adk-tracegauge 0.1.0__tar.gz

adk_tracegauge-0.1.0/LICENSE

@@ -0,0 +1,202 @@
+
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the
+      purposes of this License, Derivative Works shall not include works
+      that remain separable from, or merely link (or bind by name) to the
+      interfaces of, the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including the
+      original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright [yyyy] [name of copyright owner]
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.

adk_tracegauge-0.1.0/PKG-INFO

@@ -0,0 +1,108 @@
+Metadata-Version: 2.4
+Name: adk-tracegauge
+Version: 0.1.0
+Summary: Per-invocation cost-in-USD evaluator for Google's Agent Development Kit (ADK), built on tracegauge's cost engine. Raw dollars and tokens, no calibrated bands, no fabricated numbers for unknown models.
+Author-email: Gaurav Gandhi <gaurav.gandhi2411@gmail.com>
+License-Expression: Apache-2.0
+Project-URL: Repository, https://github.com/gaurav-gandhi-2411/adk-tracegauge
+Keywords: adk,agent-development-kit,google-adk,evaluation,cost,token-efficiency
+Classifier: Development Status :: 3 - Alpha
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3.10
+Classifier: Programming Language :: Python :: 3.11
+Classifier: Programming Language :: Python :: 3.12
+Classifier: Programming Language :: Python :: 3.13
+Classifier: Topic :: Software Development :: Quality Assurance
+Classifier: Topic :: Utilities
+Classifier: Intended Audience :: Developers
+Classifier: Environment :: Console
+Classifier: Operating System :: OS Independent
+Requires-Python: >=3.10
+Description-Content-Type: text/markdown
+License-File: LICENSE
+Requires-Dist: google-adk[eval]<2.7.0,>=2.6.0
+Requires-Dist: tracegauge<0.11.0,>=0.10.0
+Dynamic: license-file
+
+# adk-tracegauge
+
+Per-invocation dollar cost for [Google ADK](https://github.com/google/adk-python) agent evaluations, built on [tracegauge](https://github.com/gaurav-gandhi-2411/token-efficiency-scorer)'s cost engine. ADK's own `evaluation` package reports zero cost or token-efficiency data on any built-in metric — this fills exactly that gap, nothing more.
+
+[![License: Apache-2.0](https://img.shields.io/badge/License-Apache--2.0-blue.svg)](LICENSE)
+[![Python 3.10+](https://img.shields.io/badge/python-3.10%2B-blue.svg)](https://pypi.org/project/adk-tracegauge/)
+
+---
+
+## Required: wrap your agent in an App with the plugin
+
+**This is a hard requirement, not a caveat.** ADK's `Invocation` objects — what an evaluator normally receives — never carry token usage or model identity; ADK strips that down to author+content before an evaluator ever sees it. The real data lives on the `LlmResponse` passed to `BasePlugin.after_model_callback` during inference. That part fires for any real ADK run, App or not — but ADK's **eval harness specifically** (`LocalEvalService`, which `AgentEvaluator` and the `adk eval` CLI both sit on top of) only pulls plugins from `app.plugins` when you pass it an `App`. Evaluate against a bare `root_agent` with no `App`, and `LocalEvalService` builds its Runner with only its own internal eval plugins — **your plugins, including this one, are silently never included.** This package will report `no usage captured` for every single invocation, not an error.
+
+```python
+from adk_tracegauge import TraceGaugeUsagePlugin
+from google.adk.apps import App
+
+app = App(name="my_app", root_agent=root_agent, plugins=[TraceGaugeUsagePlugin()])
+```
+
+`App`'s constructor requires `name` — there's no default, and there's no `agent=` alias for `root_agent=`, easy typos to make since other ADK APIs use `agent`. Then reference `adk_tracegauge_cost_usd` as a `metric_name` in your eval config, and make sure `AgentEvaluator` picks up this `app` (it looks for an `app` attribute in your agent module automatically).
