aiming-lab · QianhuiWu · Sep 16, 2026 · May 28, 2026 · Sep 12, 2026 · Sep 12, 2026
diff --git a/.assets-revision b/.assets-revision
@@ -1,72 +1,18 @@
-# Hugging Face dataset holding webharbor's heavy static assets
-# (instance_seed/*.db, static/images/, static/external_cache/).
-#
-# fetch_assets.sh uses the `hf download` CLI. The pin below
-# is a git revision (branch name like `main`, a tag, or a specific commit
-# sha). Override at runtime with the ASSETS_REVISION env var.
-#
-# The revision is a commit sha, never a moving branch name.
-#
-# Current pin = c32018ca3b3d67e7b858b1b85fb101aea5090cd7, the head commit of HF
-# dataset `main` (verified 2026-09-15): the squash-merge commit of HF dataset PR
-# #91 "berkeley: synthetic imagery bundle (164 files)" (merged 2026-09-15T04:00:54Z),
-# which added berkeley.tar.gz (sha256
-# ab9d2716ae8d06540a181b5e60c37f613d87b103864b467511da546b1b173789,
-# 6951483 bytes, 171 managed members) — the site's first archive, so that
-# fetch_assets.sh can install sites/berkeley/static/images/.
-#
-# The berkeley bundle first shipped through the interim pin `refs/pr/91` while that
-# dataset PR was open; the pin was moved to this merged commit once the PR merged,
-# because a `refs/pr/<n>` ref moves whenever its branch is updated. Checked at both
-# pin changes (2026-09-14 and 2026-09-15): for all 29 sites registered before
-# berkeley, the archive size and LFS oid on this commit are identical to the
-# b7e605c0 pin below, and all 30 site archives on this commit are byte-identical to
-# the ones on `refs/pr/91` (verified by LFS oid and by comparing the downloaded
-# berkeley.tar.gz byte for byte), so the pin change alters no site's assets. This
-# revision carries an archive for all 30 registered sites.
-#
-# Previous pin = b7e605c0ec5fc47de85b09e7427162cc50e38980, the commit HF dataset
-# `main` pointed at before this pin (verified 2026-09-14): the squash-merge commit
-# of HF dataset PR #85 ("Upload nvidia.tar.gz with distinct official product and
-# hero images (#107 follow-up)"), which sits on top of the merged PR #84 ("Upload
-# repacked nvidia.tar.gz with de-duplicated official product images (#107)"), PR
-# #75 ("Add reviewed NVIDIA asset bundle") and PR #74 ("Upload kaggle.tar.gz with
-# huggingface_hub").
-#
-# Its nvidia.tar.gz is the repacked bundle (sha256
-# 617a3e3740ba6706bcab786c8a5c3f9a22ecbb39eff5728ad2c12e4992cb098b, 16340955 bytes,
-# 37 file members and no directory entries), which carries the official NVIDIA
-# renders for 33 product/news images plus three dedicated hero images under
-# static/images/heroes/: 36 images with 36 distinct sha256, so
-# sites/nvidia/static/images/ has no byte-identical duplicate and no page reuses
-# one file twice. It supersedes the 9927312-byte PR #84 archive (sha256
-# ee8c6ba966e7a8f7fb5ad2d7ff0134ab98e7b80d6cc77f3328217405b8b34e2f) and the
-# original 6706395-byte bundle (blob 0f1d8068af602748d0dec19ef04b52c7559fef28),
-# which carried duplicates.
-#
-# Pin before that = 64264d065cdb0b7755ee99dab356be5733d9ddef, the commit HF dataset
-# `main` pointed at (verified 2026-09-13). It carries the merged HF dataset PR
-# #74 ("Upload kaggle.tar.gz with huggingface_hub") and the later PR #75 ("Add
-# reviewed NVIDIA asset bundle").
-#
-# Why these pins and not the other candidate (refs/pr/37, resolved
-# 600a3e1de158ae56dc82a5fab56c2ca25acb1e27):
-#   * the pinned revision carries an archive for every registered site (30 of 30 for
-#     c32018ca and for refs/pr/91: the 29 main-site archives plus berkeley.tar.gz;
-#     29 of 29 for b7e605c0), and
-#     the kaggle seed DB it ships was found identical to the seed the site's own
-#     code generates (13/13 tables, row by row) in the PR #106 audit;
-#   * refs/pr/37 ships an older kaggle seed (models table has 10 rows instead of
-#     11: densenet121-chestxray is absent) and its archives cover only 17 of the
-#     30 registered sites, so pinning there would break fetch_assets.sh.
-#   * the pinned revision's kaggle.tar.gz (sha256
-#     dd6f1ab34f99989b300e3366225a8f8b0932d49d033b48f141938d16f1c0607d,
-#     11278106 bytes) passes the repository's own
-#     scripts/validate_asset_archive.py:
-#         [fetch] validated 73 managed members for kaggle
-#     That 73-member pack has no directory entries, so the earlier
-#     `ValueError: unexpected managed path: 'kaggle/static'` rejection (from the
-#     superseded 11278264-byte pack, sha256 c533c283...) no longer applies and
-#     `scripts/fetch_assets.sh` installs sites/kaggle's assets from this pin.
+# Hugging Face dataset holding WebHarbor's heavy static assets.
+# Pin an immutable commit SHA; never a moving branch or refs/pr/<n>.
+#
+# Merged dataset-main commit from HF asset PR #92.
+# https://huggingface.co/datasets/ChilleD/WebHarbor/discussions/92
+# Its complete asset tree matches the tested candidate commit
+# f09e586eec8bf1bca0bc0881e08b77f3c2a5508e.
+#
+# Adds bh_photo.tar.gz: 79,658,793 bytes, 511 managed archive members.
+# SHA-256: 867363d5484eb114d647e236991017992d5ac91ae3415996ad43bf654d99bd9a
+# All 32 pre-existing archives are unchanged from the previous main pin
+# c32018ca3b3d67e7b858b1b85fb101aea5090cd7 (Berkeley, HF PR #91).
+# This revision has archives for all 31 registered sites. The two additional
+# archives (bandcamp and drugs_com) are ignored by fetch_assets.sh.
+# B&H's bundle contains images and external cache; Docker generates its seed
+# from tracked catalog data (sites/bh_photo/.build-generated-seed).
 repo: ChilleD/WebHarbor
-revision: c32018ca3b3d67e7b858b1b85fb101aea5090cd7
+revision: fa1e8a5b9e8e5d0e42764cd658825f4dea088d8f
diff --git a/AGENTS.md b/AGENTS.md
@@ -4,7 +4,7 @@ A coding agent (Claude Code, Cursor, Aider, Codex, ...) is reading this. Read on
 
