Gen Con Scheduler

Cast your wishlist

Paste your profile YAML below and submit. The scheduler ranks every Gen Con event against your interests, builds an optimal personal schedule, composes a 50-row wishlist, and runs simulated annealing to refine the priority order.

Total runtime is roughly 30–90 seconds with live progress. Need a starting point? The textarea is pre-populated with the template profile — edit the name, days, and preferences to match your party. Profile YAML is limited to 65,536 bytes.

# =====================================================================
# Gen Con 2026 — Preference Profile Template
# =====================================================================
#
# This file describes WHO you are and WHAT you want out of Gen Con. The
# scheduler reads it and produces:
#   1. A ranked list of every event in the catalog (`ranked.csv`)
#   2. An optimal personal schedule  (`schedule.json` / `schedule.md`)
#   3. A 50-event Wish List with backups (`wishlist.csv` / `wishlist_paste.txt`)
#   4. An ICS calendar feed              (`schedule.ics`)
#   5. A decision report explaining the picks (`decision_report.md`)
#
# How to use this file
# --------------------
#  1. Copy: `cp profiles/_template.yaml profiles/<your_name>.yaml`
#  2. Fill in your `name`, `max_budget`, `baseline_venue`, days, and
#     preference lists. Most of the file is already calibrated to a
#     reasonable default — focus on the sections marked with `>>> EDIT:`.
#  3. Validate: `make lint-profile P=<your_name>`
#     (catches typos, missing fields, contradictions, dead keywords)
#  4. Generate: `make schedule P=<your_name>`
#  5. Read the decision report and tweak. Common iterations:
#       - too few good picks  → broaden `strong_positive_preferences`
#       - too many off-tone   → add to `negative_preferences`
#       - one system dominates → use `preferred_systems_ranked` instead
#         of the flat list, and weight your favorites accordingly
#       - want a specific kind of slot guaranteed → add a
#         `must_include_categories` entry (see worked example below)
#
# Mental model: the scheduler runs in three layers.
#   * HARD CONSTRAINTS  — filter (an event either qualifies or it doesn't)
#   * SOFT PREFERENCES  — score   (events compete on a 0..3-ish scale)
#   * WISHLIST COMPOSER — guarantee (reserved slots even if score didn't win)
#
# Anything in this file with `# >>> EDIT:` is something you should
# customize. Everything else has a sensible default that works without
# tuning. See `profiles/alex.yaml` for a complete worked example.
# =====================================================================

# >>> EDIT: short slug used in CLI commands and output paths.
# `make schedule P=name` -> outputs/<name>/...
name: your_name_here

# >>> EDIT: hard ceiling on total event ticket cost across the weekend
# (USD). Generic events are usually $2-6; premium events (True Dungeon,
# Mothership megagames) can be $30-80. A single weekend usually fits
# in $200-400 even with a couple of premium picks.
max_budget: 250

# >>> EDIT: where you start your day (your hotel, or ICC if commuting).
# Walking-distance penalties are measured from here when no prior
# event sets the location. Use one of the canonical venue names from
# `data/walking_minutes.csv`:
#   ICC, JW, Marriott, Hyatt, Westin, Hilton, Omni, Crowne Plaza,
#   Stadium, Embassy Suites, Le Meridien, Taxman CityWay, Union Station,
#   St. Elmo, 416 Wabash.
# Pick the closest match if you're staying somewhere not on the list.
baseline_venue: "ICC"

# =====================================================================
# HARD CONSTRAINTS — events that violate any of these are pre-filtered
# out before scoring. Tighten only if you really mean it; an over-tight
# constraint produces an empty schedule.
# =====================================================================
hard_constraints:

# >>> EDIT: which days you'll attend. Drop any you're skipping.
  # The con runs Thursday Jul 30 - Sunday Aug 2, 2026.
  days: [Thursday, Friday, Saturday, Sunday]

# Earliest you'll start an event (24h "HH:MM"). Most attendees set
  # this to 09:00 or 10:00; set 08:00 only if you're a morning person
  # who's already at the ICC by then.
  earliest_start: "10:00"