+
+## Install
+
+```bash
+pip install "adk-tracegauge"
+```
+
+`google-adk[eval]` and `tracegauge` are pulled in as dependencies. The `[eval]` extra on `google-adk` is not optional here even though this package doesn't use any of what it gates (pandas, jinja2, rouge-score, gepa, Vertex AI eval) — `google-adk`'s own `metric_evaluator_registry.py` unconditionally imports every one of its built-in evaluators at module import time, including the Vertex AI facade, which needs those packages just to import cleanly. Without the extra, `import adk_tracegauge` fails with `ModuleNotFoundError: No module named 'pandas'` — an ADK packaging quirk, not something this package controls.
+
+## What it reports, and what it deliberately doesn't
+
+- **`score`**: raw cost in USD for the invocation, summed across every real model call within it (tool loops and sub-agent delegation can mean more than one model call per invocation). Not normalized, not calibrated, not a 0–1 quality score.
+- **`rationale`** (the only other channel ADK's native reporting preserves — see "Why this metric always reports NOT_EVALUATED" below): a per-call breakdown — model, fresh/cached/output token counts, and their individual dollar costs.
+- **No calibrated efficiency bands.** tracegauge's own token-economy axis compares your numbers against a baseline built from 75 Claude Code sessions. That baseline is not used here, on purpose — applying a Claude-Code-derived baseline to ADK agent behavior would be an unvalidated transfer, and a plausible-looking-but-wrong number is worse than no number. This package reports raw counts and dollars only; set your own thresholds for what "too expensive" means for your agent.
+- **No trajectory-quality judging.** tracegauge's Ollama/Anthropic-based trajectory axis is out of scope for v1 — it's CC-specific tooling, unrelated to the cost story, and would add a dependency this package doesn't need.
+
+## Why this metric always reports `NOT_EVALUATED`
+
+ADK's built-in pass/fail convention is `PASSED if score >= threshold else FAILED` — hardcoded, higher-is-better. Cost is lower-is-better. There is no lower-is-better or inverted-metric convention anywhere in `google.adk.evaluation` to plug into (checked directly against the source, not assumed). Silently negating the score to make the built-in gate technically "work" would misrepresent the number to anyone reading it — a `score: -0.0043` needs an explanation to even parse. So this metric doesn't participate in ADK's pass/fail gating: it always reports `eval_status=NOT_EVALUATED`, whatever threshold you configure. Read `score` directly, or write your own comparison against it.
+
+## Gemini pricing
+
+`tracegauge`'s bundled price table covers Claude models only (its own domain — Claude Code sessions). ADK is Gemini-native, so this package ships and owns its own Gemini price table (`src/adk_tracegauge/data/gemini_prices.json`), passed explicitly into tracegauge's cost engine.
+
+- Every entry carries its own `source_url` and `fetched_on` date — currently `https://ai.google.dev/gemini-api/docs/pricing`, fetched 2026-08-13. **Prices change without notice.** Verify against the source before relying on a number for a real budget decision.
+- Standard tier only (no batch/flex/priority). Covers `gemini-2.5-pro`, `gemini-2.5-flash`, `gemini-2.5-flash-lite`, `gemini-2.0-flash` (deprecated, kept for historical sessions), `gemini-3.5-flash`, `gemini-3.5-flash-lite`, `gemini-3.6-flash`.
+- **Known gap:** `gemini-2.5-pro` is priced at its ≤200k-token-context rate only. Gemini bills >200k-context Pro calls at roughly double that rate; this table has no per-call tiered-pricing logic, so long-context Pro invocations will be under-priced. Not silently wrong — documented here and in the price table's own per-entry `note` field.
+- Cache-read discount is `0.1x` the model's fresh input rate for every model (verified against the published rate for each model individually, not assumed from tracegauge's Claude convention — it happens to match). Cache-*write* multipliers are `0.0`: Gemini's default automatic caching has no write-time surcharge and no `cache_creation` token field in its usage metadata at all, unlike Anthropic's explicit cache-write billing.
