feat(adaptive): add bounded algorithm scout

.github/workflows/build_and_test.yml

@@ -180,6 +180,7 @@ jobs:
             tests.test_adaptive_evidence_contracts \
             tests.test_adaptive_environment_profile \
             tests.test_adaptive_bundle_registry \
+            tests.test_adaptive_algorithm_scout \
             tests.test_adaptive_managed_output \
             tests.test_adaptive_qualification \
             tests.test_adaptive_recommendation \
@@ -249,6 +250,7 @@ jobs:
             tests.test_adaptive_vision_roadmap_generator \
             tests.test_adaptive_environment_profile \
             tests.test_adaptive_bundle_registry \
+            tests.test_adaptive_algorithm_scout \
             tests.test_adaptive_managed_output \
             tests.test_adaptive_qualification \
             tests.test_adaptive_recommendation \
@@ -332,6 +334,7 @@ jobs:
             tests/test_generated_cli_reference.py \
             tests/test_ssot_capability_coverage.py \
             tests/test_adaptive_vision_roadmap_generator.py \
+            tests/test_adaptive_algorithm_scout.py \
             tests/test_adaptive_evidence_activation.py \
             tests/test_adaptive_qualification.py \
             tests/test_adaptive_recommendation.py \

.github/workflows/publish.yml

@@ -43,6 +43,7 @@ jobs:
             tests.test_adaptive_evidence_activation \
             tests.test_adaptive_environment_profile \
             tests.test_adaptive_bundle_registry \
+            tests.test_adaptive_algorithm_scout \
             tests.test_adaptive_managed_output \
             tests.test_adaptive_recommendation \
             tests.test_adaptive_processing \
@@ -69,6 +70,8 @@ jobs:
           required = {
             "yolozu/data/manifest/adaptive_vision_roadmap.json",
             "yolozu/data/schemas/adaptive_vision_roadmap.schema.json",
+            "yolozu/data/schemas/algorithm_scout_sources.schema.json",
+            "yolozu/data/schemas/algorithm_scout_report.schema.json",
             "yolozu/data/schemas/local_artifact_inventory.schema.json",
             "yolozu/data/schemas/qualification_report.schema.json",
             "yolozu/data/schemas/evidence_activation_record.schema.json",
@@ -198,6 +201,7 @@ jobs:
             tests.test_adaptive_evidence_activation \
             tests.test_adaptive_environment_profile \
             tests.test_adaptive_bundle_registry \
+            tests.test_adaptive_algorithm_scout \
             tests.test_adaptive_managed_output \
             tests.test_adaptive_recommendation \
             tests.test_adaptive_processing \
@@ -222,6 +226,8 @@ jobs:
           required = {
             "yolozu/data/manifest/adaptive_vision_roadmap.json",
             "yolozu/data/schemas/adaptive_vision_roadmap.schema.json",
+            "yolozu/data/schemas/algorithm_scout_sources.schema.json",
+            "yolozu/data/schemas/algorithm_scout_report.schema.json",
             "yolozu/data/schemas/local_artifact_inventory.schema.json",
             "yolozu/data/schemas/qualification_report.schema.json",
             "yolozu/data/schemas/evidence_activation_record.schema.json",

CHANGELOG.md

@@ -12,6 +12,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - Add Experimental dry-run-by-default pinned local image processing through MCP.
 - Verify adaptive routing boundaries across source, candidate artifacts, and installed MCP calls.
 - Register the existing model zoo as non-promoted adaptive Candidate baselines.
+- Add an Experimental bounded official-source algorithm scout and nonselectable candidate inbox.
 
 ### Fixed
 - Keep repository-wrapper prediction and TTA/TTT log paths anchored to the checkout when invoked from another working directory.

README.md

@@ -129,6 +129,12 @@ non-promoted Candidate baselines matching the existing model zoo: YOLOX-S,
 Detectron2 Faster R-CNN R50-FPN 1x, and MMDetection Faster R-CNN R50-FPN 1x.
 Their pinned weights are fetchable metadata, but their adaptive execution binding
 is explicitly unbound. Loading and validation do not import a model runtime.
+Experimental `yolozu scout-algorithms` now validates one canonical official-source
+allowlist and, only with `--collect`, writes a dated candidate inbox. The default
+command is a network-free, write-free plan. Collected pages are untrusted metadata;
+raw documents are not retained, and the inbox cannot be loaded as an
+AlgorithmBundle registry or used as qualification, support, recommendation,
+adoption, or promotion evidence.
 Explicit workspace catalogs remain operator-asserted and nonselectable. The Experimental
 `yolozu qualify-image-pipeline` command now implements pinned no-follow input
 and asset preflight, a frozen repeat/soak protocol, bounded child-process
@@ -164,6 +170,8 @@ qualification evidence or proof of human adoption.
 