# Latest end time. Three knobs because Gen Con's days are uneven:
  #   - Thursday/Friday: weeknight ceiling (probably 23:00 or 00:00)
  #   - Saturday: gaming runs LATE — 02:00 means "rolls into Sunday
  #     morning". Set equal to latest_end_weeknight if you want to
  #     opt out of late-night gaming.
  #   - Sunday: con effectively ends mid-afternoon.
  latest_end_weeknight: "23:00"
  latest_end_sunday: "15:00"
  latest_end_saturday: "02:00"

# Minimum minutes between events, ON TOP OF the walking time between
  # the two venues. 20 = "I always want a 20-minute breather". The
  # scheduler reads `data/walking_minutes.csv` for venue-to-venue
  # walking times, multiplies by `hustle_factor`, and adds this buffer.
  min_transfer_buffer_minutes: 20

# Reserved meal windows. An event overlapping one of these is NOT
  # blocked outright — it incurs a soft "meal displacement" score
  # penalty. A truly excellent event can win against a meal (see
  # `soft_preferences.meal_protection_threshold` below).
  #
  # >>> EDIT: shift these if you eat earlier/later. If you don't care
  # about meals at all, set to `[]` and `min_free_meal_blocks_per_day`
  # to 0.
  meal_blocks:
    - "12:30-13:30"
    - "18:30-19:30"

# Hard floor on free meal blocks per day. With two meal blocks above
  # and `min_free=1`, the optimizer can skip lunch OR dinner for a
  # killer event but never both.
  min_free_meal_blocks_per_day: 1

# Soft daily ceiling on event hours. Leaves room for the exhibit
  # hall, food, naps, walking, line-waiting, and the inevitable
  # "let's grab a drink" detour. 7-8 hours is typical; 6 if you
  # know you crash easily.
  max_scheduled_hours_per_day: 8

# Per-day USD spend cap (in addition to the weekend `max_budget`).
  # Set to a number to enable, `null` to disable. Useful if you want
  # to avoid a single $80 Saturday-night megagame chewing through
  # your whole budget.
  max_cost_per_day: null

# Per-day cognitive-load cap. Each event adds load by type:
  #   RPG / Workshop / True Dungeon-style    = 3
  #   Hobby / Board game / Seminar           = 2
  #   Entertainment / Film                   = 1
  # 9 fits "3 RPGs" or "2 RPGs + 1 workshop + 1 entertainment".
  # Lower this if you find yourself burning out by Saturday.
  max_cognitive_load_per_day: null

# ---- Exhibit Hall (Phase D Tier 2 — flexible window) ----------------
  # The hall is open 10:00-18:00. Instead of pre-booking a fixed window
  # (which often fights with a great event), this version reserves a
  # MINUTE BUDGET inside an 8-hour window — the optimizer picks the
  # cheapest place to spend it.
  #
  # If `combines_with_meal=true`, a meal block sitting inside the
  # window counts toward your hall budget. In other words: "Uber Eats
  # + walk the hall" collapses lunch and the hall walk into the same
  # hour. Set to false if you want hall and food kept separate.
  reserve_exhibit_hall_blocks: true
  exhibit_hall_window: "10:00-18:00"
  exhibit_hall_minutes_per_day: 90       # 60 = quick browse / 120 = serious
  exhibit_hall_combines_with_meal: true
  exhibit_hall_days: [Thursday, Friday, Saturday]
  # Legacy field, ignored when the four flexible fields above are set.
  # Kept for back-compat with old profiles. You can delete it.
  exhibit_hall_block: "14:00-16:00"

# ---- Walking-time tuning -------------------------------------------
  # The walking matrix is conservative ("planning minutes"). In
  # practice you can move faster on demand. `hustle_factor` is the
  # discount: 0.7 means "I can shave 30% off planning minutes when
  # I have to". Set to 1.0 for strict.
  hustle_factor: 0.7

# Walking-minute threshold that puts a pair into the "Avoid back-to-
  # back" band. Pairs above this are HARD-rejected unless separated by
  # `long_break_minutes` (a real meal/rest). Lower if you hate rushing.
  max_back_to_back_minutes: 25
  long_break_minutes: 60

