picks:

- `--common-card-threshold <fraction>`: Excludes candidate example cards that already appear in more than this fraction of themes (default 0.18 = 18%). This curbs ubiquitous staples (e.g., Sol Ring) from crowding lists.
- `--print-dup-metrics`: After a run (dry-run or apply) prints global frequency distribution so thresholds can be tuned.

Typical loop (dry run with metrics):

```powershell
python code/scripts/synergy_promote_fill.py --fill-example-cards --common-card-threshold 0.18 --print-dup-metrics
```

Adjustment guidance:
- If many mid-frequency utility cards still leak through: lower the threshold (e.g., 0.15).
- If high-value, legitimately thematic cards are being suppressed prematurely: raise slightly (e.g., 0.22) and re-run with metrics to confirm net effect.

These controls complement existing staple suppression heuristics (`staples_block` list) and color-fallback restraint flags.

## 📝 Theme Editorial Enrichment

Phase D introduces automated editorial helpers to surface representative cards and synergy-informed commander suggestions per theme while preserving manual curation:

### Key scripts
- `code/scripts/generate_theme_editorial_suggestions.py`: Scans card & commander CSVs to propose `example_cards`, `example_commanders`, and a derived `synergy_commanders` list.
  - Synergy commanders: up to 3 / 2 / 1 picks from the theme's top three synergies (3/2/1 pattern), falling back to legendary card hits if a synergy lacks direct commander-tagged entries.
  - Promotion: ensures a minimum (default 5) example commanders by promoting synergy picks with annotations.
  - Annotation format: `Name - Synergy (Synergy Theme)` for both promoted examples and residual synergy-only entries.
  - Duplicate filtering: commanders already present (base name) in `example_commanders` are omitted from `synergy_commanders` to avoid redundancy.
  - Augmentation (`--augment-synergies`): heuristically pads sparse `synergies` arrays (e.g., injects `Counters Matter`, `Proliferate` for counter variants) before deriving synergy picks.
- `code/scripts/lint_theme_editorial.py`: Validates editorial consistency (annotation correctness, min counts, deduplication, size bounds) and emits warnings (non-fatal by default).

### Usage examples
```powershell
# Dry run (first 25 themes)
python code/scripts/generate_theme_editorial_suggestions.py

# Apply to all themes with augmentation and minimum example commanders of 5
python code/scripts/generate_theme_editorial_suggestions.py --apply --augment-synergies --min-examples 5

# Lint results
python code/scripts/lint_theme_editorial.py
```

### Non-determinism & source control guidance
The editorial output depends on current CSV card data, ranks, and tagging heuristics. Regeneration on a different machine or after data refresh may yield a different ordering or composition. For that reason:
- Commit the tooling and schema changes.
- Prefer selective or manually reviewed YAML edits (especially cornerstone themes) instead of bulk committing the entire regenerated catalog.
- Treat full-catalog regeneration as an operational task (e.g., during a release preparation) rather than every feature branch.

### Schema fields
Per-theme YAML now may include:
```yaml
example_cards: [ ... ]              # Representative non-commander cards
example_commanders: [ ... ]          # Curated + promoted synergy commanders (annotated)
synergy_commanders: [ ... ]          # Remaining annotated synergy picks not already promoted
```

### Lint expectations (default)
- example_commanders: <=12, aim for ≥5 (warnings below minimum)
- example_cards: ≤20
- synergy_commanders: ≤6 (3/2/1 pattern) after filtering duplicates
- Annotations must reference a synergy present in `synergies`.

