search for standard sites
pub-search.waow.tech
search
zig
blog
atproto
feat: rename constellation to atlas and add search
Rename all files, routes, docs, and references from "constellation"
to "atlas" (constellation conflicts with an existing ATProto service).
Add search box to atlas page — type a query, it calls the semantic
search API, matches result URIs to points on the canvas, computes a
weighted centroid, and animates the view to center on the results
with matched points highlighted. Cmd+K to focus, Escape to clear.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
+634
-392
+1
-1
.gitignore
+1
-1
.gitignore
+3
-3
README.md
+3
-3
README.md
···
61
61
62
62
the backend indexes multiple ATProto platforms - currently `pub.leaflet.*` and `site.standard.*` collections. platform is stored per-document and returned in search results.
63
63
64
-
## constellation
64
+
## atlas
65
65
66
-
a 2D semantic map of the entire document index: [pub-search.waow.tech/constellation](https://pub-search.waow.tech/constellation)
66
+
a 2D semantic map of the entire document index: [pub-search.waow.tech/atlas](https://pub-search.waow.tech/atlas)
67
67
68
68
documents are projected from 1024-dim voyage embeddings to 2D via PCA → UMAP, then clustered with HDBSCAN at two granularities. each point is colored by platform. zoom in to see finer cluster labels and individual document titles.
69
69
70
-
built with `scripts/build-constellation` (batch job, ~20s) → `site/constellation.json` → canvas renderer. see [docs/constellation.md](docs/constellation.md) for details.
70
+
built with `scripts/build-atlas` (batch job, ~20s) → `site/atlas.json` → canvas renderer. see [docs/atlas.md](docs/atlas.md) for details.
71
71
72
72
## [stack](https://bsky.app/profile/zzstoatzz.io/post/3mbij5ip4ws2a)
73
73
+1
-1
docs/bridgy-fed.md
+1
-1
docs/bridgy-fed.md
···
8
8
9
9
**attempt 1 (early 2026):** bridgy fed content flooded the index — tens of thousands of short fediverse posts mixed in with long-form articles. search results became polluted with content that wasn't meaningfully "published" in the way leaflet/whitewind/etc. content is. we added `is_bridgyfed` column to turso and marked all bridgy fed documents, then excluded them from search results.
10
10
11
-
**attempt 2 (later):** even with search exclusion, the vectors remained in turbopuffer and polluted semantic search and the constellation visualization. had to run `scripts/purge-bridgyfed-vectors` to clean up ~26k orphan vectors.
11
+
**attempt 2 (later):** even with search exclusion, the vectors remained in turbopuffer and polluted semantic search and the atlas visualization. had to run `scripts/purge-bridgyfed-vectors` to clean up ~26k orphan vectors.
12
12
13
13
**current state:** bridgy fed content is now **dropped at ingest** in the backend. the tap still receives it (can't filter at the firehose level), but the backend's ingest pipeline checks the PDS endpoint and silently drops any DID hosted on `brid.gy`. this is the cleanest solution — no storage, no cleanup needed.
14
14
+9
-9
docs/constellation.md
docs/atlas.md
+9
-9
docs/constellation.md
docs/atlas.md
···
1
-
# constellation
1
+
# atlas
2
2
3
3
2D semantic map of the document index. each document is a point on a canvas, positioned by semantic similarity and colored by platform.
4
4
5
-
**live:** [pub-search.waow.tech/constellation](https://pub-search.waow.tech/constellation)
5
+
**live:** [pub-search.waow.tech/atlas](https://pub-search.waow.tech/atlas)
6
6
7
7
## data pipeline
8
8
9
-
`scripts/build-constellation` is a batch python script (uv inline dependencies) that:
9
+
`scripts/build-atlas` is a batch python script (uv inline dependencies) that:
10
10
11
11
1. **exports vectors** from turbopuffer — paginated query with `rank_by: ["id", "asc"]`, fetches all ~12k vectors + metadata
12
12
2. **PCA 1024 → 50** — denoising pass, typically captures ~60% variance
···
16
16
- fine: `min_cluster_size=20` (~160 clusters, zoomed-in labels)
17
17
- outliers assigned to nearest cluster centroid
18
18
5. **c-TF-IDF** on document titles per cluster → 3-term labels
19
-
6. **outputs** `site/constellation.json` (~3MB, gitignored)
19
+
6. **outputs** `site/atlas.json` (~3MB, gitignored)
20
20
21
21
run time: ~20s. dependencies: `umap-learn`, `hdbscan`, `scikit-learn`, `httpx`, `numpy`, `pydantic-settings`.