+
+**An invocation whose model isn't in this table is never priced with a fallback rate.** `score` reports `None`, and the rationale names exactly which model string didn't resolve and lists every model this package knows how to price. A cost number for the wrong model is worse than no number, so this package doesn't produce one.
+
+### Staleness — what happens when Gemini's prices change and the table doesn't
+
+A per-entry `source_url`/`fetched_on` date is provenance, not a freshness guarantee — nothing stops the table from silently aging out unless something actually checks it. Two independent things do:
+
+- **At use time**: every `ResolvedModel.is_stale` check compares `fetched_on` against today; past 180 days (`STALE_THRESHOLD_DAYS` in `_pricing.py`), a priced result's `rationale` gets a `PRICE TABLE STALE` line naming the model and threshold, and a real `warnings.warn(...)` fires alongside it — visible to anyone watching logs, not only whoever reads that one invocation's output. Staleness never blocks the number; it warns and still reports it, on the position that a flagged-possibly-wrong number beats no number for something you already computed.
+- **In CI**: `test_bundled_table_is_not_currently_stale` (`tests/test_pricing.py`) re-checks every bundled entry against the same 180-day threshold on every run. Once the table crosses it, this test starts failing on its own — not only when someone happens to notice a suspicious dollar figure.
+
+**Updating the price table**: edit `src/adk_tracegauge/data/gemini_prices.json` — update `input_usd_per_mtok`/`output_usd_per_mtok` against the current values at the `source_url` already on each entry, and bump `fetched_on` to today. Run `pytest tests/test_pricing.py` to confirm the staleness test goes green and nothing else broke, then open a PR. There's no automated price-scraping here by design — a human should look at the actual pricing page before a dollar figure changes.
+
+## Compatibility risk
+
+Registration uses `google.adk.evaluation.metric_evaluator_registry`, which google-adk marks `@experimental`: "may change or be removed... at any time," with no SemVer guarantee. This package pins `google-adk[eval]>=2.6.0,<2.7.0` accordingly — narrow, re-validated deliberately on each bump, not left open-ended.
+
+If the registry API breaks in a future google-adk release, registration happens at import time as a side effect, so the failure mode is a loud, immediate `AttributeError`/`TypeError` on `import adk_tracegauge` — not a silent no-op or a subtly wrong result. That's a favorable failure mode worth naming explicitly: you'll know immediately, not after a bad number reaches a dashboard.
+
+A scheduled CI job (`.github/workflows/pypi-canary.yml`) installs the *latest* `google-adk[eval]` release (ignoring the pin) and runs the full test suite weekly, so a break surfaces on a schedule rather than via a user bug report.
+
+The plugin half of this package (`BasePlugin.after_model_callback`) sits on firmer ground — it isn't marked `@experimental` anywhere in `google-adk`. If the registry API breaks, the usage-capture half likely still works; only evaluator registration needs a patch.
+
+## Relationship to tracegauge
+
+This package depends on `tracegauge`'s `tes.cost` module (`compute_session_cost` and its digest types) as a library. `tracegauge` overall is AGPL-3.0-only, but `tes/cost.py` and `tes/_digest.py` specifically are additionally available under Apache-2.0 — see [tracegauge's license note](https://github.com/gaurav-gandhi-2411/token-efficiency-scorer#license) — which is what lets this package stay Apache-2.0 itself. Every other part of tracegauge (the token-economy/trajectory/waste axes, the CLI, the dashboard) remains AGPL-3.0-only and is not used here.