Cornerstone themes (Landfall, Reanimate, Superfriends, Tokens Matter, +1/+1 Counters) should always have curated coverage; automated promotion fills gaps but manual review is encouraged.
python code/headless_runner.py `
  --commander "Krenko, Mob Boss" `
  --primary-tag "Goblin Kindred" `
  --creature-count 35 `
  --land-count 33 `
  --ramp-count 12

# Include/exclude specific cards (semicolon for comma-containing names)
python code/headless_runner.py `
  --commander "Jace, Vryn's Prodigy" `
  --include-cards "Sol Ring;Jace, the Mind Sculptor" `
  --exclude-cards "Chaos Orb;Shahrazad" `
  --enforcement-mode strict
```

**New CLI Features:**
- **Type-safe help text**: All arguments show expected types (PATH, NAME, INT, BOOL) 
- **Ideal count arguments**: `--ramp-count`, `--land-count`, `--creature-count`, etc.
- **Theme tag names**: `--primary-tag "Theme Name"` instead of `--primary-choice 1`
- **Include/exclude CLI**: `--include-cards`, `--exclude-cards` with semicolon support
- **Console diagnostics**: Detailed summary output for validation and results

Headless submenu notes:
- If one JSON exists in `config/`, it auto-runs it
- If multiple exist, they’re listed as "Commander - Theme1, Theme2, Theme3"; `deck.json` shows as "Default"
- CSV/TXT are exported as usual; JSON run-config is exported only in interactive runs

### Run locally (no Docker)
```powershell
# Show resolved settings (no run)
python code/headless_runner.py --config config/deck.json --dry-run

