cctrack 0.5.0__tar.gz → 0.6.0__tar.gz

{cctrack-0.5.0 → cctrack-0.6.0}/CLAUDE.md

@@ -31,9 +31,11 @@ uv build
 
 This is a single-module CLI tool. All logic lives in `src/cctrack/__init__.py`:
 
-1. **Rate card** (`RATES`, `get_rates`, `calculate_cost`, `is_known_model`) - Model-to-price mapping using prefix matching. Order matters: longer/more-specific prefixes must come first (the legacy `claude-opus-4` 15/75 entry must stay below every `claude-opus-4-x`). Unknown models fall back to Sonnet rates and are flagged with `?` in the report.
+1. **Rate card** (`RATES`, `get_rates`, `calculate_cost`, `is_known_model`) - Model-to-price mapping using prefix matching. Order matters: longer/more-specific prefixes must come first (the legacy `claude-opus-4` 15/75 entry must stay below every `claude-opus-4-x`). Unknown models fall back to Sonnet rates and are flagged with `?` in the report. `get_rates`/`calculate_cost` take the usage date so time-limited `intro` pricing (e.g. Sonnet 5 through 2026-08-31) applies to past usage correctly.
+1a. **Energy model** (`ENERGY_WH_PER_MTOK`, `estimate_energy_wh`) - Opt-in (`--energy`) order-of-magnitude estimate keyed off each rate entry's `tier`. Deliberately NOT a function of price: both cache-write TTLs use one figure because a 1h write costs more money but the same compute.
 2. **File discovery** (`default_dirs`, `discover_files`) - Walks `~/.claude/projects/` and `~/.sandy/sandboxes/` for `.jsonl` files.
 3. **Remote aggregation** (`fetch_remote_jsonl`, `parse_remote_events`, `fetch_remote_hook_data`, `parse_remote_hook_data`) - SSHs to hosts, cats all JSONL logs plus statusline hook data (statusline files, monthly summaries, session baselines) with file separators, parses locally.
+3a. **Hook parsing** (`parse_hook_rows`, `compute_hook_deltas`, `_cost_increment`) - The hook is authoritative for **cost only**. Its `input_tokens`/`output_tokens`/`cache_*` fields are context-window SNAPSHOTS (`input_tokens` ≈ `cache_read + cache_write`; `output_tokens` is just the last message's), not cumulative — never use them for token counts. `cost_usd` is a per-session running total, but it RESTARTS on resume/compaction and a render can report 0, so deltas walk every row chronologically and treat a meaningful drop as a restart.
 4. **JSONL parsing** (`parse_lines`, `parse_events`, `dedup_events`) - Extracts `assistant` events with token usage. Deduplicates by `requestId` across all of a host's files (last event wins) — resumed sessions (`--resume`/`--continue`) repeat events with the same `requestId` in new files. Events without `requestId` are never deduped.
 5. **Aggregation** (`aggregate`) - Buckets events into daily and monthly summaries with token counts and costs.
 6. **Output** (`print_report`) - Prints current month-to-date summary with projections, daily breakdown, and previous month summaries. Cost is composed **per source** (one source per machine): hook cost where a source has statusline coverage, accuracy-factor-adjusted JSONL cost otherwise, summed per date. The accuracy factor is measured per source over dates strictly after that source's hook install day.
@@ -44,7 +46,9 @@ Entry point: `cctrack:main` (registered in `pyproject.toml` as a console script)
 
 - Zero external dependencies - stdlib only (json, argparse, subprocess, etc.)
 - Deduplication is global per host by `requestId` (last wins) - resumed sessions replay history into new files and must not double count. Cross-host `requestId` collisions can't occur (distinct API calls).
-- Hook data is authoritative for cost where present; JSONL-derived cost is an estimate. Never let one machine's hook coverage suppress another machine's JSONL cost - compose per source.
+- Hook data is authoritative for **cost only**; every token count comes from the JSONL logs. Never mix the two in one figure - the hook's token fields are snapshots and would double count against JSONL cache columns.
+- Never let one machine's hook coverage suppress another machine's JSONL cost - compose per source.
+- Cumulative counters from the hook are not monotonic. Any code differencing `cost_usd` must handle restarts (resume/compaction) and zero-valued renders, or whole days silently vanish.
 - `session-baselines.json` carries each session's rolled-up cumulative across the 90-day retention boundary so straddling sessions aren't re-counted from zero.
 - Build system is Hatch (`hatchling`)
 - Requires Python 3.10+