 ## What it is
 
-30 Flask mirror websites (Amazon, GitHub, BBC News, ...) packaged into one Docker image, plus a control plane on `:8101` for resetting per-site state. Used as a deterministic offline environment for web-agent benchmarks. ~3 GB image.
+31 Flask mirror websites (Amazon, GitHub, BBC News, ...) packaged into one Docker image, plus a control plane on `:8101` for resetting per-site state. Used as a deterministic offline environment for web-agent benchmarks. ~3 GB image.
 
 Two repos:
 - **code** (this one) — Flask apps, control plane, scripts.
@@ -48,17 +48,17 @@ Inside the image, sites live at `/opt/WebSyn/<site>/`. The path predates the ren
 # fresh clone
 ./scripts/fetch_assets.sh                     # pulls assets from HF
 ./scripts/build.sh                            # docker build -t webharbor:dev .
-docker run -d -p 8101:8101 -p 40000-40029:40000-40029 webharbor:dev
+docker run -d -p 8101:8101 -p 40000-40030:40000-40030 webharbor:dev
 ```
 
 Or use the published image directly:
 
 ```bash
-docker run -d -p 8101:8101 -p 40000-40029:40000-40029 \
+docker run -d -p 8101:8101 -p 40000-40030:40000-40030 \
   battalion7244/webharbor:latest
 ```
 
-Sites are on `40000`-`40029` in the order declared by `SITES=( ... )` in `websyn_start.sh`. Control plane:
+Sites are on `40000`-`40030` in the order declared by `SITES=( ... )` in `websyn_start.sh`. Control plane:
 
 | Method | Path                | Purpose                                   |
 |--------|---------------------|-------------------------------------------|
@@ -136,13 +136,13 @@ python3 -m py_compile sites/<site>/app.py
 
 # 3. run on alt ports (don't collide with anything you already have running)
 docker run -d --rm --name wh-test \
-  -p 8201:8101 -p 41000-41029:40000-40029 webharbor:dev
+  -p 8201:8101 -p 41000-41030:40000-40030 webharbor:dev
 
 # 4. control plane healthy, all sites alive
 curl -s http://localhost:8201/health | python3 -m json.tool | head
 
 # 5. every site renders 200
-for p in $(seq 41000 41029); do
+for p in $(seq 41000 41030); do
   curl -so /dev/null -w "$p:%{http_code}\n" http://localhost:$p/
 done
 

diff --git a/CONTRIBUTING.md b/CONTRIBUTING.md
@@ -24,7 +24,7 @@ git clone https://github.com/<you>/webharbor && cd webharbor
 ./scripts/fetch_assets.sh                       # pull current assets
 ./scripts/new_site.py mywebsite                 # OR edit an existing site
 ./scripts/build.sh && docker run -d --rm \
-  -p 8101:8101 -p 40000-40029:40000-40029 webharbor:dev
+  -p 8101:8101 -p 40000-40030:40000-40030 webharbor:dev
 # iterate locally...
 
 ./scripts/extract_assets.sh ../webharbor-static-pr/   # split assets out

diff --git a/Dockerfile b/Dockerfile
@@ -1,5 +1,5 @@
 # WebHarbor — slim, self-contained image.
-# 30 Flask mirror sites + control plane on :8101.
+# 31 Flask mirror sites + control plane on :8101.
 
 FROM python:3.12-slim-bookworm
 