+
+## What this is not
+
+Not a general ADK observability/tracing tool — see `traceAI-google-adk` for that. Not a statistics/confidence-interval layer for ADK evaluation results — that's a separate, harder problem ([agentgauge](https://github.com/gaurav-gandhi-2411/agentgauge)'s domain) blocked by the same `EvalMetricResult` field-stripping this package works around by using `score`+`rationale` only. Not a replacement for any of ADK's quality metrics — this reports cost alongside them, not instead of them.
+
+## License
+
+[Apache-2.0](LICENSE).

adk_tracegauge-0.1.0/README.md

@@ -0,0 +1,82 @@
+# adk-tracegauge
+
+Per-invocation dollar cost for [Google ADK](https://github.com/google/adk-python) agent evaluations, built on [tracegauge](https://github.com/gaurav-gandhi-2411/token-efficiency-scorer)'s cost engine. ADK's own `evaluation` package reports zero cost or token-efficiency data on any built-in metric — this fills exactly that gap, nothing more.
+
+[![License: Apache-2.0](https://img.shields.io/badge/License-Apache--2.0-blue.svg)](LICENSE)
+[![Python 3.10+](https://img.shields.io/badge/python-3.10%2B-blue.svg)](https://pypi.org/project/adk-tracegauge/)
+
+---
+
+## Required: wrap your agent in an App with the plugin
+
+**This is a hard requirement, not a caveat.** ADK's `Invocation` objects — what an evaluator normally receives — never carry token usage or model identity; ADK strips that down to author+content before an evaluator ever sees it. The real data lives on the `LlmResponse` passed to `BasePlugin.after_model_callback` during inference. That part fires for any real ADK run, App or not — but ADK's **eval harness specifically** (`LocalEvalService`, which `AgentEvaluator` and the `adk eval` CLI both sit on top of) only pulls plugins from `app.plugins` when you pass it an `App`. Evaluate against a bare `root_agent` with no `App`, and `LocalEvalService` builds its Runner with only its own internal eval plugins — **your plugins, including this one, are silently never included.** This package will report `no usage captured` for every single invocation, not an error.
+
+```python
+from adk_tracegauge import TraceGaugeUsagePlugin
+from google.adk.apps import App
+
+app = App(name="my_app", root_agent=root_agent, plugins=[TraceGaugeUsagePlugin()])
+```
+
+`App`'s constructor requires `name` — there's no default, and there's no `agent=` alias for `root_agent=`, easy typos to make since other ADK APIs use `agent`. Then reference `adk_tracegauge_cost_usd` as a `metric_name` in your eval config, and make sure `AgentEvaluator` picks up this `app` (it looks for an `app` attribute in your agent module automatically).
+
+## Install
+
+```bash
+pip install "adk-tracegauge"
+```
+
+`google-adk[eval]` and `tracegauge` are pulled in as dependencies. The `[eval]` extra on `google-adk` is not optional here even though this package doesn't use any of what it gates (pandas, jinja2, rouge-score, gepa, Vertex AI eval) — `google-adk`'s own `metric_evaluator_registry.py` unconditionally imports every one of its built-in evaluators at module import time, including the Vertex AI facade, which needs those packages just to import cleanly. Without the extra, `import adk_tracegauge` fails with `ModuleNotFoundError: No module named 'pandas'` — an ADK packaging quirk, not something this package controls.
+
+## What it reports, and what it deliberately doesn't
+
+- **`score`**: raw cost in USD for the invocation, summed across every real model call within it (tool loops and sub-agent delegation can mean more than one model call per invocation). Not normalized, not calibrated, not a 0–1 quality score.
+- **`rationale`** (the only other channel ADK's native reporting preserves — see "Why this metric always reports NOT_EVALUATED" below): a per-call breakdown — model, fresh/cached/output token counts, and their individual dollar costs.
+- **No calibrated efficiency bands.** tracegauge's own token-economy axis compares your numbers against a baseline built from 75 Claude Code sessions. That baseline is not used here, on purpose — applying a Claude-Code-derived baseline to ADK agent behavior would be an unvalidated transfer, and a plausible-looking-but-wrong number is worse than no number. This package reports raw counts and dollars only; set your own thresholds for what "too expensive" means for your agent.
