Struct PipelineEvent 

#[non_exhaustive]
pub struct PipelineEvent {

    pub session_hash: String,
    pub tool_call_id_hash: String,
    pub tool_name_anon: String,
    pub endpoint_class: String,
    pub response_chars: u64,
    pub shape: Shape,
    pub inner_formats: Vec<String>,
    pub content_sha_prefix_hex: String,
    pub file_path_hash: Option<String>,
    pub is_dedup_hit: bool,
    pub layer_used: Layer,
    pub template_id: Option<String>,
    pub tokens_baseline: u32,
    pub tokens_final: u32,
    pub context_partition: u32,
    pub is_sidechain: bool,
    pub ts_ms: i64,
    pub sample_rate_applied: f32,
    pub enricher_prefetched: bool,
    pub enricher_predicted_cost_tokens: u32,
    pub enricher_decline_reason: Option<String>,
    pub cited_in_next_n_turns: Option<bool>,
}

Single pipeline decision — emitted once per tool-result.

See docs/research/paper-2-mckp-format-adaptive.md §Telemetry & Observability for field-level documentation.

Fields (Non-exhaustive)

Non-exhaustive structs could have additional fields added in future. Therefore, non-exhaustive structs cannot be constructed in external crates using the traditional Struct { .. } syntax; cannot be matched against without a wildcard ..; and struct update syntax will not work.

session_hash: String

SHA-256 prefix of the session UUID (not the raw UUID).

tool_call_id_hash: String

SHA-256 prefix of this tool_use_id — identifies the response for dedup references without exposing the raw id.

tool_name_anon: String

Anonymized tool name. MCP slugs are hashed (mcp__p<hash>__verb).

endpoint_class: String

Coarse endpoint classification (e.g. git_log, curl, or the full tool_name for single-endpoint tools).

response_chars: u64

Raw byte count of the response.

shape: Shape

Structural shape of the response.

inner_formats: Vec<String>

Names of embedded formats detected inside the response (e.g. diff, log, url, hash). Empty when no embedded formats were seen.

content_sha_prefix_hex: String

Hex-encoded prefix of SHA-256 over the response bytes — 32 hex chars representing the first 16 bytes (128 bits), matching the paper’s stated fingerprint width and the Python extractor’s output.

file_path_hash: Option<String>

Anonymized file-path hash for Read/Edit/Write-family tools; None for tools that don’t operate on a file path.

is_dedup_hit: bool

Did the pipeline emit a dedup hint in lieu of full content?

layer_used: Layer

Terminal layer in the pipeline decision tree.

template_id: Option<String>

L1 template identifier if layer_used == L1.

tokens_baseline: u32

Token count before pipeline encoding (baseline).

tokens_final: u32

Token count of what the pipeline emitted.

context_partition: u32

Monotonic partition counter; increments on each compaction boundary.

is_sidechain: bool

True for subagent (sidechain) tool-results; false for main session.

ts_ms: i64

Unix milliseconds at which the response was produced.

sample_rate_applied: f32

Fraction of events kept when sampling is enabled. 1.0 when every event is recorded. Consumers of telemetry must scale counts by 1 / sample_rate_applied.

enricher_prefetched: bool

True when the planner pre-fetched this tool call (rather than the LLM emitting it directly). Drives the prefetch_hit_rate metric — paired with cited_in_next_n_turns once the post-pass scanner runs.

enricher_predicted_cost_tokens: u32

cost_model.typical_kb-derived prediction (in tokens) at the moment the planner admitted the call. Compared with tokens_baseline to compute cost_overrun_rate.

enricher_decline_reason: Option<String>

Set on declined candidates that the host emitted anyway as a telemetry-only event (so tune analyze can study what the planner skipped). One of "budget" / "low_probability" / "preempted" / "prereq_missing".

cited_in_next_n_turns: Option<bool>

Reserved for an offline citation-enrichment post-pass that re-reads the JSONL log and sets this to true when the next 1–3 LLM messages textually reference any of the response’s content_sha_prefix_hex bytes. The live pipeline never sets this; the post-pass is not shipped yet (the existing tune from-claude-logs --tools only seeds [tools.*] defaults, it does not populate citation fields). Stays None until that pass lands.

Trait Implementations

impl Clone for PipelineEvent

fn clone(&self) -> PipelineEvent

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for PipelineEvent

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl Default for PipelineEvent

fn default() -> PipelineEvent

Returns the “default value” for a type.

impl<'de> Deserialize<'de> for PipelineEvent

fn deserialize<__D>(__deserializer: __D) -> Result<Self, __D::Error>

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl Serialize for PipelineEvent

fn serialize<__S>(&self, __serializer: __S) -> Result<__S::Ok, __S::Error>

where __S: Serializer,

Serialize this value into the given Serde serializer.