@@ -103,6 +103,16 @@ os.makedirs('instance_seed', exist_ok=True); \
 shutil.copy2('instance/rotten_tomatoes.db', 'instance_seed/rotten_tomatoes.db'); \
 print('Rotten Tomatoes seed DB generated at build time.')" && rm -rf /opt/WebSyn/rotten_tomatoes/instance
 
-EXPOSE 8101 40000-40029
+# B&H's asset bundle contains images; generate the reset seed from its tracked
+# catalog in the image so a fresh checkout needs no locally prepared database.
+RUN python3 /opt/check_asset_inventory.py /opt/WebSyn/bh_photo
+RUN cd /opt/WebSyn/bh_photo && rm -rf instance instance_seed && PYTHONHASHSEED=0 python3 -c "\
+import app; \
+import os, shutil; \
+os.makedirs('instance_seed', exist_ok=True); \
+shutil.copy2('instance/bh_photo.db', 'instance_seed/bh_photo.db'); \
+print('B&H Photo seed DB generated at build time.')" && rm -rf instance
+
+EXPOSE 8101 40000-40030
 
 CMD ["/opt/websyn_start.sh"]
diff --git a/README.md b/README.md
@@ -36,17 +36,17 @@ WebHarbor takes a different approach. We leverage coding agent (e.g., Claude Cod
 - **Deep features unlocked** — carts, checkouts, accounts, all fully testable
 - **Evolving** — harder tasks drive richer mirrors; the environment grows with agents
 - **RL-ready** — sub-second database resets between rollouts
-- **Community-driven** — 30 sites today, scaling to 100+ together
+- **Community-driven** — 31 sites today, scaling to 100+ together
 
 ## 🚀 Quickstart
 
 One command to run all web environments:
 
 ```bash
-docker run -p 8101:8101 -p 40000-40029:40000-40029 battalion7244/webharbor:latest
+docker run -p 8101:8101 -p 40000-40030:40000-40030 battalion7244/webharbor:latest
 ```
 
-Then point your agent at `http://localhost:40000` through `http://localhost:40029` to explore 30 local mirrors of WebVoyager sites: `Allrecipes, Amazon, Apple, ArXiv, BBC News, Booking, GitHub, Google Flights, Google Maps, Google Search, Hugging Face, Wolfram Alpha, Cambridge Dictionary, Coursera, ESPN, Merriam-Webster, IKEA, Phys.org, Target, TED, Ohio State University, Rotten Tomatoes, Compass, Walmart Careers, FedEx, WebMD Doctor, Healthline, Kaggle, NVIDIA, and UC Berkeley`.
+Then point your agent at `http://localhost:40000` through `http://localhost:40030` to explore 31 local mirrors of WebVoyager sites: `Allrecipes, Amazon, Apple, ArXiv, BBC News, Booking, GitHub, Google Flights, Google Maps, Google Search, Hugging Face, Wolfram Alpha, Cambridge Dictionary, Coursera, ESPN, Merriam-Webster, IKEA, Phys.org, Target, TED, Ohio State University, Rotten Tomatoes, Compass, Walmart Careers, FedEx, WebMD Doctor, Healthline, Kaggle, NVIDIA, UC Berkeley, and B&H Photo`.
 
 For sub-second reset between rollouts, expose the control plane and call `/reset/<site>`:
 
@@ -63,29 +63,29 @@ git clone https://github.com/aiming-lab/WebHarbor && cd WebHarbor
 ./scripts/build.sh                                 # docker build -t webharbor:dev .
 ```
 