+- **No trajectory-quality judging.** tracegauge's Ollama/Anthropic-based trajectory axis is out of scope for v1 — it's CC-specific tooling, unrelated to the cost story, and would add a dependency this package doesn't need.
+
+## Why this metric always reports `NOT_EVALUATED`
+
+ADK's built-in pass/fail convention is `PASSED if score >= threshold else FAILED` — hardcoded, higher-is-better. Cost is lower-is-better. There is no lower-is-better or inverted-metric convention anywhere in `google.adk.evaluation` to plug into (checked directly against the source, not assumed). Silently negating the score to make the built-in gate technically "work" would misrepresent the number to anyone reading it — a `score: -0.0043` needs an explanation to even parse. So this metric doesn't participate in ADK's pass/fail gating: it always reports `eval_status=NOT_EVALUATED`, whatever threshold you configure. Read `score` directly, or write your own comparison against it.
+
+## Gemini pricing
+
+`tracegauge`'s bundled price table covers Claude models only (its own domain — Claude Code sessions). ADK is Gemini-native, so this package ships and owns its own Gemini price table (`src/adk_tracegauge/data/gemini_prices.json`), passed explicitly into tracegauge's cost engine.
+
+- Every entry carries its own `source_url` and `fetched_on` date — currently `https://ai.google.dev/gemini-api/docs/pricing`, fetched 2026-08-13. **Prices change without notice.** Verify against the source before relying on a number for a real budget decision.
+- Standard tier only (no batch/flex/priority). Covers `gemini-2.5-pro`, `gemini-2.5-flash`, `gemini-2.5-flash-lite`, `gemini-2.0-flash` (deprecated, kept for historical sessions), `gemini-3.5-flash`, `gemini-3.5-flash-lite`, `gemini-3.6-flash`.
+- **Known gap:** `gemini-2.5-pro` is priced at its ≤200k-token-context rate only. Gemini bills >200k-context Pro calls at roughly double that rate; this table has no per-call tiered-pricing logic, so long-context Pro invocations will be under-priced. Not silently wrong — documented here and in the price table's own per-entry `note` field.
+- Cache-read discount is `0.1x` the model's fresh input rate for every model (verified against the published rate for each model individually, not assumed from tracegauge's Claude convention — it happens to match). Cache-*write* multipliers are `0.0`: Gemini's default automatic caching has no write-time surcharge and no `cache_creation` token field in its usage metadata at all, unlike Anthropic's explicit cache-write billing.
+
+**An invocation whose model isn't in this table is never priced with a fallback rate.** `score` reports `None`, and the rationale names exactly which model string didn't resolve and lists every model this package knows how to price. A cost number for the wrong model is worse than no number, so this package doesn't produce one.
+
+### Staleness — what happens when Gemini's prices change and the table doesn't
+
+A per-entry `source_url`/`fetched_on` date is provenance, not a freshness guarantee — nothing stops the table from silently aging out unless something actually checks it. Two independent things do:
+
+- **At use time**: every `ResolvedModel.is_stale` check compares `fetched_on` against today; past 180 days (`STALE_THRESHOLD_DAYS` in `_pricing.py`), a priced result's `rationale` gets a `PRICE TABLE STALE` line naming the model and threshold, and a real `warnings.warn(...)` fires alongside it — visible to anyone watching logs, not only whoever reads that one invocation's output. Staleness never blocks the number; it warns and still reports it, on the position that a flagged-possibly-wrong number beats no number for something you already computed.
+- **In CI**: `test_bundled_table_is_not_currently_stale` (`tests/test_pricing.py`) re-checks every bundled entry against the same 180-day threshold on every run. Once the table crosses it, this test starts failing on its own — not only when someone happens to notice a suspicious dollar figure.