# ---- Phase H5: early-exit semantics --------------------------------
  # The strict gap rule above ("walking + buffer between every pair")
  # produces robust schedules but doesn't reflect that most attendees
  # WILL leave a non-essential anchor 10-15 min early to make a tight
  # cross-venue transfer. Without an explicit override, the audit and
  # the optimizer reject schedules you'd actually accept.
  #
  # `transfer_overrides` is the declarative knob: when enabled, the
  # audit subtracts up to `max_early_exit_minutes` from the required
  # gap, BUT only for pairs where both rows' roles are in
  # `apply_only_to_roles` AND neither falls in `never_leave_early_roles`.
  # Default lists guard `must_include` / `dont_miss` rows, which you'd
  # never leave early in practice.
  #
  # Leave commented-out for the strict default (recommended for first
  # iteration). Uncomment when you trust your hard constraints and
  # want to encode the "I'll hustle" behavior explicitly.
  #
  # transfer_overrides:
  #   allow_early_exit: true
  #   max_early_exit_minutes: 15
  #   apply_only_to_roles: [anchor, extra, backup]
  #   never_leave_early_roles: [must_include, dont_miss]

# ---- Anti-clone caps ------------------------------------------------
  # Same normalized title (run/part/session suffixes stripped) cannot
  # appear more than this many times in your wishlist anchors —
  # protects you from "5 sessions of Cattle Raid Tier 1" crowding out
  # variety. Tight defaults (2 / 1) are usually right.
  max_sessions_per_title: 2
  max_sessions_per_title_per_day: 1

# ---- Optional shape controls (defaults are off) --------------------
  # Stop dense chains of one-hour filler events:
  max_events_per_day: null
  # Force a real break after a long run of events. 240 = "after 4
  # straight scheduled hours, take a break before the next pick":
  max_consecutive_scheduled_minutes: null
  # Ensure at least N rest blocks remain unscheduled per day. Useful
  # if you want a guaranteed afternoon coffee/decompress slot.
  min_rest_blocks_per_day: 0
  rest_blocks: []   # e.g. ["15:00-16:00", "17:00-18:00"]

# =====================================================================
# STRONG POSITIVE PREFERENCES — your "I love this" list. Used both for
# semantic ranking (the embedding sees these phrases as a query) AND for
# keyword bonuses (substring matches in title/description score).
# =====================================================================
#
# >>> EDIT: BE SPECIFIC AND VERBOSE. Use multiple phrasings for the
# same idea (e.g. "D&D" AND "Dungeons and Dragons" AND "5e"). The
# embedding generalizes pretty well, but the keyword bonus does
# substring matching, so spell out the variations you care about.
#
# Group your interests mentally into clusters — themes (the *kind* of
# story you like), craft (running games / making content), and
# systems/styles (specific games or mechanics). The retrieval engine
# breaks these into query clusters automatically and takes the best
# match across them, so dilution is limited.
#
# Examples that work well:
#   - moral dilemma, political intrigue, faction conflict
#   - heist design, mystery structure, faction clocks
#   - Dungeons & Dragons, 5e, Forged in the Dark
#   - miniature painting, terrain building, 3D printing for tabletop
#   - GM craft, GM seminar, encounter pacing
#
# Anti-patterns:
#   - too vague: "fantasy" (matches half the catalog)
#   - too long: 100+ entries dilute every cluster — see sizing below
#   - just system names with no themes — you'll get one-shot demos
#     instead of the kind of *content* you actually wanted
#
# Recommended cluster sizes:
#   15-30 phrases: casual players (low effort, broad coverage)
#   30-60 phrases: focused / portfolio players (alex.yaml is here)
#   60+ phrases:   ONLY safe if the cluster-aware retrieval is breaking
#                  the list into themed sub-queries — verify in
#                  `decision_report.md` that you see distinct cluster
#                  matches per event, not one amorphous mega-query.
#                  If everything cites the same 3 phrases, prune.
strong_positive_preferences:
  # >>> EDIT: replace these placeholders.
  - example interest 1
  - example interest 2

