---@meta
---@class AlcResultAssessed
---@field answer string
---@field confidence number @Self-assessed confidence 0.0–1.0
---@field total_llm_calls number
---@class AlcResultCalibrated
---@field answer string
---@field confidence number @Initial self-assessed confidence
---@field escalated boolean @Whether fallback was triggered
---@field fallback_detail? table @Fallback strategy result (voted/paneled)
---@field strategy "direct"|"retry"|"panel"|"ensemble"
---@field total_llm_calls number
---@class AlcResultConformalDecided
---@field action "commit"|"escalate"|"anomaly" @Three-way decision per Proposition 3
---@field card_id? string @Emitted Card id (only when auto_card=true)
---@field coverage_level number @1 - alpha (finite-sample guarantee)
---@field p_social table<string, number> @Linear opinion pool output { [label] = prob }
---@field prediction_set string[] @Labels y with P_social(y|x) >= tau
---@field q_hat number @Calibration quantile of nonconformity scores
---@field selected? string @Committed label (nil when action != 'commit')
---@field tau number @1 - q_hat (prediction-set threshold)
---@class AlcResultDeepPaneled
---@field abort_reason? string @Abort reason (nil when not aborted)
---@field aborted boolean @True if early-abort triggered
---@field answer any @Plurality answer (nil on abort)
---@field anti_jury boolean @Condorcet anti-jury detection at Stage 1
---@field branches table @{ [bkey] = { approach, answer, best_score, tree_stats } }
---@field confidence number @Meta-confidence estimate
---@field decomp? table @ensemble_div.decompose output (nil if Stage 3b skipped)
---@field diversity? table @{ n_distinct, distinctness, decomp_status }
---@field expected_accuracy number @Condorcet expected majority accuracy
---@field margin_gap number @(top - runner_up) / n
---@field n_branches_completed number @Branches actually finished
---@field n_distinct_answers number @Count of unique normalized answers
---@field needs_investigation boolean @True if meta-confidence below threshold
---@field panel_size number @Requested panel size
---@field plurality_fraction number @Top-answer vote fraction
---@field stages table[] @Per-stage detail (heterogeneous)
---@field target_met boolean @Whether expected accuracy >= ctx.target_accuracy
---@field total_llm_calls number
---@field unanimous boolean @All normalized votes identical
---@field vote_counts table<string, number> @{ [normalized_answer] = count } tally
---@class AlcResultDeliberated
---@field answer string @Selected option's final answer text
---@field card_id? string @Emitted Card id (only when auto_card=true)
---@field convergence "dominance"|"no_blocking"|"fallback" @How the session converged
---@field decision_packet { minority_report: { confidence: number, position: string, rationale: string }[], next_actions: string[], reopen_triggers: string[], residual_objections: string[], selected_option: { answer: string, evidence: string[], rationale: string } } @5-component decision packet; all 5 fields MUST be non-nil
---@field history table[] @Per-stage typed-act log (14-act typed)
---@field stats { options_count: number, rounds_used: number, total_acts: number, total_llm_calls: number } @Execution statistics
---@field workspace { emerging_ideas: string[], key_frames: string[], next_actions: string[], problem_view: string, synthesis_in_progress: string, tensions: string[] } @Shared workspace 6 fields after finalization
---@class AlcResultFunnelRanked
---@field best string @Top-ranked text
---@field best_index number @Top-ranked original index (1-based)
---@field bypass_reason? string @Reason for bypass (nil when not bypassed)
---@field funnel_bypassed boolean @True when N < 6 bypasses funnel stages
---@field funnel_shape number[] @Candidate counts per stage [N, s1_out, s2_out]
---@field naive_baseline_calls number @Hypothetical full-pairwise call count
---@field naive_baseline_kind string @Baseline method identifier
---@field ranking { original_index: number, pairwise_score: number, rank: number, text: string }[] @Final ranking
---@field savings_percent? number @LLM call savings vs baseline (nil on bypass)
---@field stages ({ both_tie_pairs: number, input_count: number, llm_calls: number, name: "direct_pairwise", output_count: number, position_bias_splits: number }|{ input_count: number, llm_calls: number, name: "listwise_rank", output_count: number, window_size: number }|{ input_count: number, llm_calls: number, name: "listwise_skipped", output_count: number, reason: string }|{ axes_count: number, input_count: number, llm_calls: number, name: "multi_axis_scoring", output_count: number, parse_failures: number, score_range: { max: number, min: number } }|{ both_tie_pairs: number, input_count: number, llm_calls: number, name: "pairwise_rank_allpair", output_count: number, position_bias_splits: number }|{ input_count: number, llm_calls: number, name: "scoring_skipped", output_count: number, reason: string })[] @Per-stage detail (discriminated by name)
---@field total_llm_calls number
---@field warnings { code: string, data: table, message: string, severity: "warn"|"critical" }[] @Diagnostic warnings
---@class AlcResultListwiseRanked
---@field best string @Top-ranked text
---@field best_index number @Top-ranked original index (1-based)
---@field killed { index: number, rank: number, text: string }[] @Eliminated candidates
---@field n_candidates number
---@field ranked { index: number, rank: number, text: string }[] @Full ranking
---@field top_k { index: number, rank: number, text: string }[] @Top-k subset
---@field total_llm_calls number
---@class AlcResultPairwiseRanked
---@field best string @Top-ranked text
---@field best_index number @Top-ranked original index (1-based)
---@field both_tie_pairs number @Pairs that tied in both directions
---@field killed { index: number, rank: number, score: number, text: string }[] @Eliminated candidates
---@field method "allpair"|"sorting" @Comparison strategy
---@field n_candidates number
---@field position_bias_splits number @Position-bias correction splits
---@field ranked { index: number, rank: number, score: number, text: string }[] @Full ranking with scores
---@field score_semantics "copeland"|"rank_inverse" @Score interpretation
---@field top_k { index: number, rank: number, score: number, text: string }[] @Top-k subset
---@field total_llm_calls number
---@class AlcResultPaneled
---@field arguments { role: string, text: string }[] @Per-role position statements
---@field synthesis string @Moderator synthesis
---@class AlcResultQuickVoted
---@field answer string @Leader answer from sample 1 (cleaned, not normalized)
---@field leader_norm string @Normalized leader key used for agreement tests
---@field n_samples number @Total samples drawn (1 leader + k agreement observations)
---@field needs_investigation boolean @True only when outcome == 'truncated' (evidence inconclusive at declared α/β). 'rejected' is a conclusive verdict and does NOT set this flag.
