Struct EncoderTrace 

pub struct EncoderTrace {

    pub term_substitutions: u32,
    pub if_prefix: u32,
    pub success: u32,
    pub failure: u32,
    pub arrow: u32,
    pub vs: u32,
    pub and: u32,
    pub filler_removed: u32,
    pub ansi_stripped: u32,
    pub trailing_ws: u32,
    pub blank_lines: u32,
    pub json_minified: u32,
    pub json_records_table: u32,
    pub numeric_range_lines: u32,
    pub repeated_chunk_dict: u32,
    pub repeated_lines: u32,
    pub tool_schema_semantic_table: u32,
    pub bytes_saved: [u64; 17],
}

Per-rule fire counts emitted alongside an encode pass.

Each field counts how many match instances the rule produced on this input. A non-firing rule reports 0; rules with no semantic effect on a given input are useful exactly because they cost nothing to evaluate. Use EncoderTrace::any_fired to gate telemetry on whether the encoder did real work.

Fields

term_substitutions: u32

Total term-substitution match count across the 35-pair table.

if_prefix: u32

Number of if prefix removals.

success: u32

Success-glyph (\u{2713}) substitutions.

failure: u32

Failure-glyph (\u{2717}) substitutions.

arrow: u32

Arrow-glyph (\u{2192}) substitutions.

vs: u32

vs substitutions.

and: u32

+ substitutions for and/plus.

filler_removed: u32

Number of filler words removed.

ansi_stripped: u32

Number of ANSI-escape sequences stripped from tool_result content blocks. Applied shim-side before the encoder runs to claw back tokens burned on terminal-color bytes the LLM never needs to see. Lossless for model consumption.

trailing_ws: u32

Number of lines where trailing [ \t]+ was removed from tool_result content. Pure-waste whitespace that tokenizes into real input-billed tokens; LLMs ignore trailing spaces semantically. Lossless for model consumption.

blank_lines: u32

Number of excess blank lines collapsed in tool_result content. Runs of 3+ consecutive \n are squeezed to exactly two (one blank line preserved as a paragraph break). Preserves semantic section separation while dropping layout-only padding.

json_minified: u32

Number of JSON string payloads minified inside tool_result content. Lossless for valid JSON: only insignificant whitespace is removed.

json_records_table: u32

Number of homogeneous JSON record arrays rendered as compact TOON-like tables inside tool_result content. This preserves field names and scalar values but is no longer JSON syntax, so it is canary-only by default.

numeric_range_lines: u32

Number of numeric line ranges compacted in tool_result content. This is deterministic for lines that differ only by an incrementing number.

repeated_chunk_dict: u32

Number of repeated chunk dictionaries emitted in tool_result content. Repeated log lines/sentences are represented once plus an explicit sequence, preserving reconstruction information.

repeated_lines: u32

Number of consecutive repeated-line runs collapsed in tool_result content. Repetition count is preserved as xN, so log volume is compacted without hiding that repetition occurred.

tool_schema_semantic_table: u32

Number of simple tool-schema payloads rendered as compact semantic tables. Preserves tool name, description, required parameters, and property schemas for the supported simple schema subset; canary-only by default.

bytes_saved: [u64; 17]

Per-rule byte-savings attribution, indexed identically to the pairs returned by as_pairs (alphabetical by rule name).

PLAN_2026-04-24.md Step 9 substrate. Each entry is the UTF-8 byte count a rule removed from the payload on THIS encoder pass — pre-rule body length minus post-rule body length. A non-firing rule contributes 0. Rules that never fire on a given input contribute 0, so zero-valued entries are signal, not noise.

Default is an all-zero array (matches the default-non-firing trace). The actual per-rule measurement hooks inside encode_symbolic_traced_with will land in a follow-up commit so this schema can deploy without the rule-pass refactor. Until then, callers SHOULD treat any bytes_saved they read as “may be zero because the rule did not run OR because the measurement hook is not yet wired” — any_fired() on the same trace is the disambiguator.

Implementations

impl EncoderTrace

pub const IDX_AND: usize = 0

Positional index of each rule in bytes_saved / as_pairs. Alphabetically ordered, so adding a new rule between existing ones shifts subsequent indices — readers MUST use these constants instead of hardcoded integers. Step-9 scaffold.

pub const IDX_ANSI_STRIPPED: usize = 1

pub const IDX_ARROW: usize = 2

pub const IDX_BLANK_LINES: usize = 3

pub const IDX_FAILURE: usize = 4

pub const IDX_FILLER_REMOVED: usize = 5

pub const IDX_IF_PREFIX: usize = 6

pub const IDX_JSON_MINIFIED: usize = 7

pub const IDX_JSON_RECORDS_TABLE: usize = 8

pub const IDX_NUMERIC_RANGE_LINES: usize = 9

pub const IDX_REPEATED_CHUNK_DICT: usize = 10

pub const IDX_REPEATED_LINES: usize = 11

pub const IDX_SUCCESS: usize = 12

pub const IDX_TERM_SUBSTITUTIONS: usize = 13

pub const IDX_TOOL_SCHEMA_SEMANTIC_TABLE: usize = 14

pub const IDX_TRAILING_WS: usize = 15

pub const IDX_VS: usize = 16

pub fn any_fired(&self) -> bool

True if any rule produced at least one substitution or removal.

pub fn bytes_saved_pairs(&self) -> [(&'static str, u64); 17]

Render byte-savings attribution as a stable, sorted list of (rule_name, bytes_saved) pairs. Positional order matches as_pairs; non-firing rules contribute 0. Added with Step 9.

Pairs with as_pairs for post-hoc analysis: counts tell you HOW OFTEN each rule fired, bytes_saved tells you WHICH BYTES each rule removed. Combined they let the bandit attribute savings to specific rules without re-running the encoder.

pub fn as_pairs(&self) -> [(&'static str, u32); 17]

Render as a stable, sorted list of (rule_name, count) pairs. Used by pithy-cli analyze and the audit ledger so a future rule rename does not break replay.

pub fn merge(&mut self, other: EncoderTrace)

Sum-merge fire counts from other into self. Used by the structural encoder to aggregate traces across compressed paragraph segments.

Trait Implementations

impl Clone for EncoderTrace

fn clone(&self) -> EncoderTrace

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for EncoderTrace

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

Formats the value using the given formatter.

impl Default for EncoderTrace

fn default() -> EncoderTrace

Returns the “default value” for a type.

impl PartialEq for EncoderTrace

fn eq(&self, other: &EncoderTrace) -> bool

Tests for self and other values to be equal, and is used by ==.

fn ne(&self, other: &Rhs) -> bool

Tests for !=. The default implementation is almost always sufficient, and should not be overridden without very good reason.

impl Eq for EncoderTrace

impl StructuralPartialEq for EncoderTrace