{cctrack-0.5.0 → cctrack-0.6.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: cctrack
-Version: 0.5.0
+Version: 0.6.0
 Summary: Claude Code cost tracker — scans JSONL logs and reports token usage + estimated cost
 Project-URL: Homepage, https://github.com/rappdw/cctrack
 Project-URL: Repository, https://github.com/rappdw/cctrack
@@ -98,19 +98,39 @@ cctrack --install-hook --remote dgx macbook-air
 
 ## Rate card
 
-Prices per million tokens, from [Anthropic's pricing page](https://www.anthropic.com/pricing). Cache writes are priced per TTL: 5-minute at 1.25× input, 1-hour at 2× input.
+Prices per million tokens, from [Anthropic's pricing page](https://platform.claude.com/docs/en/about-claude/pricing). Cache writes are priced per TTL: 5-minute at 1.25× input, 1-hour at 2× input.
 
 | Model | Input | Output | Cache Read | Cache Write 5m | Cache Write 1h |
 |---|---|---|---|---|---|
 | Fable 5 / Mythos 5 | $10.00 | $50.00 | $1.00 | $12.50 | $20.00 |
-| Opus 4.8 / 4.7 / 4.6 / 4.5 | $5.00 | $25.00 | $0.50 | $6.25 | $10.00 |
+| Opus 5 / 4.8 / 4.7 / 4.6 / 4.5 | $5.00 | $25.00 | $0.50 | $6.25 | $10.00 |
 | Opus 4 / 4.1 | $15.00 | $75.00 | $1.50 | $18.75 | $30.00 |
-| Sonnet 5 / 4.x / 3.7 | $3.00 | $15.00 | $0.30 | $3.75 | $6.00 |
+| Sonnet 5 (through 2026-08-31) | $2.00 | $10.00 | $0.20 | $2.50 | $4.00 |
+| Sonnet 5 (from 2026-09-01) / 4.x / 3.7 | $3.00 | $15.00 | $0.30 | $3.75 | $6.00 |
 | Haiku 4.x | $1.00 | $5.00 | $0.10 | $1.25 | $2.00 |
 | Haiku 3.5 | $0.80 | $4.00 | $0.08 | $1.00 | $1.60 |
 | Haiku 3 | $0.25 | $1.25 | $0.03 | $0.30 | $0.50 |
 
-Unknown models fall back to Sonnet rates and are flagged with `?` in the report's Model column.
+Usage is priced with the rates in effect on the day it happened, so time-limited introductory pricing is applied correctly to past usage.
+
+Unknown models fall back to Sonnet rates and are flagged with `?` in the report's Model column — a prompt to update the rate card rather than a silent guess.
+
+## Energy estimate
+
+`cctrack --energy` adds an estimated inference energy column (Wh/kWh):
+
+```bash
+cctrack --energy
+```
+
+This is a **rough estimate, not a measurement** — Anthropic does not publish per-token energy figures or model sizes. It is modelled from logged tokens using the published full-stack per-prompt figures the major labs have disclosed (Google: 0.24 Wh median Gemini text prompt; OpenAI: ~0.34 Wh average ChatGPT query — both include accelerator, host, cooling and idle capacity), apportioned across token classes by relative compute cost and scaled by model tier:
+
+- **output tokens dominate** — decode is sequential and memory-bandwidth bound
+- **input and cache writes** are prefill: parallel, far cheaper per token
+- **cache reads** skip prefill entirely (~0.1× input)
+- both cache-write TTLs use the same figure — a 1-hour write costs more money but the same compute
+
+Treat results as ±1 order of magnitude. It covers datacenter inference only: not model training, and not your own machine. Scale the whole model with `CCTRACK_ENERGY_SCALE=1.5` if you have better numbers for your workload.
 