# Run with a specific config file
python code/headless_runner.py --config config/deck.json

# Point to a folder; if exactly one config exists, it's auto-used
python code/headless_runner.py --config config

# Override via CLI
python code/headless_runner.py --commander "Pantlaza, Sun-Favored" --primary-choice 2 --secondary-choice 0 --add-lands true --fetch-count 3
```

### CLI flag reference

Each flag below mirrors an environment variable (env vars override JSON, CLI overrides both). Types match `--help` output.

#### Core selection & flow

| Flag | Type | Description | JSON key / Env var |
| --- | --- | --- | --- |
| `--config PATH` | string | Path to JSON config file or directory (auto-detects a single file) | `DECK_CONFIG` |
| `--commander NAME` | string | Commander search term | `commander` / `DECK_COMMANDER` |
| `--primary-choice INT` | int | Primary theme menu index (1-based) | `primary_choice` / `DECK_PRIMARY_CHOICE` |
| `--secondary-choice INT` | optional int | Secondary theme index | `secondary_choice` / `DECK_SECONDARY_CHOICE` |
| `--tertiary-choice INT` | optional int | Tertiary theme index | `tertiary_choice` / `DECK_TERTIARY_CHOICE` |
| `--primary-tag NAME` | string | Primary theme name (auto-maps to index) | `primary_tag` / `DECK_PRIMARY_TAG` |
| `--secondary-tag NAME` | string | Secondary theme name | `secondary_tag` / `DECK_SECONDARY_TAG` |
| `--tertiary-tag NAME` | string | Tertiary theme name | `tertiary_tag` / `DECK_TERTIARY_TAG` |
| `--bracket-level 1-5` | int | Power bracket selection | `bracket_level` / `DECK_BRACKET_LEVEL` |
| `--dry-run` | flag | Print the resolved configuration and exit | — |

#### Deck composition & counts

| Flag | Type | Description | JSON key / Env var |
| --- | --- | --- | --- |
| `--ramp-count INT` | int | Target ramp spells | `ideal_counts.ramp` |
| `--land-count INT` | int | Total land target | `ideal_counts.lands` |
| `--basic-land-count INT` | int | Minimum basic lands | `ideal_counts.basic_lands` |
| `--creature-count INT` | int | Creature count target | `ideal_counts.creatures` |
| `--removal-count INT` | int | Spot removal target | `ideal_counts.removal` |
| `--wipe-count INT` | int | Board wipe target | `ideal_counts.wipes` |
| `--card-advantage-count INT` | int | Card advantage target | `ideal_counts.card_advantage` |
| `--protection-count INT` | int | Protection spell target | `ideal_counts.protection` |
| `--fetch-count INT` | optional int | Requested fetch land count | `fetch_count` / `DECK_FETCH_COUNT` |
| `--dual-count INT` | optional int | Dual land request | `dual_count` / `DECK_DUAL_COUNT` |
| `--triple-count INT` | optional int | Three-color land request | `triple_count` / `DECK_TRIPLE_COUNT` |
| `--utility-count INT` | optional int | Utility land request | `utility_count` / `DECK_UTILITY_COUNT` |

#### Card type toggles

| Flag | Type | Description | JSON key / Env var |
| --- | --- | --- | --- |
| `--add-lands BOOL` | bool | Run the land builder sequence | `add_lands` / `DECK_ADD_LANDS` |
| `--add-creatures BOOL` | bool | Add creatures | `add_creatures` / `DECK_ADD_CREATURES` |
| `--add-non-creature-spells BOOL` | bool | Bulk add non-creature spells (if available) | `add_non_creature_spells` / `DECK_ADD_NON_CREATURE_SPELLS` |
