Make every denominator explicit

{
  "id": "game_draw_rate",
  "version": "1.0.0",
  "grain": "one provider/game_id",
  "inputs": [
    "fact_game"
  ],
  "filters": "rated Standard games, excluding source-marked bots, completed results only",
  "numerator": "count of eligible games with result 1/2-1/2",
  "denominator": "count of eligible games with result in 1-0, 0-1, 1/2-1/2",
  "formula": "numerator / denominator",
  "null_rules": "unknown results excluded and counted; empty denominator yields null",
  "units": "fraction",
  "result_schema": {
    "numerator": "integer",
    "denominator": "integer",
    "value": "number|null",
    "unknown_results": "integer",
    "marked_bots": "integer"
  },
  "caveats": [
    "Historical foundation corpus is not evidence about contemporary players.",
    "Excluding marked bots does not establish all other players are human.",
    "No opening, time-pressure, or causal claims are supported by this metric."
  ]
}

{
  "id": "opening_usage",
  "version": "1.0.0",
  "grain": "one provider/game_id in the observed bounded prefix",
  "inputs": [
    "fact_game.source_opening",
    "fact_game.result",
    "fact_game.marked_bot"
  ],
  "classification": "source opening family is the trimmed text before the first colon in the provider tag; no local position classifier",
  "numerator": "eligible games in selected source opening family",
  "denominator": "eligible games with a nonempty source opening family",
  "eligible": "rated Standard accepted games with completed result, excluding source-marked bots",
  "null_rules": "missing/empty source opening excluded from denominator and reported separately",
  "units": "fraction",
  "caveats": [
    "The 100,000-game prefix is a convenience sample, not an August 2026 population sample.",
    "Opening tags are provider supplied; opening choice is confounded."
  ]
}

{
  "id": "opening_player_score",
  "version": "1.0.0",
  "grain": "one player-game at selected color, opening family, exact time control, and rating band",
  "inputs": [
    "fact_game",
    "fact_player_game"
  ],
  "filters": "completed rated Standard, no source-marked bot, known source opening, known focal rating; half-open rating band",
  "win_rate": "wins / (wins + draws + losses)",
  "score_rate": "(wins + 0.5 * draws) / (wins + draws + losses)",
  "null_rules": "unknown result and missing focal rating excluded and counted in coverage",
  "units": "fraction",
  "low_support_threshold_games": 100,
  "caveats": [
    "Descriptive association in the observed prefix only.",
    "Do not infer a causal advantage from opening choice.",
    "Focal-player games are dependent across players and opponents; M4 uncertainty method must account for this."
  ]
}

{
  "id": "opening_adjusted_score",
  "version": "1.0.0",
  "grain": "one focal player-game in a selected source Opening family",
  "filters": "completed non-marked-bot games; exact color, half-open player rating band, exact base+increment, two named source opening families; recorded opponent rating required for adjustment",
  "score": "wins plus 0.5 times draws divided by eligible player-games",
  "strata": "player rating minus recorded opponent rating: below -100, -100 to 99, and 100 or more",
  "common_weights": "pooled eligible player-game counts over strata represented in both openings, normalized to sum to one",
  "adjusted_result": "within-stratum score rates for each opening averaged using identical common weights",
  "interval": "400 seeded focal-player cluster bootstrap repetitions within each opening; percentile 2.5 and 97.5 bounds among valid resamples",
  "low_support": "fewer than 100 eligible player-games is a project warning",
  "null_rules": "missing opponent rating is excluded from adjustment but remains in raw score; a missing common stratum yields no adjusted rate",
  "units": "score fraction",
  "caveats": [
    "Observed ordered prefix only; no month-wide inference.",
    "Association, not opening-choice causation.",
    "Bootstrap does not model opponent dependence or archive-prefix selection."
  ]
}

