Module stats 

Structs

CepSessionSnapshot

Point-in-time snapshot of CEP scores for a single session.

CepStats

Aggregated CEP (Cognitive Efficiency Protocol) metrics across sessions.

CommandStats

Per-command token statistics: invocation count and input/output totals.

CostBreakdown

Detailed cost comparison: with vs. without lean-ctx compression.

CostModel

LLM pricing model for estimating dollar savings from token compression.

DayStats

Daily aggregate: command count and token totals for one calendar day.

GainSummary

High-level token savings summary for display.

StatsStore

Persistent store for all-time token savings, command stats, and daily history.

Enums

TrafficClass

Whether a recorded command’s output is controlled by compression.

Constants

DEFAULT_INPUT_PRICE_PER_M

Average LLM pricing per 1M tokens (blended across Claude, GPT, Gemini).

DEFAULT_OUTPUT_PRICE_PER_M

Functions

adjust_savings

Adjust saved tokens after post-processing (terse, hints) changed the output size. Positive delta = savings were over-reported, negative = under-reported.

flush

flush_if_due

Debounced flush: persists at most once per FLUSH_INTERVAL_SECS. Call from post_dispatch on every tool call to bound data loss on abrupt MCP termination to at most 2 seconds of events.

format_cep_report

Renders the full CEP (Cognitive Efficiency Protocol) report with themes.

format_gain

format_gain_body

The dashboard body without the trailing footer (tips / Context OS / hints). Used to compose gain --deep, where the extra themed sections must appear before the footer instead of in the middle of the output.

format_gain_daily

Renders a daily breakdown table of token savings with totals.

format_gain_footer

The standalone gain dashboard footer (contextual tip, Context OS, hints).

format_gain_graph

Renders a 30-day token savings bar chart with sparkline.

format_gain_hero

Renders the concise “hero” gain output — 3 key metrics, gain score, trend, next actions.

format_gain_hero_themed

Hero gain with specific theme.

format_gain_json

Returns the full stats store as pretty-printed JSON.

format_gain_themed

Renders the token savings dashboard with a specific theme.

format_gain_themed_at

Renders the token savings dashboard at a specific animation tick (with footer).

gain_live

Runs the live-updating gain dashboard (1s refresh loop, Ctrl+C to exit).

load

load_for_display

Loads stats for display, summing across every auto-resolved data dir that holds a stats.json (#500). When the MCP server process and the CLI resolve different XDG dirs — the documented #408/#414 split, common when an agent host (e.g. a containerised Hermes) launches the MCP server with a different HOME or XDG_* than the user’s shell — the bulk of the savings can land in a sibling tree. Reading only the primary dir then makes gain report 0 while the real data sits one directory over. Folding the siblings in keeps the headline honest regardless of which process wrote where.

load_stats

record

record_at_turn

Records a tool result against an observed provider turn. turn == 0 keeps daemon-free callers honest: first injection is known, re-read count is not.

record_cep_session

record_reread

reset_all

reset_cep

save