+
+**Updating the price table**: edit `src/adk_tracegauge/data/gemini_prices.json` — update `input_usd_per_mtok`/`output_usd_per_mtok` against the current values at the `source_url` already on each entry, and bump `fetched_on` to today. Run `pytest tests/test_pricing.py` to confirm the staleness test goes green and nothing else broke, then open a PR. There's no automated price-scraping here by design — a human should look at the actual pricing page before a dollar figure changes.
+
+## Compatibility risk
+
+Registration uses `google.adk.evaluation.metric_evaluator_registry`, which google-adk marks `@experimental`: "may change or be removed... at any time," with no SemVer guarantee. This package pins `google-adk[eval]>=2.6.0,<2.7.0` accordingly — narrow, re-validated deliberately on each bump, not left open-ended.
+
+If the registry API breaks in a future google-adk release, registration happens at import time as a side effect, so the failure mode is a loud, immediate `AttributeError`/`TypeError` on `import adk_tracegauge` — not a silent no-op or a subtly wrong result. That's a favorable failure mode worth naming explicitly: you'll know immediately, not after a bad number reaches a dashboard.
+
+A scheduled CI job (`.github/workflows/pypi-canary.yml`) installs the *latest* `google-adk[eval]` release (ignoring the pin) and runs the full test suite weekly, so a break surfaces on a schedule rather than via a user bug report.
+
+The plugin half of this package (`BasePlugin.after_model_callback`) sits on firmer ground — it isn't marked `@experimental` anywhere in `google-adk`. If the registry API breaks, the usage-capture half likely still works; only evaluator registration needs a patch.
+
+## Relationship to tracegauge
+
+This package depends on `tracegauge`'s `tes.cost` module (`compute_session_cost` and its digest types) as a library. `tracegauge` overall is AGPL-3.0-only, but `tes/cost.py` and `tes/_digest.py` specifically are additionally available under Apache-2.0 — see [tracegauge's license note](https://github.com/gaurav-gandhi-2411/token-efficiency-scorer#license) — which is what lets this package stay Apache-2.0 itself. Every other part of tracegauge (the token-economy/trajectory/waste axes, the CLI, the dashboard) remains AGPL-3.0-only and is not used here.
+
+## What this is not
+
+Not a general ADK observability/tracing tool — see `traceAI-google-adk` for that. Not a statistics/confidence-interval layer for ADK evaluation results — that's a separate, harder problem ([agentgauge](https://github.com/gaurav-gandhi-2411/agentgauge)'s domain) blocked by the same `EvalMetricResult` field-stripping this package works around by using `score`+`rationale` only. Not a replacement for any of ADK's quality metrics — this reports cost alongside them, not instead of them.
+
+## License
+
+[Apache-2.0](LICENSE).

adk_tracegauge-0.1.0/pyproject.toml

@@ -0,0 +1,87 @@
+[build-system]
+requires = ["setuptools>=70", "wheel"]
+build-backend = "setuptools.build_meta"
+
+[project]
+name = "adk-tracegauge"
+version = "0.1.0"
+description = "Per-invocation cost-in-USD evaluator for Google's Agent Development Kit (ADK), built on tracegauge's cost engine. Raw dollars and tokens, no calibrated bands, no fabricated numbers for unknown models."