22
22
23
23
```bash
24
-
./scripts/build-constellation # writes site/constellation.json
25
-
./scripts/build-constellation -o out.json # custom output path
24
+
./scripts/build-atlas # writes site/atlas.json
25
+
./scripts/build-atlas -o out.json # custom output path
26
26
```
27
27
28
28
## frontend
29
29
30
-
`site/constellation.html` + `site/constellation.js` + `site/constellation.css`
30
+
`site/atlas.html` + `site/atlas.js` + `site/atlas.css`
31
31
32
32
- **canvas 2D** renderer — no libraries, sprite-based (pre-rendered offscreen canvas per platform)
33
33
- **pan/zoom** via wheel, drag, touch/pinch (max 15×)
···
38
38
39
39
## recomputing
40
40
41
-
the constellation is a point-in-time snapshot. rerun the build script when the index changes meaningfully:
41
+
the atlas is a point-in-time snapshot. rerun the build script when the index changes meaningfully:
42
42
43
43
```bash
44
-
./scripts/build-constellation
44
+
./scripts/build-atlas
45
45
cd site && wrangler pages deploy . --project-name leaflet-search
46
46
```
47
47
+356
scripts/build-atlas
+356
scripts/build-atlas
···
1
+
#!/usr/bin/env -S uv run --script --quiet
2
+
# /// script
3
+
# requires-python = ">=3.12"
4
+
# dependencies = ["httpx", "numpy", "scikit-learn", "umap-learn", "hdbscan", "pydantic-settings"]
5
+
# ///
6
+
"""
7
+
Build atlas.json — 2D semantic map of the document index.
8
+
9
+
Exports vectors from turbopuffer, projects to 2D via PCA+UMAP,
10
+
clusters with HDBSCAN at two granularities, labels via c-TF-IDF.
11
+
12
+
Usage:
13
+
./scripts/build-atlas # writes site/atlas.json
14
+
./scripts/build-atlas --output out.json
15
+
"""
16
+
17
+
import argparse
18
+
import json
19
+
import os
20
+
import re
21
+
import sys
22
+
import time
23
+
from collections import Counter
24
+
from pathlib import Path
25
+
26
+
import httpx
27
+
import numpy as np
28
+
from pydantic_settings import BaseSettings, SettingsConfigDict
29
+
from sklearn.decomposition import PCA
30
+
from sklearn.feature_extraction.text import TfidfVectorizer
31
+
32
+
33
+
def log(msg: str) -> None:
34
+
print(msg, flush=True)
35
+
36
+
37
+
class Settings(BaseSettings):
38
+
model_config = SettingsConfigDict(
39
+
env_file=os.environ.get("ENV_FILE", ".env"), extra="ignore"
40
+
)
41
+
42
+
turbopuffer_api_key: str
43
+
turbopuffer_namespace: str = "leaflet-search"
44
+
45
+
@classmethod
46
+
def settings_customise_sources(cls, settings_cls, **kwargs):
47
+
"""Dotenv file wins over environment variables."""
48
+
return (
49
+
kwargs["init_settings"],
50
+
kwargs["dotenv_settings"],
51
+
kwargs["env_settings"],
52
+
kwargs["file_secret_settings"],
53
+
)
54
+
55
+
56
+
def tpuf_export(client: httpx.Client, settings: Settings) -> list[dict]:
57
+
"""Export all vectors from turbopuffer via paginated query."""
58
+
url = f"https://api.turbopuffer.com/v2/namespaces/{settings.turbopuffer_namespace}/query"
59
+
headers = {
60
+
"Authorization": f"Bearer {settings.turbopuffer_api_key}",
61
+
"Content-Type": "application/json",
62
+
}
63
+
attrs = ["uri", "title", "platform", "base_path", "did", "vector"]
64
+
all_rows = []
65
+
last_id = None
66
+
67
+
while True:
68
+
body: dict = {
69
+
"rank_by": ["id", "asc"],
70
+
"limit": 10000,
71
+
"include_attributes": attrs,
72
+
}
73
+
if last_id is not None:
74
+
body["filters"] = ["id", "Gt", last_id]
75
+
76
+
resp = client.post(url, headers=headers, json=body, timeout=120)
77
+
resp.raise_for_status()
78
+
rows = resp.json().get("rows", [])
79
+
if not rows:
80
+
break
81
+
82
+
all_rows.extend(rows)
83
+
last_id = rows[-1]["id"]
84
+
log(f" fetched {len(all_rows)} vectors so far...")
85
+
86
+
if len(rows) < 10000:
87
+
break
88
+
89
+
return all_rows
90
+
91
+
92
+
def extract_terms(title: str) -> list[str]:
93
+
"""Extract lowercase alphanumeric tokens from a title."""