| `--add-ramp BOOL` | bool | Add ramp spells | `add_ramp` / `DECK_ADD_RAMP` |
| `--add-removal BOOL` | bool | Add removal spells | `add_removal` / `DECK_ADD_REMOVAL` |
| `--add-wipes BOOL` | bool | Add board wipes | `add_wipes` / `DECK_ADD_WIPES` |
| `--add-card-advantage BOOL` | bool | Add card draw | `add_card_advantage` / `DECK_ADD_CARD_ADVANTAGE` |
| `--add-protection BOOL` | bool | Add the protection suite | `add_protection` / `DECK_ADD_PROTECTION` |

#### Include / exclude controls

| Flag | Type | Description | JSON key / Env var |
| --- | --- | --- | --- |
| `--include-cards LIST` | string | Force-include cards (comma or semicolon separated) | `include_cards` |
| `--exclude-cards LIST` | string | Exclude cards | `exclude_cards` |
| `--enforcement-mode warn|strict` | string | Handling when includes cannot be satisfied | `enforcement_mode` |
| `--allow-illegal BOOL` | bool | Permit non-legal cards in include/exclude lists | `allow_illegal` |
| `--fuzzy-matching BOOL` | bool | Enable fuzzy card name matching | `fuzzy_matching` |

#### Random mode (web parity)

| Flag | Type | Description | Env var / JSON key |
| --- | --- | --- | --- |
| `--random-mode` | flag | Force the headless random builder path | `HEADLESS_RANDOM_MODE` / `random_mode` |
| `--random-theme NAME` | string | Legacy single-theme alias (maps to primary) | `RANDOM_THEME` / `random.theme` |
| `--random-primary-theme NAME` | string | Primary theme slug | `RANDOM_PRIMARY_THEME` / `random.primary_theme` |
| `--random-secondary-theme NAME` | string | Secondary theme slug | `RANDOM_SECONDARY_THEME` / `random.secondary_theme` |
| `--random-tertiary-theme NAME` | string | Tertiary theme slug | `RANDOM_TERTIARY_THEME` / `random.tertiary_theme` |
| `--random-auto-fill BOOL` | bool | Auto-fill missing theme slots | `RANDOM_AUTO_FILL` / `random.auto_fill` |
| `--random-auto-fill-secondary BOOL` | bool | Override secondary auto-fill | `RANDOM_AUTO_FILL_SECONDARY` / `random.auto_fill_secondary` |
| `--random-auto-fill-tertiary BOOL` | bool | Override tertiary auto-fill | `RANDOM_AUTO_FILL_TERTIARY` / `random.auto_fill_tertiary` |
| `--random-strict-theme-match BOOL` | bool | Require exact theme matches when selecting commanders | `RANDOM_STRICT_THEME_MATCH` / `random.strict_theme_match` |
| `--random-attempts INT` | int | Retry attempts before giving up | `RANDOM_MAX_ATTEMPTS` / `random.attempts` |
| `--random-timeout-ms INT` | int | Timeout per attempt (milliseconds) | `RANDOM_TIMEOUT_MS` / `random.timeout_ms` |
| `--random-seed VALUE` | int or string | Deterministic seed for reproducible runs | `RANDOM_SEED` / `random.seed` |
| `--random-constraints JSON_OR_PATH` | string | Inline JSON or path to constraint file | `RANDOM_CONSTRAINTS` / `RANDOM_CONSTRAINTS_PATH` / `random.constraints` |
| `--random-output-json PATH` | string | Write random build payload to file or directory | `RANDOM_OUTPUT_JSON` / `random.output_json` |

Booleans accept: 1/0, true/false, yes/no, on/off.

### Random mode parity (web → headless)

The headless runner now shares the web UI's random builder pipeline. Set `--random-mode` (or `HEADLESS_RANDOM_MODE=1`) to route through the Surprise/Reroll flow with multi-theme support, auto-fill assistance, constraints, and deterministic seeds. Combine flags as needed:

