Struct DsfbDebugEngine 

pub struct DsfbDebugEngine<const MAX_SIGNALS: usize, const MAX_MOTIFS: usize> { /* private fields */ }

Main DSFB debugging engine — stateless, deterministic, read-only.

The engine evaluates telemetry residuals through the DSFB pipeline and produces typed, auditable debugging episodes.

Const Generic Parameters

• MAX_SIGNALS: maximum number of monitored signals (span durations, error rates, etc.)
• MAX_MOTIFS: maximum heuristics bank entries

Non-Intrusion Contract

All public methods accept only &self and &[T] (shared immutable references). There is NO mutable write path into any upstream data structure. The Rust type system enforces this at compile time.

Implementations

impl<const S: usize, const M: usize> DsfbDebugEngine<S, M>

pub fn new(config: EngineConfig) -> Result<Self>

Create a new engine with the given configuration.

pub fn paper_lock() -> Result<Self>

Create a new engine with paper-lock configuration.

pub fn config(&self) -> &EngineConfig

Get the engine configuration (read-only)

pub fn heuristics_bank(&self) -> &HeuristicsBank<M>

Get the engine’s heuristics bank (read-only). Exposed for the fusion harness’s consensus-aware scoring pass; operators reading this can call bank.match_episode_with_consensus(...) against any closed DebugEpisode directly.

pub fn evaluate_signal( &self, residual_norms: &[f64], k: usize, rho: f64, signal_index: u16, window_index: u64, was_imputed: bool, recent_raw_states: &[GrammarState], persistence_count: usize, ) -> SignalEvaluation

Evaluate a single window of telemetry for a single signal.

Non-Intrusion Contract

All inputs are shared immutable references. Cannot modify upstream data.

Arguments

• residual_norms - historical residual norms for this signal (immutable)
• k - current window index into the norms array
• rho - envelope radius for this signal
• signal_index - index of this signal
• window_index - global window index
• was_imputed - whether this observation was imputed (missing data)
• recent_raw_states - last n_confirm raw grammar states
• persistence_count - consecutive Boundary windows

pub fn run_evaluation( &self, data: &[f64], num_signals: usize, num_windows: usize, fault_labels: &[bool], healthy_window_end: usize, eval_out: &mut [SignalEvaluation], episodes_out: &mut [DebugEpisode], dataset_name: &'static str, ) -> Result<(usize, BenchmarkMetrics)>

Run the full DSFB evaluation pipeline over a dataset.

This is the main entry point for benchmark evaluation.

Non-Intrusion Contract

All input slices are shared immutable references. Outputs are written into caller-owned mutable buffers.

Arguments

• data - row-major observation data [window][signal] (immutable)
• num_signals - signals per window
• num_windows - total windows
• fault_labels - per-window fault labels (immutable)
• healthy_window_end - index of last healthy window for baseline
• eval_out - output buffer for per-signal evaluations (row-major)
• episodes_out - output buffer for episodes
• dataset_name - name for metrics reporting

Returns

(episode_count, BenchmarkMetrics)

pub fn run_evaluation_with_graph( &self, data: &[f64], num_signals: usize, num_windows: usize, fault_labels: &[bool], healthy_window_end: usize, eval_out: &mut [SignalEvaluation], episodes_out: &mut [DebugEpisode], dataset_name: &'static str, service_graph: &[(u16, u16)], ) -> Result<(usize, BenchmarkMetrics)>

run_evaluation plus graph-attribution: same return value, but each closed episode’s root_cause_signal_index is populated by walking the supplied service-call graph (see crate::causality).

Backward-compatible with v0.1: callers without a graph use run_evaluation and get root_cause_signal_index = None on every episode.

pub fn verify_deterministic_replay( &self, data: &[f64], num_signals: usize, num_windows: usize, fault_labels: &[bool], healthy_window_end: usize, ) -> Result<bool>

Deterministic replay verification (Theorem 9 proof-by-construction).

Runs the evaluation twice on identical inputs and verifies identical outputs.

impl DsfbDebugEngine<256, 64>

pub fn default_size() -> Result<Self>

Create with default const generics (256 signals, 64 motifs).

The 64-slot heuristics bank holds 29 canonical motifs as of v0.2 (Session 3 expansion); the remaining 35 slots provide v0.3 / v0.4 headroom for additional site-specific findings.