-### Local NVIDIA review candidate
+### Site registry
 
-This branch registers **30 sites**, the 30 entries listed above; NVIDIA took index 28
-when #107 merged, so UC Berkeley (the site under review here) is the last entry,
-registry index 29, container port 40029 (local review host port 48029). The
-published-image quickstart above is not a claim that this review candidate has been
-published or accepted.
+This checkout registers **31 sites**. NVIDIA remains at index 28, UC Berkeley
+remains at index 29, and B&H Photo is appended at index 30. Build the image from
+this checkout to use this registry; publishing source does not update the
+published Docker image automatically.
 
-| Site | Registry position | Container port | Local review host port |
+| Site | Registry position | Container port | Example local review host port |
 | --- | --- | --- | --- |
 | NVIDIA | 28 | 40028 | 48028 |
 | UC Berkeley | 29 | 40029 | 48029 |
+| B&H Photo | 30 | 40030 | 48030 |
 
 `websyn_start.sh`, `control_server.py`, the `Dockerfile` `EXPOSE` line and every
-site's `tasks.jsonl` `web` URL agree on 30 sites and `40000-40029`;
+site's `tasks.jsonl` `web` URL agree on 31 sites and `40000-40030`;
 `scripts/check_site_registry.py` (run by `scripts/check_assets.sh`) fails when they
 drift.
 
 After preparing the candidate assets and building `webharbor:dev`, the local
 review deployment uses:
 
 ```bash
-docker run -p 127.0.0.1:48080:8101 -p 127.0.0.1:48000-48029:40000-40029 webharbor:dev
+docker run -p 127.0.0.1:48080:8101 -p 127.0.0.1:48000-48030:40000-40030 webharbor:dev
 ```
 
 NVIDIA inherits the site contribution from @KaKituken
@@ -97,28 +97,22 @@ passed.
 