# =====================================================================
# NEGATIVE PREFERENCES — substring matches in title/description/system
# trigger a heavy `bad_fit_penalty` (-1.0 by default). One match is
# enough to push an event below the wishlist threshold.
# =====================================================================
#
# >>> EDIT: be careful with short tokens. They can match unintended
# substrings.
#   GOTCHA: "YA" matches "Bryan" (a GM name). Use "young adult" instead.
#   GOTCHA: "mini" matches "Mini-Game", "minimum", etc.
#            Use "miniature" if you mean the hobby; otherwise drop it.
#
# Other tokens that frequently misfire — DO NOT add unless you've
# verified they don't shadow content you'd actually want:
#   - OSR        (matches Mörk Borg, Shadowdark, Cairn descriptions)
#   - war        (matches Warhammer, war games, "ware", "awareness")
#   - dark       (matches Shadowdark, Mörk Borg, Blades in the Dark)
#   - kids       (matches "kids-friendly" but also "all kids welcome",
#                 "kid-tested", and any benign description text)
#   - adult      (matches "adult content" but also "adult-only $10"
#                 ticket-tier rows that aren't tonally adult)
#   - romance    (matches romance-forward AND "necromance(r)" — almost
#                 always the wrong choice)
#
# Good: full words / phrases that only appear in things you actually
# want filtered.
negative_preferences:
  - horror
  # >>> EDIT: add anything else. Common picks:
  # - cosmic horror
  # - body horror
  # - bleak survival
  # - young adult       # NOT "YA" — too short
  # - kids-friendly
  # - filler seminar
  # - autograph signing
  # - cosplay photo op

# =====================================================================
# SOFT PREFERENCES — these don't filter, they tilt the score.
# =====================================================================
soft_preferences:
  # Penalize early-morning events. Set to your "no events before X"
  # time. The scheduler can still pick before this — it just costs
  # score points proportional to how much earlier than X the event
  # starts.
  prefer_start_after: "10:00"

# Reward schedules that have at least one high-score "anchor" event
  # per day. Calibrate `anchor_score_threshold` to roughly the score
  # of "a solid on-theme RPG you'd be excited about".
  prefer_one_anchor_event_per_day: true
  anchor_score_threshold: 1.5
  prefer_anchor_blocks: true
  # >>> EDIT: latest start time for an event to still count as an
  # "anchor". 21:00 lets late-night gaming anchor your Saturday.
  # null = no constraint. Try "20:00" for early-bed schedules.
  latest_anchor_start: null

# Reward schedules with a variety of event types (RPG + Workshop +
  # Hobby) instead of all-RPG-everything.
  prefer_variety_across_days: true

# Meal-displacement softening. If an event's pre-meal score exceeds
  # this threshold, the meal-displacement penalty for it is halved.
  # Higher = stricter (meals more protected). 1.5 is "must be a strong
  # event"; 2.0 is "must be excellent".
  meal_protection_threshold: 1.5

# =====================================================================
# OPTIONAL EXPLICIT LISTS — substring-matched against the corresponding
# event field. Leave [] to fall back on `strong_positive_preferences`.
# =====================================================================

# >>> EDIT: which event-type codes you'll consider. Common ones:
#   RPG - Roleplaying Game
#   BGM - Board Game
#   CGM - Collectible Card Game
#   ENT - Entertainment Events
#   FLM - Film Festival
#   LRP - Live Action Role Playing
#   MHE - Miniature Hobby Events
#   NMN - Non-Historical Miniatures
#   RPGA - Role Playing Game Association
#   SEM - Seminar
#   TCG - Tradeable Card Game
#   TDA - True Dungeon (premium)
#   TRD - Trade Day Events
#   WKS - Workshop
preferred_event_types: []
# Example:
# preferred_event_types:
#   - RPG - Roleplaying Game
#   - WKS - Workshop
#   - SEM - Seminar
#   - MHE - Miniature Hobby Events

# >>> EDIT: substring matches against `game_system` field. Use exact
# casing of the system name when possible. Multiple synonyms ok.
preferred_systems: []
# Example:
# preferred_systems:
#   - Dungeons & Dragons
#   - D&D
#   - 5th Edition
#   - Pathfinder

# >>> EDIT: substring matches against `gm_names` field (often "First
# Last; First Last; Company Name").
preferred_gms_or_companies: []
# Example:
# preferred_gms_or_companies:
#   - Wizards of the Coast
#   - MCDM
#   - Free League