 ## Origins
 

{cctrack-0.5.0 → cctrack-0.6.0}/README.md

@@ -73,19 +73,39 @@ cctrack --install-hook --remote dgx macbook-air
 
 ## Rate card
 
-Prices per million tokens, from [Anthropic's pricing page](https://www.anthropic.com/pricing). Cache writes are priced per TTL: 5-minute at 1.25× input, 1-hour at 2× input.
+Prices per million tokens, from [Anthropic's pricing page](https://platform.claude.com/docs/en/about-claude/pricing). Cache writes are priced per TTL: 5-minute at 1.25× input, 1-hour at 2× input.
 
 | Model | Input | Output | Cache Read | Cache Write 5m | Cache Write 1h |
 |---|---|---|---|---|---|
 | Fable 5 / Mythos 5 | $10.00 | $50.00 | $1.00 | $12.50 | $20.00 |
-| Opus 4.8 / 4.7 / 4.6 / 4.5 | $5.00 | $25.00 | $0.50 | $6.25 | $10.00 |
+| Opus 5 / 4.8 / 4.7 / 4.6 / 4.5 | $5.00 | $25.00 | $0.50 | $6.25 | $10.00 |
 | Opus 4 / 4.1 | $15.00 | $75.00 | $1.50 | $18.75 | $30.00 |
-| Sonnet 5 / 4.x / 3.7 | $3.00 | $15.00 | $0.30 | $3.75 | $6.00 |
+| Sonnet 5 (through 2026-08-31) | $2.00 | $10.00 | $0.20 | $2.50 | $4.00 |
+| Sonnet 5 (from 2026-09-01) / 4.x / 3.7 | $3.00 | $15.00 | $0.30 | $3.75 | $6.00 |
 | Haiku 4.x | $1.00 | $5.00 | $0.10 | $1.25 | $2.00 |
 | Haiku 3.5 | $0.80 | $4.00 | $0.08 | $1.00 | $1.60 |
 | Haiku 3 | $0.25 | $1.25 | $0.03 | $0.30 | $0.50 |
 
-Unknown models fall back to Sonnet rates and are flagged with `?` in the report's Model column.
+Usage is priced with the rates in effect on the day it happened, so time-limited introductory pricing is applied correctly to past usage.
+
+Unknown models fall back to Sonnet rates and are flagged with `?` in the report's Model column — a prompt to update the rate card rather than a silent guess.
+
+## Energy estimate
+
+`cctrack --energy` adds an estimated inference energy column (Wh/kWh):
+
+```bash
+cctrack --energy
+```
+
+This is a **rough estimate, not a measurement** — Anthropic does not publish per-token energy figures or model sizes. It is modelled from logged tokens using the published full-stack per-prompt figures the major labs have disclosed (Google: 0.24 Wh median Gemini text prompt; OpenAI: ~0.34 Wh average ChatGPT query — both include accelerator, host, cooling and idle capacity), apportioned across token classes by relative compute cost and scaled by model tier:
+
+- **output tokens dominate** — decode is sequential and memory-bandwidth bound
+- **input and cache writes** are prefill: parallel, far cheaper per token
+- **cache reads** skip prefill entirely (~0.1× input)
+- both cache-write TTLs use the same figure — a 1-hour write costs more money but the same compute
+
+Treat results as ±1 order of magnitude. It covers datacenter inference only: not model training, and not your own machine. Scale the whole model with `CCTRACK_ENERGY_SCALE=1.5` if you have better numbers for your workload.
 
 ## Origins
 

{cctrack-0.5.0 → cctrack-0.6.0}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "cctrack"
-version = "0.5.0"
+version = "0.6.0"
 description = "Claude Code cost tracker — scans JSONL logs and reports token usage + estimated cost"
 readme = "README.md"
 license = "MIT"