+authors = [{ name = "Gaurav Gandhi", email = "gaurav.gandhi2411@gmail.com" }]
+requires-python = ">=3.10"
+readme = "README.md"
+license = "Apache-2.0"
+license-files = ["LICENSE"]
+classifiers = [
+    "Development Status :: 3 - Alpha",
+    "Programming Language :: Python :: 3",
+    "Programming Language :: Python :: 3.10",
+    "Programming Language :: Python :: 3.11",
+    "Programming Language :: Python :: 3.12",
+    "Programming Language :: Python :: 3.13",
+    "Topic :: Software Development :: Quality Assurance",
+    "Topic :: Utilities",
+    "Intended Audience :: Developers",
+    "Environment :: Console",
+    "Operating System :: OS Independent",
+]
+keywords = ["adk", "agent-development-kit", "google-adk", "evaluation", "cost", "token-efficiency"]
+dependencies = [
+    # [eval] is required, not optional: google-adk's own
+    # metric_evaluator_registry.py unconditionally imports every built-in
+    # evaluator at module import time, including the Vertex AI eval facade,
+    # which transitively needs pandas/jinja2/rouge-score/tabulate/gepa --
+    # gated behind this extra. Without it, `import adk_tracegauge` fails
+    # with ModuleNotFoundError: No module named 'pandas', even though this
+    # package uses none of those libraries itself. Confirmed by reading
+    # google-adk's own pyproject.toml requires list.
+    "google-adk[eval]>=2.6.0,<2.7.0",
+    "tracegauge>=0.10.0,<0.11.0",
+]
+
+[dependency-groups]
+dev = [
+    "ruff>=0.15,<1",
+    "mypy>=1.14,<2",
+    "pytest>=8.3,<9",
+    "pytest-asyncio>=0.25,<1",
+    "pytest-cov>=6.0,<7",
+    "pytest-mock>=3.14,<4",
+]
+
+[project.urls]
+Repository = "https://github.com/gaurav-gandhi-2411/adk-tracegauge"
+
+[tool.setuptools.packages.find]
+where = ["src"]
+include = ["adk_tracegauge*"]
+
+[tool.setuptools.package-data]
+"adk_tracegauge" = ["data/gemini_prices.json"]
+
+[tool.ruff]
+line-length = 100
+target-version = "py311"
+src = ["src"]
+
+[tool.ruff.lint]
+select = ["E", "F", "I", "UP", "B", "SIM"]
+ignore = ["E501"]
+
+[tool.mypy]
+python_version = "3.12"
+strict = true
+ignore_missing_imports = true
+warn_unused_ignores = true
+
+[tool.pytest.ini_options]
+testpaths = ["tests"]
+asyncio_mode = "auto"
+addopts = "-v"
+pythonpath = [".", "src"]
+
+[tool.coverage.run]
+omit = ["tests/*"]
+
+[tool.coverage.report]
+fail_under = 0
+exclude_lines = ["pragma: no cover", "if TYPE_CHECKING:"]

adk_tracegauge-0.1.0/setup.cfg

@@ -0,0 +1,4 @@
+[egg_info]
+tag_build = 
+tag_date = 0
+

adk_tracegauge-0.1.0/src/adk_tracegauge/__init__.py

@@ -0,0 +1,40 @@
+"""adk-tracegauge — per-invocation cost-in-USD evaluator for the Agent Development Kit.
+
+Importing this package registers the "adk_tracegauge_cost_usd" metric into
+google-adk's DEFAULT_METRIC_EVALUATOR_REGISTRY as a side effect, matching
+the registration pattern google-adk itself documents for third-party
+metrics. If google-adk's @experimental registry API has changed
+incompatibly, this import-time call fails loudly (AttributeError/TypeError)
+rather than silently doing nothing -- see README, "Compatibility risk".
+
+Required setup (not optional -- see README for why):
+
+    from adk_tracegauge import TraceGaugeUsagePlugin
+    app = App(agent=root_agent, plugins=[TraceGaugeUsagePlugin()])
+
+Then reference "adk_tracegauge_cost_usd" as a metric_name in your eval_config.json.