# =====================================================================
# RANKED PREFERENCES (Phase D Tier 1, optional but RECOMMENDED).
# =====================================================================
#
# When you set `preferred_systems_ranked` (or `..._gms_..._ranked`),
# the WEIGHTED map supersedes the flat list above for scoring. The
# implementation short-circuits in `gencon/scoring.py` (see
# `_score_system` and `_score_gm`): when a ranked weight is found,
# the function returns immediately and the flat list is not consulted.
# This means a system at ranked weight 0.0 (or any low value) cannot
# silently revert to the flat list's full credit — once a system has a
# ranked weight, that weight is authoritative.
#
# Use weights in [0.0, 1.0]:
#   1.0 = "favorite, will pick this over anything else"
#   0.7 = "would happily play"
#   0.4 = "familiar, but not a draw"
#   0.2 = "I'll tolerate it for an excellent specific event"
#
# This solves the "all my favorite systems look identical in
# ranked.csv" tie cluster — you'll see Draw Steel ranking above
# Pathfinder even when both match keywords equally well.
#
# Pick weights with INTENT: at most 3-4 systems should be at 1.0,
# otherwise everything ties again. Aim for a clear top tier, a strong
# middle, and a long tail.
preferred_systems_ranked: {}
# Example:
# preferred_systems_ranked:
#   # Tier 1: top-2 favorites
#   Dungeons & Dragons: 1.0
#   D&D: 1.0
#   # Tier 2: very high
#   Daggerheart: 0.85
#   Mothership: 0.85
#   Shadowdark: 0.85
#   # Tier 3: strong
#   Forged in the Dark: 0.7
#   Blades in the Dark: 0.7
#   The One Ring: 0.7
#   # Tier 4: low
#   Pathfinder: 0.4
#   Starfinder: 0.3

preferred_gms_or_companies_ranked: {}
# Example:
# preferred_gms_or_companies_ranked:
#   MCDM: 1.0
#   Darrington Press: 1.0
#   Free League: 0.95
#   Tuesday Knight: 0.85
#   Goodman Games: 0.7
#   Wizards of the Coast: 0.5

# =====================================================================
# DESIGN-STUDY EXCEPTIONS (optional, advanced).
# =====================================================================
#
# Use this for systems you'd normally filter out (because of tone/genre)
# but whose adventure/encounter design is worth studying. Matching
# events get a `design_extraction_value` bonus AND can selectively
# whitelist specific negative keywords for themselves.
#
# Worked example: you've banned "horror" in negative_preferences, but
# Mothership module-design seminars are some of the best in the
# industry. A `design_study_exception` with `system_match: [Mothership]`
# and `suppress_negative_keywords: [horror]` gets you the seminars
# without unbanning horror globally.
#
# When to use this:
#   - You're a GM who wants to learn from systems you don't play
#   - There's a specific *kind* of event (seminar, designer talk) in
#     a system you avoid
#
# When NOT to use it:
#   - You actually like the system in general — just put it in
#     `preferred_systems_ranked`
design_study_exceptions: []
# Example:
# design_study_exceptions:
#   - system_match: ["Mothership"]
#     reason: >
#       Best-in-market module/adventure design — study layout, panic
#       economy, and table-runnable adventure structure.
#     prefer_formats:
#       - seminar
#       - workshop
#       - module design
#       - Warden
#     avoid_formats:
#       - extreme horror
#       - body horror
#       - late night
#     extraction_targets:
#       - layout
#       - room design
#       - faction clock
#       - panic
#     suppress_negative_keywords:
#       - horror
#     bonus_weight: 1.0