```powershell
python code/headless_runner.py `
  --random-mode `
  --random-primary-theme Artifacts `
  --random-secondary-theme Tokens `
  --random-auto-fill true `
  --random-output-json deck_files/random/latest.json
```

The build prints a full summary (commander, seed, theme stack, fallback reasons, attempts vs timeout, auto-fill status) and writes the optional JSON payload when `--random-output-json`/`RANDOM_OUTPUT_JSON` is provided. Use `--random-constraints` (or `RANDOM_CONSTRAINTS[_PATH]`) to supply pool limits, and `--random-seed` for reproducible reruns. Pair any random flag with `--dry-run` to inspect the resolved configuration without building.

### JSON export in headless
- By default, headless runs do not export a JSON run-config to avoid duplicates.
- Opt-in with:
  ```powershell
  $env:HEADLESS_EXPORT_JSON = "1"
  python code/headless_runner.py --config config/deck.json
  ```
- Tip: when opting in, prefer using `--config` instead of a `DECK_CONFIG` file path to avoid creating both a stem-based JSON and a second explicit-path JSON.

Example JSON (`config/deck.json`):
```json
{
  "commander": "Pantlaza",
  "bracket_level": 3,
  "primary_choice": 2,
  "secondary_choice": 2,
  "tertiary_choice": 2,
  "tag_mode": "OR", // OR or AND; Web UI default is OR
  "add_lands": true,
  "fetch_count": 3,
  "ideal_counts": { "ramp": 10, "lands": 36, "basic_lands": 16, "creatures": 28, "removal": 8, "wipes": 3, "card_advantage": 8, "protection": 3 }
}
```
Notes: headless honors `ideal_counts` (leave prompts blank to accept). Only `fetch_count` is tracked/exported for lands.

#### JSON fields for Combos (Web Configs)
When running from a JSON config in the Web UI, the following fields are supported and exported from interactive runs:
```json
{
  "prefer_combos": true,
  "combo_target_count": 3,
  "combo_balance": "mix"
}
```
Auto-Complete Combos runs before theme fill/monolithic spells when `prefer_combos` is true.

## 🔒 Bracket policy & enforcement

- Policy source: `config/brackets.yml` defines per-bracket limits for categories like `game_changers`, `extra_turns`, `mass_land_denial`, `tutors_nonland`, and `two_card_combos`.
- Authoritative lists: JSON under `config/card_lists/` provides names for enforcement and reporting (`game_changers.json`, `extra_turns.json`, `mass_land_denial.json`, `tutors_nonland.json`). A `list_version` may be included for badges.
- Global safety prune: when a category has a limit of 0, the builder removes matching cards from the card pool up-front so they cannot be selected (Game Changers by name; others by tags when present). This runs in both Web and headless builds.
- Preemptive filters: spells and creatures phases also apply bracket-aware pre-filters.
- Inline enforcement (Web): if violations still occur, Step 5 shows them before the summary. You must replace or remove flagged cards before you can Continue or Rerun. Alternatives are role-consistent, exclude the replaced/in-deck/locked/commander cards, and bias toward owned when enabled.
- Game Changer fallback: if `config/card_lists/game_changers.json` is empty, enforcement and reporting fall back to the in-code `builder_constants.GAME_CHANGERS` list so low-bracket decks still exclude those cards.
- Auto-enforce (optional): set `WEB_AUTO_ENFORCE=1` to automatically apply the enforcement plan after build and re-export CSV/TXT when violations are detected.

Status levels
- PASS: within limits; panel remains collapsed by default.
- WARN: advisory thresholds met (from `<category>_warn` keys in `config/brackets.yml` or conservative defaults for low brackets on tutors/extra turns). Panel opens automatically and shows WARN tiles with a subtle amber border and badge; no gating or automatic enforcement.
- FAIL: over hard limits or disallowed categories for the selected bracket. Panel opens automatically, tiles show with red accents, and Continue/Rerun are disabled until resolved; enforcement actions are available.

Compliance report
- Every run writes `[stem]_compliance.json` next to your deck exports, including per-category counts/limits, status (PASS/FAIL), detected cheap/early two-card combos, and list versions.
- Headless JSON builds and Web builds use the same engine; summary and exports are consistent.

