feat: add behavior-validated context selection

CHANGELOG.md

@@ -4,6 +4,17 @@ All notable changes to this project are documented here. The format follows [Kee
 
 ## [Unreleased]
 
+### Added
+- Added `denser minimize-context` and a versioned context-bundle manifest. The
+  selector greedily tests optional components, keeps behavior-changing or
+  uncertain components, requires a sensitive negative control, and repeats a
+  final audit against a caller-set full-input reduction target.
+- Added bounded concurrent replay for the Codex CLI backend. Per-call metadata
+  remains thread-isolated and each CLI turn remains ephemeral and read-only.
+- Added a two-task local-file Codex pilot. It removed an unrelated archived
+  handbook, retained the release and CI policies whose removal regressed, kept
+  all final cases at 3/3, and reduced provider-reported full input by 12.47%.
+
 ## [0.2.0-alpha.3] — 2026-08-18
 
 ### Changed

README.md

@@ -20,43 +20,55 @@
 
 ---
 
-## Featured: measurable Codex input reduction for text-only tasks
+## Featured: automatic context pruning in a normal Codex tool workflow
 
-The first result that clears denser's end-to-end bar comes from capability
-selection, not prose compression. For replay tasks that need no files, shell,
-network, plugins, apps, skills, or memory, the Codex CLI adapter can use an
-explicit `text-only` profile and omit those unused capabilities from the model
-input. `standard` remains the default.
+`denser minimize-context` takes a manifest of visible context components,
+tries removing optional components one at a time, and keeps a removal only when
+behavior remains identical and a known-bad control still fails. Errors,
+improvements that change behavior, and insensitive tests all fail closed.
 
-With Codex CLI 0.147.0, `gpt-5.6-sol`, and medium reasoning:
+The first tool-using pilot used Codex CLI 0.147.0, `gpt-5.6-sol`, medium
+reasoning, and the `standard` capability profile. Both tasks had to read a local
+JSON file whose identifier and arbitrary policy code were absent from the
+prompt, so the expected answers could not be recovered from a text-only input.
 
-| Workload | Quality | Full input per call | Reduction |
+| Workload | Complete context | Selected context |
+|---|---:|---:|
+| Local release-record decision | 3/3 | 3/3 |
+| Local CI-record decision | 3/3 | 3/3 |
+
+The selector removed a 17,236-byte archived handbook but rejected attempts to
+remove the release or CI policy because each caused a covered regression. The
+final three-trial audit completed with zero operational errors:
+
+| Measurement | Complete context | Selected context | Reduction |
 |---|---:|---:|---:|
-| Release-operation decisions | 27/27 in each profile | 20,294.11 → 18,154.00 | 10.55% |
-| Automation permission routing | 15/15 in each profile | 20,619.00 → 18,434.00 | 10.60% |
-
-This clears the predeclared rule of at least two real scenarios with at least
-10% provider-reported full-input reduction and no observed quality loss. The
-final run made 84 authenticated calls in seeded randomized order, with three
-trials per case, zero operational errors, and zero transport fallbacks. It is
-not a general coding mode: tasks that need tools must use `standard`.
-
-The first strict run caught one regression: without tools, one case asked for
-more context instead of following its fixed output contract. The `text-only/v1`
-wrapper now states that all required input is already present, and the complete
-84-call audit was rerun rather than patching the single failure. This is the
-kind of false confidence denser is designed to expose.
-
-The earlier 10.3%-shorter instruction rewrite reduced full Codex input by only
-about 0.25%. That negative result remains important: rewriting a small file is
-not enough when the larger cost is unused runtime context.
-
-See the [case study and reproduction
-guide](docs/CODEX_TEXT_ONLY_CASE_STUDY.md), plus the complete per-call outputs,
-token counts, source hashes, runtime settings, and limitations in the
+| Provider-reported full input, 6 calls | 277,871 | 243,210 | **12.47%** |
+| Visible bundle estimate | 4,626 | 303 | 93.45% |
+
+Shell access, plugins, and skill search were not disabled. The benchmark's
+standard profile did disable apps, memories, and multi-agent execution for
+reproducibility, identically on both sides; the measured 12.47% delta comes
+from the selected bundle, not from changing that runtime profile. This is one
+synthetic two-task pilot, not proof that every repository can remove 12%.
+
+See the [component-selection case study](docs/CODEX_CONTEXT_SELECTION_CASE_STUDY.md),
+the [manifest and tool fixtures](examples/context_bundles/tool_workflows/), and
+the [complete final audit](examples/context_bundles/tool_workflows/selection.codex-standard.3x.2026-08-18.json).
+
+### Earlier result: text-only capability selection
+
+For pre-bundled decisions that need no files or tools, the explicit
+`text-only/v1` profile reduced full input by 10.55% and 10.60% across two
+synthetic workloads, with 42/42 expected decisions per profile. That remains a
+narrow capability-selection result, not the main product claim. A
+10.3%-shorter instruction rewrite had reduced complete Codex input by only
+about 0.25%, which is why denser now targets whole visible context bundles.
+
+See the earlier [text-only case study](docs/CODEX_TEXT_ONLY_CASE_STUDY.md) and
 [`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
+### Earlier 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
@@ -102,6 +114,11 @@ end-to-end input usage separately from asset-only length.
 ## What denser does
 
 ```bash
+denser minimize-context context-bundle.json --suite replay.json \
+  --backend codex-cli --codex-capability-profile standard \
+  --selection-trials 1 --validation-trials 3 --parallelism 6 \
+  --out selected-context.md --json-out selection-evidence.json
+
 denser audit AGENTS.md AGENTS.variant.md --type claude_md \
   --suite replay.holdout.json \
   --negative-control AGENTS.negative-control.md \
@@ -121,7 +138,15 @@ denser replay --type claude_md AGENTS.md --suite replay.json \
   --backend codex-cli --codex-capability-profile text-only
 ```
 