# =====================================================================
# MUST-INCLUDE CATEGORIES (Phase B, optional but POWERFUL).
# =====================================================================
#
# Reserved Wish List slots. The composer guarantees that at least
# `minimum` events matching each category land in the 50-item Wish
# List, by injecting the highest-scoring ranked match as an anchor at
# the front of the priority order if the optimizer didn't already
# include enough.
#
# Use this when you'd rather cut a generic anchor than miss a SPECIFIC
# kind of slot. Common use cases:
#   - "Reserve one Mothership/Draw Steel/Daggerheart-style session,
#     even if my preferences pulled the optimizer toward 5 D&D one-shots"
#   - "Always include one Workshop with painting/terrain content"
#   - "Always include one Game Design Academy / FEPH / Writers
#     Symposium event"
#
# CRITICAL: do NOT use must_include_categories for "sounds nice".
# Use it ONLY for trip-defining categories you would happily cut a
# generic anchor for. Everything else belongs in
# `nice_to_have_categories` — those are tagged role='anchor' and skip
# the cart-aware promotion pass, so they won't fight your real anchors.
#
# Match semantics (AND between matchers; OR within each matcher's list):
#   - `match_systems`        — substring match against `game_system`
#   - `match_event_types`    — substring match against `event_type`
#                              (e.g. "WKS - Workshop" matches anything
#                              starting with "WKS")
#   - `require_text_any_of`  — substring match against the FULL search
#                              text (title + description + system + gm
#                              + special_category). Be specific to
#                              avoid description false-positives.
must_include_categories: []
# Example:
# must_include_categories:
#   - name: mothership_design_study
#     match_systems: ["Mothership"]
#     # Empty require_text_any_of: any Mothership event qualifies.
#     # If you want only seminars/workshops, add specific terms.
#     require_text_any_of: []
#     minimum: 1
#
#   - name: design_seminar_or_writers_symposium
#     match_event_types: ["SEM - Seminar", "WKS - Workshop"]
#     # Use full words; avoid 3-char tokens like "GM" that match too
#     # broadly across the catalog descriptions.
#     require_text_any_of:
#       - game design
#       - playtest
#       - writers symposium
#       - encounter design
#       - module design
#       - worldbuilding
#     minimum: 1
#
#   - name: miniatures_or_terrain_workshop
#     match_event_types:
#       - WKS - Workshop
#       - MHE - Miniature Hobby Events
#       - NMN - Non-Historical Miniatures
#     # Use FULL words. "mini" alone matches "Mini-Game" — false positive.
#     require_text_any_of:
#       - miniature
#       - painting
#       - terrain
#       - sculpt
#       - airbrush
#       - 3d print
#       - dwarven forge
#       - hobby
#     minimum: 1

# =====================================================================
# MANUAL TRANSFER EXCEPTIONS (Phase H5) — surgical early-exit allowances
# =====================================================================
#
# Per-pair allowance for the WALKING_BUFFER_INSUFFICIENT audit. When a
# wishlist row pair matches `from_event_title` -> `to_event_title` by
# case-insensitive substring on title, the audit downgrades the WARN to
# an `EARLY_EXIT_REQUIRED` INFO line and applies
# `allow_leave_early_minutes` of slack on top of any
# `transfer_overrides` allowance.
#
# Use this for *one-off* known transfers you've vetted, not as a
# general policy (use `transfer_overrides` for that). Common case: an
# anchor and a must_include that touch at exactly the building boundary
# but you've decided you'll leave the anchor early.
manual_transfer_exceptions: []
# Example:
# manual_transfer_exceptions:
#   - from_event_title: "The Black Ziggurat"
#     to_event_title: "You Meet in an Ambush"
#     allow_leave_early_minutes: 15
#     reason: "Draw Steel is protected; willing to leave Coriolis early."

# =====================================================================
# NICE-TO-HAVE CATEGORIES (Phase G3) — soft floors. Same matcher
# semantics as `must_include_categories` but rows are tagged
# `role='anchor'` and skip the cart-aware promotion pass. If a higher-
# priority earlier row blocks a nice_to_have row, it simply won't land.
# =====================================================================
#
# When to use which:
#   - `must_include_categories`  — "I have explicitly reserved this slot;
#                                   if blocked, promote it above the
#                                   blocker (incl. dont_miss)."
#   - `nice_to_have_categories`  — "This category should appear if the
#                                   catalog allows, but never at the
#                                   cost of demoting a real must_include
#                                   or dont_miss."
#
# Typical use case: hobby/exhibit-hall floors that you want to reserve
# space for if it doesn't conflict, but are willing to skip if your
# core RPG portfolio fills the morning.
#
# Worked example (commented out by default — uncomment + edit):
#
# nice_to_have_categories:
#   - name: miniatures_or_terrain_workshop
#     match_event_types:
#       - WKS - Workshop
#       - MHE - Miniature Hobby Events
#       - NMN - Non-Historical Miniatures
#     require_text_any_of:
#       - miniature
#       - painting
#       - terrain
#       - sculpt
#       - airbrush
#       - 3d print
#       - dwarven forge
#       - hobby
#     minimum: 1
nice_to_have_categories: []