## 🧩 Theme combine mode (AND/OR)

- OR (default): Recommend cards that match any selected themes, prioritizing overlap.
- AND: Prioritize multi-theme intersections. For creatures, an AND pre-pass first picks "all selected themes" creatures up to a cap, then fills by weighted overlap.

UI tips:
- Step 2 includes AND/OR radios with a tooltip explaining trade-offs.
- In staged build view, "Creatures: All-Theme" shows which selected themes each card hits.

## 🕹️ Usage (interactive)
1) Start the app (Docker or from source)
2) Pick Build a New Deck
3) Search/confirm commander
4) Pick primary/secondary/tertiary themes (or stop at primary); choose AND/OR combine mode
5) Choose power bracket and review ideal counts
6) Deck builds; CSV/TXT export to `deck_files/`

## ⚙️ Environment variables (common)
- DECK_MODE=headless
- DECK_CONFIG=/app/config/deck.json
- DECK_COMMANDER, DECK_PRIMARY_CHOICE, DECK_SECONDARY_CHOICE, DECK_TERTIARY_CHOICE
- DECK_ADD_LANDS, DECK_FETCH_COUNT
- DECK_ADD_CREATURES, DECK_ADD_NON_CREATURE_SPELLS, DECK_ADD_RAMP, DECK_ADD_REMOVAL, DECK_ADD_WIPES, DECK_ADD_CARD_ADVANTAGE, DECK_ADD_PROTECTION
- DECK_BRACKET_LEVEL
- **ALLOW_MUST_HAVES=true** (enables include/exclude card lists feature with enhanced UI validation)
- Random Modes (alpha): core toggles `RANDOM_MODES=1`, `RANDOM_UI=1`, `RANDOM_MAX_ATTEMPTS=5`, `RANDOM_TIMEOUT_MS=5000` (see detailed list below for theme/auto-fill/seed overrides)
  - `RANDOM_BUILD_SUPPRESS_INITIAL_EXPORT=0` — (optional) disable the default suppression of builder auto-export (will produce legacy duplicate artifacts; for debugging only)
 - Testing/data overrides:
   - `CSV_FILES_DIR=csv_files/testdata` — point the app/tests at a frozen dataset snapshot for determinism

Random Modes (alpha flags):
- `RANDOM_MODES=1` — enable random build endpoints and backend features
- `RANDOM_UI=1` — show Surprise/Theme/Reroll/Share controls in UI
- `RANDOM_MAX_ATTEMPTS=5` — cap retry attempts for random generation
- `RANDOM_TIMEOUT_MS=5000` — per-build timeout in milliseconds
- `RANDOM_THEME=`*slug* — legacy single-theme alias (maps to primary when others unset)
- `RANDOM_PRIMARY_THEME=`*slug* — primary theme override
- `RANDOM_SECONDARY_THEME=`*slug* — secondary theme override
- `RANDOM_TERTIARY_THEME=`*slug* — tertiary theme override
- `RANDOM_AUTO_FILL=0|1` — auto-fill missing secondary/tertiary slots when enabled
- `RANDOM_AUTO_FILL_SECONDARY=0|1` / `RANDOM_AUTO_FILL_TERTIARY=0|1` — explicit per-slot overrides
- `RANDOM_STRICT_THEME_MATCH=0|1` — require exact theme matches when rolling commanders
- `RANDOM_CONSTRAINTS=`*inline JSON* and/or `RANDOM_CONSTRAINTS_PATH=/app/config/random_constraints.json`
- `RANDOM_SEED=`*value* — deterministic run seed (int or string)
- `RANDOM_OUTPUT_JSON=/app/deck_files/random_build.json` — write random build payload to file/dir
- `HEADLESS_RANDOM_MODE=1` — force the headless runner to take the random pipeline
- `RANDOM_BUILD_SUPPRESS_INITIAL_EXPORT=0` — (optional) disable the default suppression of builder auto-export (useful for debugging legacy duplicate exports)

Testing/determinism helpers:
- CSV_FILES_DIR=csv_files/testdata — override CSV base dir for tests or pinned snapshots

Optional name-based tag overrides (mapped to indices for the chosen commander):
- DECK_PRIMARY_TAG, DECK_SECONDARY_TAG, DECK_TERTIARY_TAG

Combine mode in headless:
- JSON: set `"tag_mode": "AND" | "OR"`
- Env var: `DECK_TAG_MODE=AND|OR` (if configured in your environment)

Web UI performance tuning:
- WEB_TAG_PARALLEL=1|0
- WEB_TAG_WORKERS=<N>
 - WEB_VIRTUALIZE=1 (enable list virtualization)
 - SHOW_DIAGNOSTICS=1 (optional: diagnostics tools and virtualization overlay toggle using `v`)
 - WEB_AUTO_ENFORCE=1 (optional; after building, auto-apply enforcement and re-export when the compliance report fails)
 - SPLASH_ADAPTIVE=1 (optional: enable adaptive off-color penalty by commander color count)
 - SPLASH_ADAPTIVE_SCALE="1:1.0,2:1.0,3:1.0,4:0.6,5:0.35" (optional: tuning for adaptive scaling)

Paths and data overrides:
- `DECK_EXPORTS=/app/deck_files` — where finished deck exports are read by the Web UI
- `OWNED_CARDS_DIR=/app/owned_cards` — preferred directory for owned inventory uploads
- `CARD_LIBRARY_DIR=/app/owned_cards` — back‑compat alias for OWNED_CARDS_DIR