 ### Asset delivery status
 
-`.assets-revision` is pinned to `c32018ca3b3d67e7b858b1b85fb101aea5090cd7`, the
-head commit of HF dataset `main` and the squash-merge commit of HF dataset PR
-[#91](https://huggingface.co/datasets/ChilleD/WebHarbor/discussions/91)
-("berkeley: synthetic imagery bundle (164 files)", merged 2026-09-15T04:00:54Z).
-Every other registered site's archive on that commit has the same size and LFS oid
-as on the previous pin `b7e605c0ec5fc47de85b09e7427162cc50e38980`, and all 30 site
-archives are byte-identical to the ones the interim `refs/pr/91` pin served, so the
-pin change adds the UC Berkeley bundle without altering any other site's assets:
-
-- the pinned revision carries 32 `*.tar.gz` (30 registered sites plus
-  `bandcamp.tar.gz` and `drugs_com.tar.gz`, which `fetch_assets.sh` ignores for
-  sites this checkout does not register);
-- `nvidia.tar.gz` at that revision is the same 37-member archive as at the previous
-  pin, so `scripts/validate_asset_archive.py nvidia.tar.gz nvidia` prints
-  `[fetch] validated 37 managed members for nvidia` and exits 0;
-- `berkeley.tar.gz` at that revision has 171 managed members, so
-  `scripts/validate_asset_archive.py berkeley.tar.gz berkeley` prints
-  `[fetch] validated 171 managed members for berkeley`;
-- `./scripts/fetch_assets.sh` at this pin extracts all 30 registered sites
-  (`[fetch] done — 30 site(s) extracted into sites/`).
-
-The previous pin `b7e605c0ec5fc47de85b09e7427162cc50e38980` is the squash-merge
+`.assets-revision` pins the merged dataset commit `fa1e8a5b9e8e5d0e42764cd658825f4dea088d8f`
+from [HF asset PR #92](https://huggingface.co/datasets/ChilleD/WebHarbor/discussions/92).
+Its complete asset tree matches the tested candidate commit
+`f09e586eec8bf1bca0bc0881e08b77f3c2a5508e`.
+
+This revision adds `bh_photo.tar.gz` and preserves all 32 existing archives from
+`c32018ca3b3d67e7b858b1b85fb101aea5090cd7` byte-for-byte, including Berkeley and
+NVIDIA. It carries 33 archives for 31 registered sites plus the unregistered
+Bandcamp and Drugs.com archives, which `fetch_assets.sh` ignores. A clean asset
+fetch downloaded and extracted all 31 registered sites successfully.
+
+B&H's archive contains images and external cache. The Docker build validates
+its 508 declared assets and generates `instance_seed/bh_photo.db` from the tracked
+catalog. No manually prepared B&H database is required for a fresh build.
+
+The earlier pin `b7e605c0ec5fc47de85b09e7427162cc50e38980` is the squash-merge
 commit of HF dataset PR
 [#85](https://huggingface.co/datasets/ChilleD/WebHarbor/discussions/85) on the
 dataset's `main`. It sits on top of PR
@@ -130,6 +124,7 @@ added the first reviewed NVIDIA bundle.
 | --- | --- | --- | --- |
 | `nvidia.tar.gz` at the current pin | 37 | 16,340,955 | `617a3e3740ba6706bcab786c8a5c3f9a22ecbb39eff5728ad2c12e4992cb098b` |
 | `berkeley.tar.gz` at the current pin (HF PR #91) | 171 | 6,951,483 | `ab9d2716ae8d06540a181b5e60c37f613d87b103864b467511da546b1b173789` |
+| `bh_photo.tar.gz` at the current pin (HF PR #92) | 511 | 79,658,793 | `867363d5484eb114d647e236991017992d5ac91ae3415996ad43bf654d99bd9a` |
 | previous pin's `nvidia.tar.gz` (HF PR #84, superseded) | 34 | 9,927,312 | `ee8c6ba966e7a8f7fb5ad2d7ff0134ab98e7b80d6cc77f3328217405b8b34e2f` |
 
 PR #85 replaces five product images and adds three dedicated hero images (see

diff --git a/control_server.py b/control_server.py
@@ -30,7 +30,7 @@
     'ikea', 'phys_org', 'target', 'ted',
     'osu', 'rotten_tomatoes', 'compass', 'walmart_careers',
     'fedex', 'webmd_doctor', 'healthline', 'kaggle',
-    'nvidia', 'berkeley',
+    'nvidia', 'berkeley', 'bh_photo',
 ]
 BASE_PORT = 40000
 WEBSYN_DIR = '/opt/WebSyn'

diff --git a/sites/bh_photo/.build-generated-seed b/sites/bh_photo/.build-generated-seed
@@ -0,0 +1 @@
+The Dockerfile generates instance_seed/bh_photo.db from the tracked source_catalog.json and seed_data.py.
diff --git a/sites/bh_photo/.gitignore b/sites/bh_photo/.gitignore
@@ -0,0 +1,8 @@
+# Harvested archive HTML and the manifests built from it are build-time
+# intermediates; the runtime catalogue lives in instance_seed/bh_photo.db.
+sources/cache/
+
+# Recreated from instance_seed/ on every boot; not committed.
+instance/
+__pycache__/
+*.pyc
diff --git a/sites/bh_photo/.requires-images b/sites/bh_photo/.requires-images
@@ -0,0 +1 @@
+This site requires static/images from the pinned Hugging Face asset bundle.
diff --git a/sites/bh_photo/_health.py b/sites/bh_photo/_health.py
@@ -0,0 +1,3 @@
+"""Per-site health probe (optional, called by control_server)."""
+def health():
+    return {"ok": True, "site": "bh_photo"}
Original file line number	Diff line number	Diff line change
		@@ -0,0 +1 @@
		The Dockerfile generates instance_seed/bh_photo.db from the tracked source_catalog.json and seed_data.py.
Original file line number	Diff line number	Diff line change
		@@ -0,0 +1 @@
		This site requires static/images from the pinned Hugging Face asset bundle.