# =====================================================================
# WISHLIST DIVERSITY CAPS (Phase E2) — optional. Stops the wishlist
# composer from filling 24 of your 50 slots with one popular family
# (the "12 Deathwish5e + 7 Sablewood" monoculture problem).
# =====================================================================
#
# Two parallel mechanisms:
#
#   wishlist_system_targets — substring-matches event.game_system
#       wishlist_system_targets:
#         "Dungeons & Dragons": [12, 16]    # [min, max]
#         "Mothership":         [2,  4]
#
#   wishlist_signature_cluster_targets — matches signature `cluster_id`
#   from data/event_signatures.yaml. Use this when a single "system"
#   field hides multiple sub-families (e.g. all the Deathwish5e
#   sessions are technically D&D, but they're really their own thing).
#
#       wishlist_signature_cluster_targets:
#         deathwish5e: [1, 4]
#         sablewood:   [1, 4]
#
# Range semantics: `[min, max]`. The MAX is a hard cap on backup +
# adjacent_backup picks (anchors and must_include rows are exempt —
# they encode floor commitments). The MIN is informational; pair it
# with a `must_include_categories` entry if you need a *guaranteed*
# floor.
#
# Both maps are optional. Leave them undefined and the composer
# behaves like Phase D (no diversity caps).
#
# Worked example (commented out by default — uncomment + edit):
#
# wishlist_system_targets:
#   Dungeons & Dragons: [12, 16]
#   Daggerheart:         [2,  4]
#   Shadowdark:          [3,  5]
#   Mothership:          [2,  4]
#   Draw Steel:          [2,  4]
#   Free League:         [2,  5]
#   Pathfinder:          [0,  2]
#
# wishlist_signature_cluster_targets:
#   deathwish5e:        [1, 4]   # premium D&D variants — break monoculture
#   sablewood:          [1, 4]   # 1hr Daggerheart intros — one is enough
#   pathfinder_demo:    [0, 2]   # filler at best
#   shadowdark_module:  [1, 3]
#   free_league_module: [1, 3]
#   design_academy:     [1, 4]   # FEPH + Game Design Academy + Writers Symp