94
+
if not title:
95
+
return []
96
+
return [t for t in re.findall(r"[a-z0-9]+", title.lower()) if len(t) > 1]
97
+
98
+
99
+
def cluster_labels(titles_per_cluster: dict[int, list[str]], n_terms: int = 3) -> dict[int, str]:
100
+
"""Compute c-TF-IDF labels for clusters from document titles."""
101
+
cluster_ids = sorted(titles_per_cluster.keys())
102
+
if not cluster_ids:
103
+
return {}
104
+
105
+
# build one "document" per cluster: concatenated titles
106
+
docs = []
107
+
for cid in cluster_ids:
108
+
docs.append(" ".join(titles_per_cluster[cid]))
109
+
110
+
# TF-IDF across cluster pseudo-documents
111
+
vectorizer = TfidfVectorizer(
112
+
max_features=5000,
113
+
stop_words="english",
114
+
token_pattern=r"[a-z0-9]{2,}",
115
+
lowercase=True,
116
+
)
117
+
try:
118
+
tfidf = vectorizer.fit_transform(docs)
119
+
except ValueError:
120
+
return {cid: f"cluster {cid}" for cid in cluster_ids}
121
+
122
+
feature_names = vectorizer.get_feature_names_out()
123
+
labels = {}
124
+
for i, cid in enumerate(cluster_ids):
125
+
row = tfidf[i].toarray().flatten()
126
+
top_idx = row.argsort()[-n_terms:][::-1]
127
+
top_terms = [feature_names[j] for j in top_idx if row[j] > 0]
128
+
labels[cid] = " ".join(top_terms) if top_terms else f"cluster {cid}"
129
+
130
+
return labels
131
+
132
+
133
+
def assign_outliers(coords_2d: np.ndarray, labels: np.ndarray, centroids: dict[int, np.ndarray]) -> np.ndarray:
134
+
"""Assign outlier points (label == -1) to nearest cluster centroid."""
135
+
result = labels.copy()
136
+
outlier_mask = result == -1
137
+
if not outlier_mask.any() or not centroids:
138
+
return result
139
+
140
+
centroid_ids = sorted(centroids.keys())
141
+
centroid_arr = np.array([centroids[c] for c in centroid_ids])
142
+
outlier_coords = coords_2d[outlier_mask]
143
+
144
+
# nearest centroid for each outlier
145
+
dists = np.linalg.norm(outlier_coords[:, None] - centroid_arr[None, :], axis=2)
146
+
nearest = dists.argmin(axis=1)
147
+
result[outlier_mask] = np.array(centroid_ids)[nearest]
148
+
149
+
return result
150
+
151
+
152
+
def main():
153
+
parser = argparse.ArgumentParser(description="Build atlas.json")
154
+
parser.add_argument(
155
+
"--output", "-o",
156
+
default=str(Path(__file__).resolve().parent.parent / "site" / "atlas.json"),
157
+
help="Output path (default: site/atlas.json)",
158
+
)
159
+
args = parser.parse_args()
160
+
161
+
try:
162
+
settings = Settings() # type: ignore
163
+
except Exception as e:
164
+
print(f"error loading settings: {e}", file=sys.stderr)
165
+
print("required env vars: TURBOPUFFER_API_KEY", file=sys.stderr)
166
+
sys.exit(1)
167
+
168
+
t_start = time.monotonic()
169
+
170
+
# --- step 1: export vectors from turbopuffer ---
171
+
log("exporting vectors from turbopuffer...")
172
+
client = httpx.Client()
173
+
rows = tpuf_export(client, settings)
174
+
client.close()
175
+
log(f" got {len(rows)} documents")
176
+
177
+
if len(rows) < 10:
178
+
print("too few documents to build atlas", file=sys.stderr)
179
+
sys.exit(1)
180
+
181
+
# parse into arrays
182
+
vectors = []
183
+
metadata = []
184
+
for row in rows:
185
+
vec = row.get("vector")
186
+
if not vec:
187
+
continue
188
+
vectors.append(vec)
189
+
metadata.append({
190
+
"uri": row.get("uri", row.get("id", "")),
191
+
"title": row.get("title", ""),
192
+
"platform": row.get("platform", "other"),
193
+
"basePath": row.get("base_path", ""),
194
+
"did": row.get("did", ""),
195
+
})
196
+
197
+
X = np.array(vectors, dtype=np.float32)
198
+
log(f" {X.shape[0]} vectors, {X.shape[1]} dims")
199
+
200
+
# --- step 2: dimensionality reduction ---
201
+
log("PCA 1024 → 50...")
