Contract 3.13.0

Every aggregate ships the control next to the number (feedback #54, #55, #56, #58, #59, #60: six wrong conclusions from correct data in one agent session). battles_decks rows carry modes (battles, wins, losses per mode group), dominant_mode and dominant_mode_share, mean_level_gap, own_mean_level, opponent_mean_level and level_gap_battles; the response carries comparable, false when the rows were played in different modes or at level gaps half a level apart, and a note then names the rows that clash. battles_cards rows carry modes (a count per mode group) and mean_level_gap; the response carries modes_in_window (battles and mean level gap per mode group) and comparable, with a note when modes with different matchmaking were pooled. battles_levels monthly_trend points carry actual_win_rate, expected_from_levels, mean_gap, opponent_mean_level, mean_starting_trophies and modal_arena {id, name}; a note fires when the modal arena or the mean starting trophies moved between two points; a new arena_id argument holds the pool fixed; methodology.n says what n counts (the scored player's battles, once) and methodology.adjusts_for says the score adjusts for card levels, never opponent skill. battles_performance carries trophy_floor (floor, arena, source, floored, on_floor_losses, losses_landing_on_floor, ladder_battles, trophy_range) when the window holds ladder battles and the arena's floor is known, with a note when a loss touched it; group_by week rows a window clips carry partial: true and covers {from, to}, with a note naming them. battles_query full verbosity: elixir_leaked on every teammate and opponent, elixir_leaked_differential on me (me minus the one opponent, null on duels, 2v2 and unreported sides), a note that neither is a skill measure, and a note when ladder losses carry trophy_change null (a loss ON the arena's floor). war_current and war_history say decks_used is the race week's cumulative count and a duel consumes one deck per round played. New error code internal for a server failure (it was reported as bad_request); a result_too_large hint says which limit would have fit. battles_opponents with any window failed since 3.11.1 (a SQL alias) and answers again. Additive.