mechanics_research_brief

generated_by: scripts/analyze.py (track C) snapshot_date: 2026-04-30 filter: playtime ≤ 60min, weight ≤ 3.0, users_rated ≥ 1000, no social mechanisms min_games_per_mechanic: 3 plan: docs/plans/v4-archetype-fit-pipeline.md

This brief accompanies mechanics_research_dataset_v1.csv — a mechanic-grain summary of the v4-wide LLM extraction over the BGG translation-eligible corpus. Archetype-blind: every signal here is a property of the mechanic itself (how strongly it produces each MDA aesthetic, what session length its core loop tolerates, how procgen-friendly its content is), not yet biased toward any specific iOS game form factor.

Methodology

Corpus: 5988 game-mechanic rows aggregated to 152 mechanics (mean 39.4 games/mechanic; mechanics with fewer than 3 eligible games dropped).
Eligibility filter: ≤60min playtime, ≤3.0 weight, ≥1k BGG ratings, no social mechanisms (Hidden Roles, Voting, Negotiation, Trading, Bluffing, Acting). Preserves the existing translation-fit eligibility floor used since v2.
Per-game extraction: Sonnet 4.6 against prompts/extraction_v4_wide.md, schema in game_llm_features_v4. 100% schema compliance held across all 1501 rows. See docs/llm_throughput_reference.md for cost.
Mechanic aggregation: per-game scores averaged within each BGG mechanic. top_mech_interaction_primitives ranks the controlled-vocab primitives that co-occur with each mechanic. top_5_exemplar_games are the games scoring highest on discovery_score (composite of rank-trajectory delta + recency-hit density + translation-difficulty residual) within the mechanic.
Discovery score (archetype-blind): signed log of (rank_2019 / rank_now) plus recency density of post-2020 top-1000 hits plus the residual of translation-difficulty after weight-controlled regression. High = under-translated mechanic with rising recent attention.

Headline findings

Top 10 mechanics by mean discovery score

Rank	Mechanic	n	Discovery	Session min	Extensibility	Top exemplar
1	Force Commitment	4	+1.73	20	1.25	Hanamikoji
2	Worker Placement, Different Worker Types	6	+1.32	18	1.50	Victorian Masterminds
3	Increase Value of Unchosen Resources	12	+1.24	14	1.58	Age of Civilization
4	Tech Trees / Tech Tracks	7	+1.23	19	2.00	Age of Civilization
5	Resource to Move	7	+1.17	15	1.43	The Quest for El Dorado
6	Take That	4	+1.12	16	2.00	Tyrants of the Underdark
7	Card Play Conflict Resolution	17	+1.10	12	1.59	Air, Land, & Sea
8	Delayed Purchase	14	+1.06	14	2.21	War Chest
9	Chaining	10	+1.02	14	1.50	Blue Lagoon
10	Multi-Use Cards	27	+0.91	14	1.96	Fantastic Factories

Bottom 5 (mechanics with the most negative discovery score)

Mechanic	n	Discovery	Reading
Pattern Movement	3	-0.65	falling rank-trajectory or low recency density
Critical Hits and Failures	3	-0.45	falling rank-trajectory or low recency density
Movement Template	3	-0.37	falling rank-trajectory or low recency density
Stacking and Balancing	22	-0.33	falling rank-trajectory or low recency density
Stat Check Resolution	6	-0.32	falling rank-trajectory or low recency density

How to use this dataset

Designers: use mean_mda_aesthetic_vector to find mechanics that produce a specific aesthetic (e.g., high Discovery + high Challenge for roguelike-feel) and top_5_exemplar_games to ground each mechanic in concrete play.
Researchers: cite the corpus filter and provenance_run_id from each row. Re-running this analysis against a future BGG snapshot lets you measure mechanic drift over time.
iOS-fit overlay (Workstream 2 of the v4 plan): joins this dataset with game_llm_archetype_fit_v4 to produce per-archetype rankings. The W2 overlay is internal; this W1 dataset is the publishable layer.

Provenance

BGG SQLite snapshot: 2026-04-30
v4-wide extraction runs surveyed: 1 distinct run IDs
Vocab version range: see vocab_version column on game_llm_features_v4
Plan: docs/plans/v4-archetype-fit-pipeline.md