{
  "version": "1.1.0",
  "name": "observed_white_opening_positions",
  "source_grain": "checked source snapshot \u00d7 provider \u00d7 game ID \u00d7 White decision ply",
  "analysis_grain": "visit snapshot \u00d7 cohort \u00d7 optional recorded opening family \u00d7 canonical position",
  "position_key_version": "standard-fen-first-four-legal-ep-1.0",
  "position_key": "Piece placement, side to move, castling rights, and en passant only if legal; first four normalized FEN fields. SHA-256 truncated to 24 hex characters; retained key and collision check.",
  "history": "Full FEN, full UCI prefix and repeated-position flag remain in local visit shards. Draw counters and repetition history are not part of the recognition key.",
  "default_window": {
    "learner_color": "white",
    "min_ply": 6,
    "max_ply": 20,
    "step": 2
  },
  "game_count": "A game contributes once per canonical position within the window; the earliest visit supplies its route and continuation.",
  "frequency": "Distinct games reaching the position / all eligible games in that cohort and, when selected, recorded opening family. Not a personal encounter estimate.",
  "opening_family": "Recorded game opening tag before its first colon. ECO is retained separately. Labels do not define or merge board positions.",
  "recurring": "At least two distinct games reach the exact position.",
  "acyclic_route_count": "Distinct complete UCI arrival prefixes with no repeated canonical position anywhere in that prefix. Looped prefixes are counted in game frequency but not in this route count.",
  "transposing": "At least two distinct acyclic arrival prefixes reach the exact position.",
  "ranking": "Descending distinct-game count, with position ID as deterministic tie breaker; up to 20 recurring and 20 transposing positions per view.",
  "route_share": "Games with this earliest arrival prefix / position games; up to three non-looped examples are published, so displayed shares need not sum to one.",
  "continuation_share": "Games choosing this legal next White move at the earliest visit / position games, including games without a next move in the denominator. Report next_move_games separately. Up to five moves are published.",
  "top_set_coverage": "Union of distinct games encountering any first-N frequency-ranked position / eligible view games, N=1..20. Overlapping position frequencies must not be added.",
  "endpoint_compression": "At a fixed ply, 1 \u2212 distinct canonical endpoint positions / distinct UCI prefixes. Exclude any prefix with a repeated canonical position. Descriptive convergence, not a measured learning benefit.",
  "publication": "Separate Elite and broader public cohorts. Publish all opening-family inventory counts, plus all-opening rankings and up to twelve largest known families with at least 1000 eligible games. Local hashes bind retained inputs, extraction plan, parser implementation and artifacts.",
  "limits": "Elite is a curated November 2025 month; public sources are nine ordered first-day archive prefixes. Frequency is not move quality, strategic importance, or causal evidence. No engine scoring, personal weighting, drills, or live provider requests.",
  "alternative_route_share": "1 - dominant acyclic route games / acyclic position games; acyclic position games = position games - repetition-prefix games. Null for zero acyclic games. Route cards retain all position games as denominator.",
  "coverage_curve": "For each N=1..20 in the frequency ranking, union distinct games reaching any first-N position / eligible view games. additional_games is the difference from N-1. This is not a greedy coverage ranking.",
  "opening_label_breadth": "Count all distinct known recorded families and all distinct known ECO codes separately; unknown-family games reported separately. Five displayed family/ECO pairs are examples, not exhaustive breadth. Labels can reflect later play.",
  "route_remainder": "Other acyclic route games + repetition-prefix games + top-three acyclic route games = position games.",
  "public_lenses": {
    "d4-g3": {
      "label": "1.d4 with g3",
      "moves": [
        "d2d4",
        "g2g3"
      ]
    },
    "d4-g3-catalan": {
      "label": "Catalan-style fianchetto",
      "moves": [
        "d2d4",
        "c2c4",
        "g2g3",
        "f1g2",
        "e7e6",
        "d7d5"
      ]
    },
    "d4-g3-kings-indian": {
      "label": "King's Indian fianchetto",
      "moves": [
        "d2d4",
        "c2c4",
        "g2g3",
        "f1g2",
        "g8f6",
        "g7g6",
        "f8g7",
        "d7d6"
      ]
    }
  },
  "lens_policy": "First White move d4; all required UCI moves present by max_ply. Curated move-sequence facets, not exhaustive taxonomy. Denominator is matching games; include eligible positions before and after g3. No personal histories.",
  "scatter_scope": "Published ranked positions in the selected view only, deduplicated by position ID; not a corpus-wide distribution."
}

