Add preregistered uv public-project validation

CHANGELOG.md

@@ -4,6 +4,15 @@ All notable changes to this project are documented here. The format follows [Kee
 
 ## [Unreleased]
 
+## [0.2.0-alpha.2] — 2026-08-18
+
+### Added
+- Added a preregistered public-project transfer check derived from frozen Astral
+  uv issue-triage and workflow-failure prompts. Across 84 authenticated Codex
+  calls, both profiles produced all expected decisions while `text-only/v1`
+  reduced provider-reported input per call by 10.50% and 10.74%, with no
+  operational errors or transport fallbacks.
+
 ## [0.2.0-alpha.1] — 2026-08-18
 
 ### Added — context behavior audit

README.md

@@ -56,6 +56,24 @@ guide](docs/CODEX_TEXT_ONLY_CASE_STUDY.md), plus the complete per-call outputs,
 token counts, source hashes, runtime settings, and limitations in the
 [`paired three-trial audit`](examples/project_instructions/codex-text-only-profile-audit.paired-3x-final.2026-08-17.json).
 
+### Public-project transfer check: Astral uv
+
+The same frozen profile was then tested against decision rules adapted from
+public Astral uv agent prompts at commit
+[`5cc226096`](https://github.com/astral-sh/uv/tree/5cc226096ea4424d021be17259bae51d761a827b).
+The 14 cases were committed before execution, then run three times per profile:
+
+| Workload | Quality | Full input per call | Reduction |
+|---|---:|---:|---:|
+| uv issue-triage decisions | 24/24 in each profile | 20,335.25 -> 18,200.75 | 10.50% |
+| uv workflow-failure decisions | 18/18 in each profile | 20,345.83 -> 18,160.83 | 10.74% |
+
+All 84 calls completed with zero operational errors and zero transport
+fallbacks. This is an external-project corpus run by denser's maintainers, not
+an independent reproduction or an endorsement by Astral. See the
+[frozen corpus and boundaries](examples/project_instructions/03_uv_public_pilot/README.md)
+and the [complete per-call report](examples/project_instructions/03_uv_public_pilot/codex-profile-audit.paired-3x.2026-08-18.json).
+
 ---
 
 ## The problem

THIRD_PARTY_NOTICES.md

@@ -20,3 +20,19 @@ evaluation. The case README records the provenance, modification boundary,
 frozen hashes, and reproduction limits.
 
 The Apache License 2.0 text is provided in [`LICENSE`](LICENSE).
+
+## Astral uv public agent rules
+
+The decision-only replay assets under
+`examples/project_instructions/03_uv_public_pilot/` are denser-created
+projections of public agent rules from the Astral `uv` repository at commit
+`5cc226096ea4424d021be17259bae51d761a827b`:
+
+- Upstream repository: <https://github.com/astral-sh/uv>
+- Issue triage prompt: <https://github.com/astral-sh/uv/blob/5cc226096ea4424d021be17259bae51d761a827b/agents/prompts/triage-issue.md>
+- Workflow diagnosis prompt: <https://github.com/astral-sh/uv/blob/5cc226096ea4424d021be17259bae51d761a827b/agents/prompts/diagnose-workflow-failure.md>
+- License: MIT or Apache License 2.0
+
+The local assets are modified, reduced projections rather than unmodified
+copies. Astral and the uv maintainers do not endorse or validate this study.
+The Apache License 2.0 text is provided in [`LICENSE`](LICENSE).

benchmarks/README.md

@@ -22,6 +22,16 @@ that need no files, commands, network, plugins, apps, skills, or memory. The
 published run and exact interpretation are documented in
 [`docs/CODEX_TEXT_ONLY_CASE_STUDY.md`](../docs/CODEX_TEXT_ONLY_CASE_STUDY.md).
 
+For the preregistered external-project corpus derived from uv's public Codex
+rules, use:
+
+```bash
+python benchmarks/codex_profile_audit.py \
+  --scenario-set uv-public-pilot \
+  --trials 3 --workers 8 --seed 20260818 --respect-system-proxy \
+  --output build/uv-public-pilot.json
+```
+
 ## What it does
 
 1. Iterates over all curated example pairs in `examples/`

benchmarks/codex_profile_audit.py

@@ -41,7 +41,7 @@ class CallUnit:
     trial: int
 
 
-SCENARIOS = (
+DEFAULT_SCENARIOS = (
     Scenario(
         name="release_operations",
         asset=ROOT / "examples" / "project_instructions" / "01_codex_release_ops" / "AGENTS.md",
@@ -66,6 +66,48 @@ class CallUnit:
     ),
 )
 
+UV_PUBLIC_PILOT_SCENARIOS = (
+    Scenario(
+        name="uv_issue_triage_snapshot",
+        asset=(
+            ROOT
+            / "examples"
+            / "project_instructions"
+            / "03_uv_public_pilot"
+            / "issue-triage-rules.md"
+        ),
+        suite=(
+            ROOT
+            / "examples"
+            / "project_instructions"
+            / "03_uv_public_pilot"
+            / "issue-triage.replay.json"
+        ),
+    ),
+    Scenario(
+        name="uv_workflow_failure_snapshot",
+        asset=(
+            ROOT
+            / "examples"
+            / "project_instructions"
+            / "03_uv_public_pilot"
+            / "workflow-failure-rules.md"
+        ),
+        suite=(
+            ROOT
+            / "examples"
+            / "project_instructions"
+            / "03_uv_public_pilot"
+            / "workflow-failure.replay.json"
+        ),
+    ),
+)
+
+SCENARIO_SETS = {
+    "built-in": DEFAULT_SCENARIOS,
+    "uv-public-pilot": UV_PUBLIC_PILOT_SCENARIOS,
+}
+
 