- `DECK_CONFIG=/app/config/deck.json` — default headless config path
- `CSV_FILES_DIR=/app/csv_files` — override CSV base dir (tests, snapshots, or alternate datasets)
- `CARD_INDEX_EXTRA_CSV=path/to/synthetic_cards.csv` — inject an extra CSV into the card index for experiments/tests

Theme catalog scan & prewarm:
- `THEME_CATALOG_YAML_SCAN_INTERVAL_SEC=2.0` — poll for YAML changes (dev); omit in production
- `WEB_THEME_FILTER_PREWARM=1` — prewarm common filters for faster first renders

Theme preview cache & Redis (optional):
- `THEME_PREVIEW_CACHE_MAX=400` — max previews cached in memory
- `WEB_THEME_PREVIEW_LOG=0` — 1=verbose cache logs
- `THEME_PREVIEW_ADAPTIVE=0` — 1=adaptive cache policy
- `THEME_PREVIEW_EVICT_COST_THRESHOLDS=5,15,40` — cost tiers for eviction policy
- `THEME_PREVIEW_BG_REFRESH=0` — 1=background refresh worker; `THEME_PREVIEW_BG_REFRESH_INTERVAL=120` seconds
- TTL tuning: `THEME_PREVIEW_TTL_BASE=300`, `THEME_PREVIEW_TTL_MIN=60`, `THEME_PREVIEW_TTL_MAX=900`, `THEME_PREVIEW_TTL_BANDS=0.2,0.5,0.8`, `THEME_PREVIEW_TTL_STEPS=2,4,2,3,1`
- Redis: `THEME_PREVIEW_REDIS_URL=redis://localhost:6379/0`, `THEME_PREVIEW_REDIS_DISABLE=0` (1=disable redis even if URL set)

Rarity weighting and diversity (advanced):
- `RARITY_W_MYTHIC=1.2`, `RARITY_W_RARE=0.9`, `RARITY_W_UNCOMMON=0.65`, `RARITY_W_COMMON=0.4` — diminish or boost rarity selection
- `RARITY_DIVERSITY_TARGETS="mythic:0-1,rare:0-2,uncommon:0-4,common:0-6"` — soft targets per rarity band
- `RARITY_DIVERSITY_OVER_PENALTY=-0.5` — penalty when exceeding targets

Misc utility land tuning (Step 7):
- MISC_LAND_DEBUG=1                 Write debug CSVs for misc land step (candidates/post-filter). Otherwise suppressed unless SHOW_DIAGNOSTICS=1.
- MISC_LAND_EDHREC_KEEP_PERCENT_MIN=0.75  Lower bound of dynamic EDHREC keep range (0–1). When MIN & MAX are both set, a random % in [MIN,MAX] is rolled per build.
- MISC_LAND_EDHREC_KEEP_PERCENT_MAX=1.0   Upper bound of dynamic EDHREC keep range (0–1).
- MISC_LAND_EDHREC_KEEP_PERCENT=0.80      Legacy fixed keep % used only if MIN/MAX not both present.
- MISC_LAND_THEME_MATCH_BASE=1.4          Base multiplier when at least one selected theme tag matches a candidate land.
- MISC_LAND_THEME_MATCH_PER_EXTRA=0.15    Incremental multiplier per additional matching theme tag beyond the first.
- MISC_LAND_THEME_MATCH_CAP=2.0           Cap for total theme multiplier after stacking base + extras.

Notes:
- Fetch lands are fully excluded from the misc step (handled earlier).
- Mono-color decks auto-filter broad rainbow/any-color lands except an explicit always-keep list.
- Land Alternatives endpoint: for land seeds, returns land-only suggestions; 12 random picks each request from a randomly sized window within the top 60–100 ranked candidates (per-card, uncached) for variety.

## 📁 Folders
- `deck_files/` — CSV/TXT exports
- `csv_files/` — card data
- `logs/` — logs
- `config/` — JSON configs (optional)
- `owned_cards/` — your owned cards lists (`.txt`/`.csv`); used for owned-only builds and Owned flagging

## 🧰 Troubleshooting
- Use `docker compose run --rm` (not `up`) for interactive sessions
- Ensure volumes are mounted so files persist (`deck_files`, `logs`, `csv_files`)
- For headless with config, mount `./config:/app/config` and set `DECK_CONFIG`
- Card data refresh: if `csv_files/cards.csv` is missing or older than 7 days, the app refreshes data and re-tags automatically. A `.tagging_complete.json` file in `csv_files/` indicates tagging completion.
 - Web health: the header dot polls `/healthz` — green is OK, red is degraded. If red, check logs.
 - Error details: HTMX errors show a toast with a “Copy details” button including X-Request-ID; include that when filing issues.
 - Logs page: set `SHOW_LOGS=1` to enable `/logs` and `/status/logs?tail=200` (read‑only) for quick diagnostics.
  - Diagnostics page: set `SHOW_DIAGNOSTICS=1` to enable the nav link and `/diagnostics` test tools.

Data integrity notes:
- Banned cards: per-color/guild CSVs now consistently respect the Commander banned list using exact, case‑insensitive name matching across `name` and `faceName`.

## 🧪 Development and tests

Set up a virtual environment, install dependencies, and run the test suite.