-`audit` is the primary interface. It runs paired baseline/variant replay,
+`minimize-context` is the automatic selective-loading interface. A versioned
+manifest names the visible context components, marks non-removable components,
+and declares which required component to drop for the known-bad control. The
+selector tests optional components largest first, retains every uncertain or
+behavior-changing component, then repeats a final audit. It operates on
+user-supplied visible text; it does not inspect hidden provider prefixes or
+compact conversation history.
+
+`audit` is the primary lower-level interface. It runs paired baseline/variant replay,
 compares every covered case, checks whether a known-bad negative control causes
 a regression, and reports both asset-only estimates and provider-reported full
 input usage. Equal scores without a detected negative control are

denser/__init__.py

@@ -18,6 +18,16 @@
     audit_context,
 )
 from denser.compress import CompressionResult, compress
+from denser.context_selection import (
+    CONTEXT_BUNDLE_SCHEMA_VERSION,
+    CONTEXT_SELECTION_SCHEMA_VERSION,
+    ComponentAttempt,
+    ContextBundle,
+    ContextComponent,
+    ContextSelectionReport,
+    load_context_bundle,
+    minimize_context,
+)
 from denser.curve import DensityCurve, DensityPoint, curve
 from denser.eval import (
     CaseResult,
@@ -96,9 +106,15 @@
     "CaseResult",
     "ComparisonReport",
     "CompressionResult",
+    "CONTEXT_BUNDLE_SCHEMA_VERSION",
+    "CONTEXT_SELECTION_SCHEMA_VERSION",
+    "ComponentAttempt",
     "ContractCategory",
     "ContractItem",
     "ContractItemResult",
+    "ContextBundle",
+    "ContextComponent",
+    "ContextSelectionReport",
     "ContextAuditReport",
     "DensityCurve",
     "DensityPoint",
@@ -147,9 +163,11 @@
     "evaluate",
     "inspect",
     "load_golden_tasks",
+    "load_context_bundle",
     "load_replay_suite",
     "load_replay_tasks",
     "optimize",
+    "minimize_context",
     "replay",
     "verify",
 ]

denser/audit.py

@@ -213,6 +213,7 @@ def audit_context(
     n_trials: int = 1,
     seed: int = 0,
     on_progress: Callable[[ReplayProgress], None] | None = None,
+    parallelism: int = 1,
 ) -> ContextAuditReport:
     """Audit a context variant and require a sensitive suite for a positive verdict.
 