---@field outcome "confirmed"|"rejected"|"truncated" @Terminal state: confirmed=H1 accepted, rejected=H0 accepted, truncated=no verdict at max_n
---@field params { alpha: number, beta: number, max_n: number, min_n: number, p0: number, p1: number } @Echoed parameter values
---@field samples { answer: string, norm: string, reasoning: string }[] @Per-sample reasoning + extracted answer
---@field sprt { a_bound: number, b_bound: number, log_lr: number, n: number } @Final SPRT state snapshot
---@field total_llm_calls number @2 × n_samples (reasoning + extract per sample)
---@field verdict "accept_h1"|"accept_h0"|"continue" @Underlying SPRT verdict from the final decide()
---@field vote_counts table<string, number> @{ [norm] = count } tally across all samples
---@class AlcResultSafePaneled
---@field abort_reason? string @Abort reason (nil when not aborted)
---@field aborted boolean @True if early-abort triggered
---@field answer? string @Consensus answer (nil on abort)
---@field anti_jury boolean @Condorcet anti-jury detection
---@field confidence number @Meta-confidence estimate
---@field expected_accuracy number @Condorcet expected majority accuracy
---@field is_safe boolean @Vote-prefix stability safe flag
---@field margin_gap number @(top - runner_up) / n
---@field n_distinct_answers number @Count of unique answers
---@field needs_investigation boolean @True if meta-confidence below threshold
---@field panel_size number @Actual panel size used
---@field plurality_fraction number @Top-answer vote fraction
---@field stages ({ confidence: number, escalated: boolean, llm_calls: number, name: "calibrate", needs_investigation: boolean, threshold: number }|{ expected_accuracy: number, name: "condorcet", p_estimate: number, recommended_n: number, target_accuracy: number, target_met: boolean }|{ is_anti_jury: boolean, name: "condorcet_anti_jury", p_estimate: number }|{ is_anti_jury: boolean, name: "condorcet_coin_flip", p_estimate: number }|{ answer: string, llm_calls: number, margin_gap: number, n_distinct_answers: number, name: "sc", panel_size: number, plurality_fraction: number, unanimous: boolean }|{ consecutive_drops: number, is_safe: boolean, name: "vote_prefix_stability", peak_idx: number, series_length: number, signal_type: string }|{ is_safe: boolean, name: "vote_prefix_stability_skipped", reason: string, series_length: number, signal_type: string })[] @Per-stage detail (discriminated by name)
---@field target_met boolean @Whether expected accuracy >= target
---@field total_llm_calls number
---@field unanimous boolean @All votes identical
---@field vote_counts table<string, number> @{ [normalized_answer] = count } tally
---@class AlcResultSmcSampled
---@field answer string @Argmax-weight particle's answer text
---@field card_id? string @Emitted Card id (only when auto_card=true)
---@field ess_trace number[] @ESS recorded at the start of each iteration (length K)
---@field iterations number @K SMC rounds actually executed
---@field particles { answer: string, history: table[], reward: number, weight: number }[] @All N particles in their final state
---@field resample_count number @Number of iterations that triggered multinomial resample
---@field stats { total_llm_calls: number, total_reward_calls: number } @Execution counters (open for diagnostics like mh_rejected)
---@field weights number[] @Final normalized weights (Σ ≈ 1)
---@class AlcResultTournament
---@field best string @Winner text
---@field best_index number @Winner original index (1-based)
---@field candidates string[] @Input candidate texts
---@field matches { a: number, b: number, reason: string, winner: number }[] @Pairwise match log
---@field total_wins number @Winner's win count
---@class AlcResultVoted
---@field answer? string @Majority answer (nil when no paths converge)
---@field answer_norm? string @Normalized vote key
---@field consensus string @LLM-synthesized majority summary
---@field n_sampled number @Number of sampled paths
---@field paths { answer: string, reasoning: string }[] @Per-path reasoning + extracted answer
---@field total_llm_calls number
---@field vote_counts table<string, number> @{ [norm] = count } tally
---@field votes string[] @Normalized vote per path, 1-indexed