 The [baseline bundle registry report](reports/adaptive_baseline_bundle_registry_2026-08-26.md)
 records the current three-way state boundary. The earlier
+[algorithm scout foundation report](reports/adaptive_algorithm_scout_foundation_2026-08-26.md)
+records the monitored-source, retention, parser, and nonselection boundaries. The earlier
 [installed-artifact verification report](reports/adaptive_routing_installed_verification_2026-08-26.md)
 checks the same boundary from source, sdist, wheel, and installed MCP calls. Its
 positive selector and executor cases use internal fixtures; they do not qualify a

Readme_jp.md

@@ -128,7 +128,12 @@ privacy-safe な live `environment_profile` を返します。probe failure は
 MMDetection Faster R-CNN R50-FPN 1xの3件を、未昇格のCandidate baselineとして
 登録しています。固定済みweightは取得可能なmetadataですが、adaptive execution
 bindingは明示的にunboundです。model runtimeをimportせずに検証して読み込めます。明示したworkspace
-catalogはoperator-assertedのままで、選択対象にはなりません。POSIX専用の
+catalogはoperator-assertedのままで、選択対象にはなりません。
+Experimental `yolozu scout-algorithms` はcanonicalなofficial-source allowlistだけを
+検証し、`--collect`を明示した場合だけ日付付きcandidate inboxを作ります。defaultは
+network-freeかつwrite-freeのplanです。取得した内容はuntrusted metadataとして扱い、raw
+documentは保持しません。このinboxをAlgorithmBundle registryとしてloadしたり、qualification、
+support、recommendation、adoption、promotionの証拠として使うことはできません。POSIX専用の
 Experimental `yolozu qualify-image-pipeline` commandは、pinned no-follow input/asset
 preflight、固定したrepeat/soak protocol、child processのbounded cancellation、
 unactivatedな`qualification_report.json`のatomic publicationを実装しています。
@@ -162,6 +167,8 @@ qualification evidenceにもhuman adoptionの証明にもなりません。
 
 [baseline bundle registry report](reports/adaptive_baseline_bundle_registry_2026-08-26.md)
 に現在の3段階の境界を記録しています。先行する
+[algorithm scout foundation report](reports/adaptive_algorithm_scout_foundation_2026-08-26.md)
+にはmonitored-source、retention、parser、nonselectionの境界を記録しています。先行する
 [installed-artifact verification report](reports/adaptive_routing_installed_verification_2026-08-26.md)
 では、source、sdist、wheel、installed MCP callで同じ境界を確認しています。positiveな
 selector/executor caseは内部fixtureによるもので、実bundleのqualificationやselectedな

docs/README.md

@@ -56,6 +56,9 @@ The environment-qualified local image-processing program targets an Experimental
   unbound execution. The evidence stream and runner maps are empty, so the
   default response records `maturity_disallowed`, abstains, and cannot run a real
   adaptive model.
+- Monitored-source inbox: [`algorithm_intake/README.md`](algorithm_intake/README.md).
+  `yolozu scout-algorithms` is network-free and write-free by default. Explicit
+  collection produces only a dated nonselectable candidate report.
 
 ## Primary Focus
 

docs/adaptive_image_routing.md

@@ -69,6 +69,16 @@ fails with an actionable error and does
 not create synthetic or no-op qualification evidence. It is not a selector,
 model adapter, or general image-processing capability.
 
+The Experimental `yolozu scout-algorithms` surface is a separate candidate inbox.
+It accepts only `docs/algorithm_intake/sources.json`. Without `--collect`, it
+validates that file and prints a bounded JSON plan without network access or writes.
+Collection uses credential-free HTTPS on port 443 for the exact scheme/host/path
+allowlist, retains summaries and provenance rather than raw source documents, and
+records unavailable fields as `unknown`. A failed or missed source remains explicit.
+The report kind is not an AlgorithmBundle registry interface contract, and the
+selector cannot load it. Discovery therefore cannot register, qualify, recommend,
+execute, or promote a model.
+
 `selected` means that one registered pipeline survived every hard filter, matched
 one active trusted qualification record for the exact measured configuration, and
 ranked first under the requested policy. `abstained` is a valid routing outcome when