 def _sha256(path: Path) -> str:
     """Hash the UTF-8 text exactly as Python passes it to the backend.
@@ -175,6 +217,7 @@ def _summarize(calls: list[dict[str, Any]]) -> dict[str, Any]:
 def _parse_args() -> argparse.Namespace:
     parser = argparse.ArgumentParser(description=__doc__)
     parser.add_argument("--model", default="gpt-5.6-sol")
+    parser.add_argument("--scenario-set", choices=tuple(SCENARIO_SETS), default="built-in")
     parser.add_argument(
         "--reasoning-effort",
         choices=("none", "low", "medium", "high", "xhigh", "max"),
@@ -205,11 +248,12 @@ def _parse_args() -> argparse.Namespace:
 def main() -> int:
     """Run the paired audit and return a process exit status."""
     args = _parse_args()
+    selected_scenarios = SCENARIO_SETS[args.scenario_set]
     profiles = (args.baseline_profile, args.variant_profile)
     assets: dict[str, str] = {}
     units: list[CallUnit] = []
     scenario_sources: dict[str, dict[str, Any]] = {}
-    for scenario in SCENARIOS:
+    for scenario in selected_scenarios:
         asset = scenario.asset.read_text(encoding="utf-8")
         suite = load_replay_suite(scenario.suite)
         if suite.freeze is not None:
@@ -262,7 +306,7 @@ def main() -> int:
             )
 
     scenarios: list[dict[str, Any]] = []
-    for scenario in SCENARIOS:
+    for scenario in selected_scenarios:
         calls_by_profile = {
             profile: sorted(
                 (
@@ -309,6 +353,7 @@ def main() -> int:
     report = {
         "schema_version": SCHEMA_VERSION,
         "source_hash_method": "utf8-lf-v1",
+        "scenario_set": args.scenario_set,
         "generated_at_utc": datetime.now(timezone.utc).isoformat().replace("+00:00", "Z"),
         "runtime": {
             "backend_kind": "codex-cli",

examples/project_instructions/03_uv_public_pilot/README.md

@@ -0,0 +1,60 @@
+# uv public-project corpus
+
+This preregistered corpus asks whether denser's Codex `text-only/v1` profile
+preserves fixed policy decisions while reducing provider-reported full input.
+It uses decision rules derived from public Astral uv agent prompts at commit
+[`5cc226096`](https://github.com/astral-sh/uv/tree/5cc226096ea4424d021be17259bae51d761a827b).
+
+## Frozen sources
+
+- [Issue triage prompt](https://github.com/astral-sh/uv/blob/5cc226096ea4424d021be17259bae51d761a827b/agents/prompts/triage-issue.md), blob `bc3777d77d9d05f03b277431cb4bfff3ad39d5c6`
+- [Issue triage schema](https://github.com/astral-sh/uv/blob/5cc226096ea4424d021be17259bae51d761a827b/agents/schemas/issue-triage.json), blob `aeb2a3a0bdd09721261e9ab4aa73bd3390f24082`
+- [Workflow failure prompt](https://github.com/astral-sh/uv/blob/5cc226096ea4424d021be17259bae51d761a827b/agents/prompts/diagnose-workflow-failure.md), blob `f285cbec1164a98788a1e31d2078c3337847ddaa`
+- [Workflow failure schema](https://github.com/astral-sh/uv/blob/5cc226096ea4424d021be17259bae51d761a827b/agents/schemas/workflow-failure.json), blob `0a5cdb5be3470a927e90071fa1b1214b221fa171`
+
+The two local rule files are reduced, decision-only projections. They replace
+live repository and GitHub lookup with complete evidence snapshots and reduce
+the output to fields that can be matched exactly. This tests a narrow external
+corpus, not uv's full automation.
+
+## Preregistered run
+
+- 8 issue cases and 6 workflow cases;
+- 3 trials under `standard` and 3 under `text-only/v1`;
+- seeded randomized submission order;
+- 84 total authenticated Codex calls;
+- success requires 42/42 exact outputs in each profile, no quality decrease,
+  at least 10% lower input per call in both scenarios, zero operational errors,
+  and zero transport fallbacks.
+
+```bash
+python benchmarks/codex_profile_audit.py \
+  --scenario-set uv-public-pilot \
+  --trials 3 --workers 8 --seed 20260818 --respect-system-proxy \
+  --output build/uv-public-pilot.json
+```
+
+## Result
+
+The preregistered run passed on 2026-08-18 with Codex CLI 0.147.0,
+`gpt-5.6-sol`, and medium reasoning:
+
+| Scenario | Standard | Text-only | Reduction | Quality |
+|---|---:|---:|---:|---:|
+| Issue triage | 20,335.25 | 18,200.75 | 10.50% | 24/24 -> 24/24 |
+| Workflow failure | 20,345.83 | 18,160.83 | 10.74% | 18/18 -> 18/18 |
+
+All 84 calls completed with zero operational errors and zero transport
+fallbacks. The complete sanitized report is
+[`codex-profile-audit.paired-3x.2026-08-18.json`](codex-profile-audit.paired-3x.2026-08-18.json).
+
+## Boundaries
+
+This is an external-project corpus authored and run by denser's maintainers. It
+is not an independent third-party reproduction and does not imply endorsement
+by Astral or uv's maintainers. No upstream issue, pull request, or comment is
+part of this study. The run tests expected-output transfer for a capability
+profile; it does not add a new negative-control execution for context rewriting.
+
+uv is available under the MIT or Apache-2.0 license. See
+[`THIRD_PARTY_NOTICES.md`](../../../THIRD_PARTY_NOTICES.md).