+"""
+
+from __future__ import annotations
+
+from google.adk.evaluation.metric_evaluator_registry import DEFAULT_METRIC_EVALUATOR_REGISTRY
+
+from ._plugin import TraceGaugeUsagePlugin
+from ._store import DEFAULT_USAGE_STORE, UsageStore
+from .evaluator import _METRIC_INFO, METRIC_NAME, CostEfficiencyEvaluator
+
+DEFAULT_METRIC_EVALUATOR_REGISTRY.register_evaluator(
+    metric_info=_METRIC_INFO,
+    evaluator=CostEfficiencyEvaluator,
+)
+
+__version__ = "0.1.0"
+
+__all__ = [
+    "CostEfficiencyEvaluator",
+    "TraceGaugeUsagePlugin",
+    "UsageStore",
+    "DEFAULT_USAGE_STORE",
+    "METRIC_NAME",
+    "__version__",
+]

adk_tracegauge-0.1.0/src/adk_tracegauge/_adapter.py

@@ -0,0 +1,92 @@
+"""adk_tracegauge/_adapter.py — Maps captured ADK usage data onto tracegauge's digest shape.
+
+tracegauge's tes.cost.compute_session_cost operates on a tes._digest.SessionDigest
+(a list of TurnDigest, one per AI turn). One ADK invocation can involve more
+than one real model call, so it maps onto a SessionDigest with one TurnDigest
+per captured call -- summed cost across the whole invocation, same as
+tracegauge sums cost across a whole Claude Code session.
+
+Model resolution happens here, once, before any TurnDigest is built. A call
+whose model_version doesn't match the Gemini price table makes the whole
+adaptation fail closed (see AdaptResult.unresolved_model) rather than
+producing a partially-priced or silently-approximate result.
+"""
+
+from __future__ import annotations
+
+from dataclasses import dataclass
+
+from tes._digest import SessionDigest, TurnDigest
+
+from ._pricing import known_model_keys, resolve_model
+from ._store import CapturedCall
+
+
+@dataclass
+class AdaptResult:
+    """Either a ready-to-price digest, or the model that blocked pricing."""
+
+    digest: SessionDigest | None
+    unresolved_model: str | None
+
+    @property
+    def ok(self) -> bool:
+        return self.digest is not None
+
+
+def build_session_digest(invocation_id: str, calls: list[CapturedCall]) -> AdaptResult:
+    """Builds a SessionDigest from captured calls, or reports the blocking model.
+
+    Every call's model_version must resolve against the Gemini price table.
+    On the first unresolved model, adaptation stops and returns it -- no
+    partial digest, no fallback pricing.
+    """
+    turns: list[TurnDigest] = []
+
+    for index, call in enumerate(calls):
+        resolved = resolve_model(call.model_version)
+        if resolved is None:
+            return AdaptResult(digest=None, unresolved_model=call.model_version)
+
+        turns.append(
+            TurnDigest(
+                turn_index=index,
+                role="ai",
+                tool_names=[],
+                content_snippet="",
+                token_count_input=call.prompt_token_count,
+                token_count_output=call.candidates_token_count,
+                cache_read=call.cached_content_token_count,
+                h2_duplicate=False,
+                cache_creation=0,
+                model=resolved.model_key,
+            )
+        )
+
+    digest = SessionDigest(
+        session_id=invocation_id,
+        domain="adk_invocation",
+        resolved=True,
+        total_tokens=sum(c.prompt_token_count + c.candidates_token_count for c in calls),
+        turn_count=len(turns),
+        h2_duplicate_count=0,
+        cache_hit_rate=0.0,
+        p25_token_ratio=0.0,
+        output_tokens_available=True,
+        task_description="",
+        turns=turns,
+    )
+    return AdaptResult(digest=digest, unresolved_model=None)
+
+
+def unknown_model_message(model_version: str) -> str:
+    known = ", ".join(known_model_keys())
+    return (
+        f"cost not computed: model '{model_version}' is not in the "
+        f"adk-tracegauge Gemini price table (known: {known}). "
+        "Check https://ai.google.dev/gemini-api/docs/pricing for a new or "
+        "renamed model, or open an issue if this model should be added."
+    )
+
+
+__all__ = ["AdaptResult", "build_session_digest", "unknown_model_message"]