@@ -232,6 +233,7 @@ def audit_context(
         n_trials=n_trials,
         seed=seed,
         on_progress=on_progress,
+        parallelism=parallelism,
     )
     comparison = ReplayComparisonReport(
         task_type=tt,

denser/backends/codex_cli.py

@@ -13,6 +13,7 @@
 import re
 import shutil
 import subprocess
+import threading
 import time
 from dataclasses import dataclass
 from pathlib import Path
@@ -269,7 +270,7 @@ def __init__(
         self._disabled_features = _DISABLED_FEATURES_BY_PROFILE[capability_profile]
         self._cli_version: str | None = None
         self._cli_version_checked = False
-        self._last_call_metadata: CodexCliCallMetadata | None = None
+        self._call_state = threading.local()
 
     @property
     def name(self) -> str:
@@ -281,6 +282,11 @@ def supports_caching(self) -> bool:
         """Report that this adapter does not expose explicit prompt caching."""
         return False
 
+    @property
+    def supports_concurrency(self) -> bool:
+        """Report that per-call metadata is isolated for concurrent replays."""
+        return True
+
     @property
     def runtime_config(self) -> dict[str, object]:
         """Return the reproducibility settings safe to include in reports."""
@@ -306,9 +312,13 @@ def runtime_config(self) -> dict[str, object]:
     @property
     def last_call_metadata(self) -> dict[str, object] | None:
         """Return sanitized evidence for the most recent invocation."""
-        if self._last_call_metadata is None:
+        metadata = getattr(self._call_state, "last_call_metadata", None)
+        if not isinstance(metadata, CodexCliCallMetadata):
             return None
-        return self._last_call_metadata.to_dict()
+        return metadata.to_dict()
+
+    def _set_last_call_metadata(self, metadata: CodexCliCallMetadata | None) -> None:
+        self._call_state.last_call_metadata = metadata
 
     def _get_cli_version(self) -> str | None:
         if self._cli_version_checked:
@@ -392,20 +402,24 @@ def _run(self, command: list[str], user: str) -> subprocess.CompletedProcess[str
             )
         except subprocess.TimeoutExpired as exc:
             duration_ms = round((time.monotonic() - started) * 1000)
-            self._last_call_metadata = CodexCliCallMetadata(
-                status="timeout",
-                exit_code=None,
-                duration_ms=duration_ms,
+            self._set_last_call_metadata(
+                CodexCliCallMetadata(
+                    status="timeout",
+                    exit_code=None,
+                    duration_ms=duration_ms,
+                )
             )
             raise BackendError(
                 f"Codex CLI timed out after {self._timeout_seconds:g} seconds"
             ) from exc
         except OSError as exc:
             duration_ms = round((time.monotonic() - started) * 1000)
-            self._last_call_metadata = CodexCliCallMetadata(
-                status="launch_error",
-                exit_code=None,
-                duration_ms=duration_ms,
+            self._set_last_call_metadata(
+                CodexCliCallMetadata(
+                    status="launch_error",
+                    exit_code=None,
+                    duration_ms=duration_ms,
+                )
             )
             raise BackendError(f"Codex CLI could not be launched: {type(exc).__name__}") from exc
 
@@ -430,7 +444,7 @@ def complete(
 
         command = self._build_command(system)
         started = time.monotonic()
-        self._last_call_metadata = None
+        self._set_last_call_metadata(None)
         completed = self._run(command, user)
         duration_ms = round((time.monotonic() - started) * 1000)
         parsed = _parse_events(completed.stdout, completed.stderr)
@@ -441,12 +455,14 @@ def complete(
             and parsed.final_message is not None
             else "failed"
         )
-        self._last_call_metadata = CodexCliCallMetadata(
-            status=status,
-            exit_code=completed.returncode,
-            duration_ms=duration_ms,
-            usage=parsed.usage,
-            transport_fallback=parsed.transport_fallback,
+        self._set_last_call_metadata(
+            CodexCliCallMetadata(
+                status=status,
+                exit_code=completed.returncode,
+                duration_ms=duration_ms,
+                usage=parsed.usage,
+                transport_fallback=parsed.transport_fallback,
+            )
         )
         if status != "completed":
             raise BackendError(