docs/algorithm_intake/README.md

@@ -0,0 +1,53 @@
+# Experimental algorithm scout
+
+`yolozu scout-algorithms` is a maintainer-only monitored-source inbox. It does
+not search the whole web or establish which algorithm is latest or best. The
+only accepted source file is `docs/algorithm_intake/sources.json`.
+
+The safe default validates the allowlist and prints a bounded JSON plan. It does
+not open a network connection or create the output directory.
+
+```bash
+yolozu scout-algorithms \
+  --sources docs/algorithm_intake/sources.json \
+  --output-dir reports/algorithm_scout \
+  --collection-date 2026-08-26 \
+  --trigger workflow_dispatch
+```
+
+`--collect` enables the only network and write path. Collection accepts exact
+allowlisted HTTPS scheme/host/path records on port 443. It rejects credentials,
+queries, fragments, IP literals, non-public DNS or peer addresses, and redirects
+outside the same explicit allowlist. No caller headers, cookies, or tokens are
+accepted. TLS 1.2 is the minimum protocol version; normal certificate-chain,
+hostname, and SNI validation remain enabled. A source gets 30 seconds, the run
+gets 12 minutes, and the surrounding
+15-minute workflow must keep three minutes for report finalization and failure
+handling.
+
+Fetched HTML, text, JSON, and PDF are untrusted input. The parser applies the
+document, process, memory, time, and byte caps recorded in the Bead and report.
+It never follows URLs found in fetched content, executes code, downloads weights,
+installs packages, or copies raw pages into the repository. The dated managed
+output retains bounded summaries, content provenance, explicit failed or missed
+status, and `unknown` for unavailable metadata. A prior candidate absent from the
+current response is retained only as `historical`, never relabeled as freshly
+collected.
+
+Exit codes are:
+
+- `0`: a valid report with no enabled-source failure;
+- `3`: a valid report with at least one failed or missed source;
+- `2`: invalid or unsafe input rejected before collection; and
+- `1`: bounded internal or finalization failure.
+
+Retention follows repository history policy for the dated metadata reports. Raw
+source documents, headers, response bodies, local paths, credentials, and parser
+temporary files are not retained. Repeated runs merge candidate history and
+deduplicate by source URL plus version or revision.
+
+The output kind is `yolozu_algorithm_scout_report` with
+`selectability=inbox_only`. It is deliberately different from the AlgorithmBundle
+registry interface contract. A later reviewed screening and registry action is
+required before any candidate can enter lifecycle, qualification, selection, or
+execution paths.

docs/algorithm_intake/sources.json

@@ -0,0 +1,126 @@
+{
+  "kind": "yolozu_algorithm_scout_sources",
+  "schema_version": 1,
+  "sources": [
+    {
+      "source_id": "detectron2-releases",
+      "enabled": true,
+      "location": {
+        "scheme": "https",
+        "host": "api.github.com",
+        "path": "/repos/facebookresearch/detectron2/releases"
+      },
+      "redirect_allowlist": [],
+      "identity": {
+        "project": "facebookresearch/detectron2",
+        "model": "release inbox"
+      },
+      "tasks": [
+        "object_detection",
+        "instance_segmentation"
+      ],
+      "availability": {
+        "local": "unknown",
+        "hosted": "unknown"
+      },
+      "license": {
+        "status": "unknown",
+        "expression": "unknown"
+      },
+      "weights": {
+        "status": "unknown"
+      },
+      "runtime_hints": []
+    },
+    {
+      "source_id": "mmdetection-releases",
+      "enabled": true,
+      "location": {
+        "scheme": "https",
+        "host": "api.github.com",
+        "path": "/repos/open-mmlab/mmdetection/releases"
+      },
+      "redirect_allowlist": [],
+      "identity": {
+        "project": "open-mmlab/mmdetection",
+        "model": "release inbox"
+      },
+      "tasks": [
+        "object_detection",
+        "instance_segmentation"
+      ],
+      "availability": {
+        "local": "unknown",
+        "hosted": "unknown"
+      },
+      "license": {
+        "status": "unknown",
+        "expression": "unknown"
+      },
+      "weights": {
+        "status": "unknown"
+      },
+      "runtime_hints": []
+    },
+    {
+      "source_id": "ultralytics-releases",
+      "enabled": true,
+      "location": {
+        "scheme": "https",
+        "host": "api.github.com",
+        "path": "/repos/ultralytics/ultralytics/releases"
+      },
+      "redirect_allowlist": [],
+      "identity": {
+        "project": "ultralytics/ultralytics",
+        "model": "release inbox"
+      },
+      "tasks": [
+        "object_detection",
+        "instance_segmentation",
+        "pose_estimation"
+      ],
+      "availability": {
+        "local": "unknown",
+        "hosted": "unknown"
+      },
+      "license": {
+        "status": "unknown",
+        "expression": "unknown"
+      },
+      "weights": {
+        "status": "unknown"
+      },
+      "runtime_hints": []
+    },
+    {
+      "source_id": "yolox-releases",
+      "enabled": true,
+      "location": {
+        "scheme": "https",
+        "host": "api.github.com",
+        "path": "/repos/Megvii-BaseDetection/YOLOX/releases"
+      },
+      "redirect_allowlist": [],
+      "identity": {
+        "project": "Megvii-BaseDetection/YOLOX",
+        "model": "release inbox"
+      },
+      "tasks": [
+        "object_detection"
+      ],
+      "availability": {
+        "local": "unknown",
+        "hosted": "unknown"
+      },
+      "license": {
+        "status": "unknown",
+        "expression": "unknown"
+      },
+      "weights": {
+        "status": "unknown"
+      },
+      "runtime_hints": []
+    }
+  ]
+}