{
  "id": "local_user_opening_analysis",
  "version": "1.0.0",
  "input": "Plain UTF-8 multi-game PGN exported from Chess.com, Lichess or another source. Mainline legal replay from the initial Standard position; variations/comments/NAGs do not contribute visits.",
  "limits": {
    "bytes": 10485760,
    "games": 5000,
    "plies": 1000,
    "variationDepth": 32,
    "files": 20,
    "worker_seconds": 60
  },
  "privacy": "Files and normalized games remain in the browser worker for the session. No account connection, API downloads, model calls, server uploads or persistent storage. Clearing/reloading terminates the worker. Report download is explicit and contains summaries rather than raw PGNs.",
  "eligibility": "Completed known results and legal Standard mainlines with named players. Reject variants, setup/FEN games, unfinished or conflicting results, duplicate headers and illegal mainlines. Do not impose reference-corpus rating floors. Missing ratings and casual games are supported; unknown metadata stays unknown.",
  "identity": "Provider + native game ID from recognized Site/Link URLs. Chess.com live/daily namespaces stay separate; Lichess uses the first eight game-ID characters. Conflicting provider/native-ID headers are rejected. Generic fallback is SHA256 of normalized player names, date and UCI mainline; source-file SHA256 is separate.",
  "deduplication": "Semantic fingerprint of normalized players, UCI moves and completed result excludes comments, variations, NAGs and other optional metadata. Same identity with equal fingerprint counts once; any differing fingerprint withholds all copies of that identity as conflicts. Provider identities never collapse across platforms. First copy supplies optional metadata.",
  "reconciliation": "Framed records = accepted unique games + duplicate copies + conflicting copies + rejected records. Failed framing imports no partial file and its unframed records are outside that count. Invalid UTF-8, byte/file/game/ply/variation-depth caps, cancellation or timeout retain no partial session.",
  "player_selection": "Explicit selected player per provider; NFC lowercase matching within that provider. A matching name on another provider is not automatically the same person. White first; matched Black games and unmatched-player records are reported separately.",
  "personal_denominator": "All selected valid completed Standard White games after explicit date/time-control/rated filters, including short games. Date-filtered undated games are excluded visibly. An optional minimum-20-ply comparison filter has its own exclusion count.",
  "position_identity": "First four canonical FEN fields with legal en passant; SHA256 truncated to 24 hex characters, matching the checked public reference. Preserve full FEN separately. Draw counters/history are not part of the recognition key.",
  "window": {
    "learner_color": "white",
    "min_ply": 6,
    "max_ply": 20,
    "step": 2
  },
  "position_metrics": "One earliest eligible visit per game/position. Frequency = position games / selected White games. Acyclic route count excludes any repeated-position prefix. Alternative-route share = 1 - largest acyclic route game count / (position games - repetition-prefix games), null for zero acyclic games.",
  "reference_scope": "Selected first-N published study positions, N=1..20, for one cohort/view/ranking. Coverage is the union of selected White games reaching those boards. No match means not in this selected set, not unknown in the full corpus or out of theory.",
  "route_comparison": "First differing move while the selected published route examples still define a next move, within the first 20 plies. An example ending stops comparison; it is not a departure. A later canonical White decision matching a selected board counts as re-entry. This is not a move-quality or mistake metric.",
  "metadata": "Date is a valid UTCDate or Date header, otherwise null; do not infer UTC time. Rated status uses explicit Rated true/false or Rated/Casual/Unrated Event prefixes; Chess.com branding alone does not establish rated status. TimeControl is retained without guessed speed/rating comparability.",
  "limits_of_inference": "Personal samples and public reference cohorts differ. Coverage is a descriptive count of the selected reference set. No engine evaluation, learning-gain claim, causal result comparison, username downloads or private-game authentication. Generic fingerprints cannot resolve every missing-ID ambiguity."
}

{
  "id": "clock_pressure_error_proxy",
  "version": "1.0.0",
  "grain": "one selected game x ply, stratified by clock-before bucket",
  "sampling": "game ID SHA256 first eight bytes modulo 20 equals zero; deterministic 5% selection from the bounded prefix, independent of annotations",
  "eligible": "non-first move by a player in a zero-increment rated Standard game with valid prior same-player clock and a completed result, excluding source-marked bots",
  "buckets_seconds": [
    "[0,10)",
    "[10,30)",
    "[30,60)",
    "[60,infinity)"
  ],
  "evaluable": "eligible move with source-provided before and after centipawn evaluations; mate transitions excluded and counted separately",
  "deterioration": "max(0, mover-perspective cp_before minus cp_after); source cp scores are White-perspective, invert both for Black",
  "numerator": "evaluable eligible moves with deterioration >= 200 centipawns",
  "denominator": "evaluable eligible moves in bucket",
  "coverage_denominator": "all eligible moves in bucket",
  "null_rules": "first same-player move, missing/invalid clock, missing evaluations and mate transitions never become zero error",
  "units": "fraction",
  "caveats": [
    "Exploratory proxy, not an official Lichess blunder label.",
    "Existing evaluation coverage is selected and source quality varies.",
    "The prefix and move sample do not represent all August games; no causal time-pressure claim."
  ]
}

{
  "id": "evaluation_coverage",
  "version": "1.0.0",
  "grain": "one selected game x ply, by clock-before bucket",
  "inputs": [
    "fact_move"
  ],
  "filters": "same eligible move and deterministic game sample as clock_pressure_error_proxy",
  "numerator": "eligible moves with comparable before and after source centipawn evaluations",
  "denominator": "all eligible moves in bucket",
  "null_rules": "missing source evals and mate transitions are non-evaluable, never zero deterioration",
  "units": "fraction",
  "caveats": [
    "Source evaluation availability can depend on which games were analyzed.",
    "Coverage is conditional on the observed prefix and deterministic move-game sample."
  ]
}