```powershell
# From repo root (PowerShell)
python -m venv .venv
& ".venv/Scripts/Activate.ps1"
pip install -r requirements.txt
pip install -r requirements-dev.txt

# Run tests (pytest is configured to look under code/tests)
python -m pytest -q
```

Notes:
- Test discovery is set to `code/tests` in `pytest.ini`.
- `pytest.ini` disables the built-in debugging plugin to avoid a stdlib module name clash with the project folder `code/`.
- Feature flags for local diagnostics: set `SHOW_DIAGNOSTICS=1` and/or `SHOW_LOGS=1` to expose `/diagnostics` and `/logs`.
 - Index testing helper: set `CARD_INDEX_EXTRA_CSV=path\to\synthetic_cards.csv` to inject a temporary CSV during tests (e.g., color identity edge cases) without altering production shards.

### What’s new (quick summary)
- Faster browsing with optional virtualized grids/lists in Step 5 and Owned (`WEB_VIRTUALIZE=1`).
- Image polish: lazy-loading, responsive `srcset/sizes`, and LQIP blur/fade for Step 5, Owned, and commander preview.
- Diagnostics overlay (opt-in with `SHOW_DIAGNOSTICS=1`): press `v` to see visible range, totals, render time, and counters.
- Accessibility: respects reduced-motion (disables blur/fade and smooth scrolling).
- Small caching wins: short‑TTL fragment caching for summary partials and suggestions.

## 📦 Releases
- Release notes are maintained in `RELEASE_NOTES_TEMPLATE.md`. Automated workflows read from this file to populate Docker Hub and GitHub releases.

### Collecting diagnostics for an issue
- Note the Request-ID from the toast or error page.
- Copy logs from `/logs` (enable with `SHOW_LOGS=1`) or `/status/logs?tail=200`.
- Include your `/healthz` JSON and environment flags (SHOW_LOGS/SHOW_DIAGNOSTICS/SHOW_SETUP) when reporting.

## Performance Baselines

A warm preview performance baseline is committed at `logs/perf/theme_preview_warm_baseline.json`.

Tooling:
- Benchmark: `python -m code.scripts.preview_perf_benchmark --all --passes 2 --extract-warm-baseline logs/perf/theme_preview_warm_baseline.json`
- Compare (manual): `python -m code.scripts.preview_perf_compare --baseline logs/perf/theme_preview_warm_baseline.json --candidate logs/perf/new_run.json --warm-only --p95-threshold 5`
- Regression helper: `python -m code.scripts.preview_perf_ci_check --baseline logs/perf/theme_preview_warm_baseline.json --p95-threshold 5` (run locally or in ad-hoc CI when desired)

Policy:
- Treat warm-only p95 regressions beyond 5% as blockers even though enforcement is now manual.
- Update the baseline only after intentional performance improvements or systemic environment shifts.

Workflow to refresh baseline:
1. Ensure no active regressions (run wrapper locally, expect pass).
2. Run a multi-pass benchmark with at least two passes over all themes.
3. Replace `theme_preview_warm_baseline.json` with new warm pass output; commit in same PR with rationale in CHANGELOG.

Optional future enhancements:
- Reintroduce automated gating once preview perf stabilizes on shared runners.
- Separate p50 threshold (e.g., >7% regression) for early warning without failing build.