docs/generated/cli_reference.md

@@ -6,10 +6,10 @@ Keep narrative docs short and link here for the full command surface.
 ## Top-level `yolozu --help`
 
 ```text
-usage: yolozu [-h] [--version] {guide,doctor,dr,list,fetch,export,export-dataset,predict-images,eval-coco,calibrate,eval-long-tail,long-tail-recipe,benchmark,parity,predictions,validate,eval-instance-seg,onnxrt,resources,migrate,import,train,train-orchestrate,test,demo,qualify-image-pipeline,activate-qualification-evidence,registry,completion,comp} ...
+usage: yolozu [-h] [--version] {guide,doctor,dr,list,fetch,export,export-dataset,predict-images,eval-coco,calibrate,eval-long-tail,long-tail-recipe,benchmark,parity,predictions,validate,eval-instance-seg,onnxrt,resources,migrate,import,train,train-orchestrate,test,demo,qualify-image-pipeline,activate-qualification-evidence,scout-algorithms,registry,completion,comp} ...
 
 positional arguments:
-  {guide,doctor,dr,list,fetch,export,export-dataset,predict-images,eval-coco,calibrate,eval-long-tail,long-tail-recipe,benchmark,parity,predictions,validate,eval-instance-seg,onnxrt,resources,migrate,import,train,train-orchestrate,test,demo,qualify-image-pipeline,activate-qualification-evidence,registry,completion,comp}
+  {guide,doctor,dr,list,fetch,export,export-dataset,predict-images,eval-coco,calibrate,eval-long-tail,long-tail-recipe,benchmark,parity,predictions,validate,eval-instance-seg,onnxrt,resources,migrate,import,train,train-orchestrate,test,demo,qualify-image-pipeline,activate-qualification-evidence,scout-algorithms,registry,completion,comp}
     guide               Show beginner-friendly routes and copy-paste commands.
     doctor (dr)         Check the environment. Use --explain for beginner-friendly next actions.
     list                List registries and built-in catalogs.
@@ -38,6 +38,7 @@ positional arguments:
                         Measure one exact local image bundle (Experimental).
     activate-qualification-evidence
                         Review one exact qualification report; dry-run unless --approve is set.
+    scout-algorithms    Plan or collect a bounded monitored-source candidate inbox (Experimental).
     registry            AI-first tool registry: list/show/validate/run tools from the canonical manifest.
     completion (comp)   Print shell completion script (bash/zsh).
 
@@ -284,6 +285,7 @@ Contact: develop@toppymicros.com
 | run_trt_pipeline | experimental | tools/run_trt_pipeline.py | Orchestrate the YOLO26 TensorRT pipeline (engine build → predictions export → parity → eval_suite → latency report). |
 | run_ttt_compare | research | tools/run_ttt_compare.py | Run a fail-closed seeded TTT comparison with full checkpoint preflight, real COCO evaluation, adaptation-cost counters, and a concise detector-response boilerplate for selected foreground class/box consistency. |
 | run_ttt_evidence_suite | research | tools/run_ttt_evidence_suite.py | Run and aggregate a fail-closed five-method clean/shifted TTT matrix for at least three seeds. |
+| scout_algorithms | experimental | tools/scout_algorithms.py | Plan by default or collect a bounded read-only Experimental inbox from exact official HTTPS sources; discovered data cannot execute, download weights, mutate the bundle registry, qualify, or promote anything. |
 | smoke | stable | scripts/smoke.sh | Run one-command offline smoke flow (doctor -> validate dataset -> validate predictions -> eval-coco dry-run -> synthgen intake smoke) with optional deep walkthrough checks. |
 | smoke_synthgen | experimental | tools/smoke_synthgen.py | Run deterministic SynthGen intake smoke, or generate and qualify a fresh cross-repo handoff with strict QA and loader checks. |
 | summarize_bop19_pose_evidence | research | tools/summarize_bop19_pose_evidence.py | Combine official BOP19 VSD/MSSD/MSPD scores with matched rotation, translation, ADD, ADD-S, symmetry-aware pose success, and optional independent reproduction. |