babaliauskas · Sep 30, 2026 · Sep 30, 2026 · Sep 30, 2026 · Sep 30, 2026 · Sep 30, 2026
diff --git a/AGENTS.md b/AGENTS.md
@@ -60,7 +60,8 @@ Detailed operating rules — commands, repo map, hard rules — live in
 uv venv --python 3.11 && source .venv/bin/activate
 uv pip install -e ".[dev]"
 pytest                        # full suite
-pre-commit run --all-files    # exactly what CI runs
+make ci                       # exactly what CI runs (also the pre-push hook)
+pre-commit run --all-files    # commit-stage hooks only (file hygiene, ruff --fix, format, mypy)
 ```
 
 - Config models use `extra="forbid"`; any new `evalshift.yaml` field needs docs

diff --git a/CHANGELOG.md b/CHANGELOG.md
@@ -83,6 +83,34 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
   whole source tree. The floor sits below the measured 94% on purpose, so a
   real regression fails CI while an honest refactor does not.
 
+- An audit of the docs against the code found them wrong in a few dozen
+  places, and all of them now say what the CLI does. Nothing about CLI
+  behaviour changed. The one that matters most concerns slices: the docs
+  said a top-level `slices:` block picks examples by tag, names the slice,
+  and scopes it to prompts. It does none of that. The block is validated
+  (the reserved `overall` name is still rejected) and recorded in the run
+  bundle, but analysis never reads it. Slices come from example `tags`, one
+  per distinct tag plus `all`, and per-slice budgets are keyed by tag under
+  `migration_policy.slices`. The docs now say the block is not applied, and
+  so does the `SliceConfig` docstring, which described `filter` as a Python
+  expression. Imported agent traces (`traces import`) stay local; the bundle
+  carries only the replay's own tool-call trace, without tool results or
+  `model_call` events. The response cache serves only tool-less examples,
+  so every `run` of an agent suite is live and full price. `--resume`
+  hashes the suite's path, not its contents. `push <run-id>` uploads an
+  existing bundle as-is instead of rebuilding it, and `bundle` needs a git
+  SHA. `--policy-gate` also fails when no `migration_policy` is configured.
+  The `init` profile table had the wrong `model-upgrade` numbers and no
+  tool-divergence column. The multi-turn suite example failed to load
+  because it had no `tools`. The failure-label list was missing
+  `TOOL_GROUND_TRUTH_MISS`. Upstream model-call failures and evaluator
+  failures are handled the same way, as errored rows excluded from the
+  statistics. The GitHub Action docs gained `require-policy` and the other
+  missing inputs. `record_model_call` examples now pass the required
+  `tools=`. DOCS.md's header said version 1.0.1; a new check in
+  `tests/unit/test_docs_currency.py` keeps the version in DOCS.md and
+  llms-full.txt equal to the package's.
+
 ## [1.1.0] - 2026-09-19
 
 Shipped as a minor deliberately. The `thresholds` removal below is breaking by

diff --git a/CONTRIBUTING.md b/CONTRIBUTING.md
@@ -35,7 +35,8 @@ pytest -m "not integration"             # unit tests only
 ruff check .                            # lint
 ruff format .                           # auto-format
 mypy --strict src/evalshift_cli             # type-check
-pre-commit run --all-files              # everything pre-commit runs
+pre-commit run --all-files              # commit-stage hooks only
+make ci                                 # exactly what CI runs (also the pre-push hook)
 ```
 
 ## Style

diff --git a/DOCS.md b/DOCS.md
diff --git a/README.md b/README.md
@@ -226,9 +226,10 @@ The short version:
   statistics, the analysis, the migration decision, economics, and the
   machine-written insights narrative.
 * **Never uploads**: provider API keys, prompt bodies and system prompts,
-  suite conversation histories, tool definitions/schemas, `raw.jsonl`, the
-  response cache, captures, and `report.html`. Prompt and dataset content is
-  replaced by SHA-256 hashes so diffs still align across runs.
+  suite conversation histories, tool definitions/schemas, `raw.jsonl`,
+  imported agent traces, the response cache, captures, and `report.html`.
+  Prompt and dataset content is replaced by SHA-256 hashes so diffs still
+  align across runs.
 * **Can still be sensitive**: inputs, expected outputs, model outputs, and
   traces carry whatever content your suite or your models put in them. Redact
   at capture time (see the SDK's redaction boundary) and inspect before

diff --git a/docs/agents.md b/docs/agents.md
@@ -46,7 +46,7 @@ row carrying its own `toolset_ref`:
 cd examples/agent
 export GOOGLE_API_KEY=<google-api-key>
 evalshift run --yes --from gemini-2.5-flash --to gemini-3.1-flash-lite-preview
-RUN_ID=$(ls .evalshift/runs/ | head -1)
+RUN_ID=$(ls -t .evalshift/runs/ | head -1)
 evalshift evaluate "$RUN_ID"
 evalshift analyze "$RUN_ID"
 evalshift report "$RUN_ID" --open
@@ -136,7 +136,7 @@ Nothing in `evalshift.yaml` wires a toolset to a prompt — dispatch reads it
 off each golden-suite *example* instead (`toolset_ref` or inline `tools`, see
 [Suite ground truth](#suite-ground-truth) below), so the same prompt can
 legitimately dispatch some examples with tools and others without, in one
-run. `tools.yaml` above is just this project's human-readable record of what
+run. [`examples/agent/tools.yaml`](https://github.com/babaliauskas/evalshift-cli/blob/main/examples/agent/tools.yaml) is just this project's human-readable record of what
 those tools are; it accepts either Anthropic-shape (`name` / `description` /
 `input_schema`) or OpenAI-shape (`{ "type": "function", "function": {...}
 }`) entries — `evalshift run` serialises whatever a toolset resolves to in
@@ -507,9 +507,11 @@ expectation are ignored here. Which keys are compared follows each expectation's
 `match_strategy`: `exact` also flags arguments the ground truth did not record,
 while `subset` (what `capture promote` writes) scores the recorded keys only.
 
-Check `evalshift doctor` first. If it warns about **tool argument shape**, your
-recorded arguments use keys the declared schema does not have, and every
-ground-truth comparison will score 0 for a reason that is not the model's fault.
+If every ground-truth comparison scores 0, check whether your recorded
+arguments use keys the declared schema does not have — a capture of a decorated
+function's parameters rather than the model's own arguments scores 0 for a
+reason that is not the model's fault. See
+[Wrapper arguments are unwrapped](#wrapper-arguments-are-unwrapped-legacy-captures).
 
 A ground-truth field that **neither** model produced is dropped from that call's
 denominator on both sides and disclosed as `unmeasured_fields` in the record's