# =====================================================================
# WEIGHTS — how the score components are combined. Defaults are
# calibrated; reach for these only if your decision_report shows a
# specific imbalance.
# =====================================================================
#
# Score formula (all positive bonuses ADD, all penalties SUBTRACT):
#
#   score =  semantic_interest * embedding_similarity
#         +  category_preference * (event matches preferred_event_types)
#         +  system_preference   * (matches preferred_systems / ranked)
#         +  gm_company_preference * (matches preferred_gms / ranked)
#         +  novelty_bonus       * (rare event_type for this profile)
#         +  practical_skill_bonus * (workshop / hands-on craft)
#         +  rarity_bonus        * (few tickets / hard-to-get)
#         +  design_extraction_value * (matches a design_study_exception)
#         +  premier_bonus       * (Gen Con Presents / premier event)
#         +  cost_premium_bonus  * (high cost-per-hour + small ticket pool)
#         +  designer_led_bonus  * (run by a curated designer/publisher)
#         +  same_cluster_bonus  * (per day, all events in one walking cluster)
#         +  signature_bonus     * (matches data/event_signatures.yaml)
#         +  immersive_set_bonus * (terrain / minis / Dwarven Forge tabling)
#         +  anchor_duration_bonus * (deep 2-4hr RPGs / 90-180min workshops)
#         +  keyword_relevance   * (FTS5 BM25 score, hybrid retrieval)
#         +  anchor_per_day_bonus * (per day with score >= anchor threshold)
#         +  variety_bonus       * (per distinct event_type in schedule)
#
#         -  time_of_day_penalty * (event starts before prefer_start_after)
#         -  travel_penalty      * (back-to-back venue distance)
#         -  fatigue_penalty     * (consecutive scheduled hours)
#         -  bad_fit_penalty     * (matches negative_preferences)
#         -  meal_displacement_penalty * (eats into a meal block)
#         -  back_to_back_penalty * (per minute under hustle-buffered gap)
#         -  event_count_penalty * (per selected event — favors depth over breadth)
#
# Tuning hints
# ------------
#  * Wish list always has filler? raise `event_count_penalty` magnitude
#    (e.g. -0.05 -> -0.10). Forces the optimizer to demand stronger picks.
#  * Hate dense schedules? raise `fatigue_penalty` magnitude.
#  * Top picks all the same system? raise `variety_bonus` to 0.2 OR
#    use `preferred_systems_ranked` to make ties explicit.
#  * Missing premier events? raise `premier_bonus` to 0.6.
#  * Want lower-cost picks? lower `cost_premium_bonus` to 0 (defaults
#    *reward* premium spending — disable if you're budget-conscious).
weights:
  # ---- Positive (rewards) -------------------------------------------
  semantic_interest: 1.0
  category_preference: 0.5
  system_preference: 0.4
  gm_company_preference: 0.3
  novelty_bonus: 0.2
  practical_skill_bonus: 0.3
  rarity_bonus: 0.15
  # Bonus per matching design_study_exception (e.g. Mothership
  # seminars). Calibrated to roughly match category_preference so a
  # matched design-study event ranks alongside on-genre RPG events.
  design_extraction_value: 0.6
  # Hybrid retrieval: keyword (FTS5/BM25) bonus on top of the
  # embedding-based semantic_interest. Catches literal matches the
  # embedding misses (specific publishers, exact module names).
  keyword_relevance: 0.4
  # Reward 2-4hr RPG sessions and 90-180min workshops/seminars over
  # one-hour filler.
  anchor_duration_bonus: 0.25
  # Soft-pref objective bonuses (set 0 to disable):
  anchor_per_day_bonus: 0.3
  variety_bonus: 0.1
  # Phase C signal-quality bonuses (safe defaults across all profiles):
  premier_bonus: 0.4
  cost_premium_bonus: 0.25
  designer_led_bonus: 0.4
  # Per-day same-cluster bonus (anchors the day at one geographic
  # cluster, e.g. ICC + neighbors). Set 0 to allow free movement.
  same_cluster_bonus: 0.15
  # Phase D Tier 1: curated event-signature bonus, multiplied by each
  # row's signature_bonus in `data/event_signatures.yaml` ([0, 1]).
  # See ADR 0012 for what this is and how to add your own entries.
  signature_bonus: 0.5
  # Phase D Tier 1: terrain/miniatures/immersive-set bonus. Catches
  # RPG sessions that *use* premium tables ("w/ Terrain, Miniatures,
  # Dice", "Dwarven Forge", "fully painted minis"). Distinct from
  # `practical_skill_bonus` which catches hobby workshops.
  immersive_set_bonus: 0.4

# ---- Negative (penalties; values are negative) ---------------------
  time_of_day_penalty: -0.4
  travel_penalty: -0.2
  fatigue_penalty: -0.3
  # bad_fit_penalty fires once per matching negative_preference token.
  # -1.0 is "one match is enough to push the event out of the wishlist".
  bad_fit_penalty: -1.0
  # Scales with how much of a meal block the event consumes.
  meal_displacement_penalty: -0.5
  # Per-minute objective penalty between adjacent events whose temporal
  # gap is shorter than (hustle * walking_minutes + buffer).
  back_to_back_penalty: -0.05
  # Per selected event. Helps the optimizer prefer fewer deep anchors
  # over many shallow micro-events.
  event_count_penalty: -0.05

# =====================================================================
# That's it! Run `make lint-profile P=<your_name>` to validate, then
# `make schedule P=<your_name>` to generate your wishlist.
#
# Outputs land in `outputs/<your_name>/`:
#   * wishlist.csv          — 50 rows, priority-ordered, with backups
#   * wishlist_paste.txt    — copy/paste-friendly slot-grouped view
#   * wishlist_final.csv    — Monte Carlo-optimized priority order
#                             (run `make optimize-wishlist P=<n>` to refresh)
#   * schedule.json/.md     — your "best case" personal schedule
#   * schedule.ics          — calendar feed for your phone
#   * decision_report.md    — what the picks are and why they ranked
#   * audit_report.md       — sanity-check findings (must_include
#                             coverage, anti-clone, day caps, etc.)
#
# Iterate by reading decision_report.md, tweaking the file, and
# re-running. The whole pipeline takes ~10 seconds end-to-end on a
# laptop.
# =====================================================================

5 jobs / hour / IP