202
+
n_components_pca = min(50, X.shape[0] - 1, X.shape[1])
203
+
pca = PCA(n_components=n_components_pca, random_state=42)
204
+
X_pca = pca.fit_transform(X)
205
+
log(f" variance explained: {pca.explained_variance_ratio_.sum():.2%}")
206
+
207
+
log("UMAP 50 → 2...")
208
+
import umap
209
+
reducer = umap.UMAP(
210
+
n_components=2,
211
+
metric="cosine",
212
+
n_neighbors=15,
213
+
min_dist=0.1,
214
+
random_state=42,
215
+
)
216
+
X_2d = reducer.fit_transform(X_pca)
217
+
218
+
# normalize to [-1, 1] range
219
+
for dim in range(2):
220
+
lo, hi = X_2d[:, dim].min(), X_2d[:, dim].max()
221
+
if hi > lo:
222
+
X_2d[:, dim] = 2 * (X_2d[:, dim] - lo) / (hi - lo) - 1
223
+
224
+
log(f" 2D range: x=[{X_2d[:,0].min():.2f}, {X_2d[:,0].max():.2f}] y=[{X_2d[:,1].min():.2f}, {X_2d[:,1].max():.2f}]")
225
+
226
+
# --- step 3: clustering ---
227
+
import hdbscan
228
+
229
+
log("HDBSCAN coarse (min_cluster_size=100)...")
230
+
clusterer_coarse = hdbscan.HDBSCAN(min_cluster_size=100, min_samples=5)
231
+
labels_coarse_raw = clusterer_coarse.fit_predict(X_2d)
232
+
n_coarse = len(set(labels_coarse_raw)) - (1 if -1 in labels_coarse_raw else 0)
233
+
n_outliers_coarse = (labels_coarse_raw == -1).sum()
234
+
log(f" {n_coarse} clusters, {n_outliers_coarse} outliers")
235
+
236
+
# compute centroids for coarse
237
+
coarse_centroids = {}
238
+
for cid in set(labels_coarse_raw):
239
+
if cid == -1:
240
+
continue
241
+
mask = labels_coarse_raw == cid
242
+
coarse_centroids[cid] = X_2d[mask].mean(axis=0)
243
+
244
+
labels_coarse = assign_outliers(X_2d, labels_coarse_raw, coarse_centroids)
245
+
# recompute centroids after outlier assignment
246
+
for cid in set(labels_coarse):
247
+
mask = labels_coarse == cid
248
+
coarse_centroids[cid] = X_2d[mask].mean(axis=0)
249
+
250
+
log("HDBSCAN fine (min_cluster_size=20)...")
251
+
clusterer_fine = hdbscan.HDBSCAN(min_cluster_size=20, min_samples=3)
252
+
labels_fine_raw = clusterer_fine.fit_predict(X_2d)
253
+
n_fine = len(set(labels_fine_raw)) - (1 if -1 in labels_fine_raw else 0)
254
+
n_outliers_fine = (labels_fine_raw == -1).sum()
255
+
log(f" {n_fine} clusters, {n_outliers_fine} outliers")
256
+
257
+
fine_centroids = {}
258
+
for cid in set(labels_fine_raw):
259
+
if cid == -1:
260
+
continue
261
+
mask = labels_fine_raw == cid
262
+
fine_centroids[cid] = X_2d[mask].mean(axis=0)
263
+
264
+
labels_fine = assign_outliers(X_2d, labels_fine_raw, fine_centroids)
265
+
for cid in set(labels_fine):
266
+
mask = labels_fine == cid
267
+
fine_centroids[cid] = X_2d[mask].mean(axis=0)
268
+
269
+
# --- step 4: labels via c-TF-IDF ---
270
+
log("computing cluster labels (c-TF-IDF on titles)...")
271
+
272
+
coarse_titles: dict[int, list[str]] = {}
273
+
fine_titles: dict[int, list[str]] = {}
274
+
for i, meta in enumerate(metadata):
275
+
title = meta.get("title", "")
276
+
if title:
277
+
coarse_titles.setdefault(int(labels_coarse[i]), []).append(title)
278
+
fine_titles.setdefault(int(labels_fine[i]), []).append(title)
279
+
280
+
coarse_labels = cluster_labels(coarse_titles)
281
+
fine_labels = cluster_labels(fine_titles)
282
+
283
+
# map fine clusters to their parent coarse cluster (majority vote)
284
+
fine_to_coarse = {}
285
+
for fine_id in set(labels_fine):
286
+
fine_mask = labels_fine == fine_id
287
+
coarse_for_fine = labels_coarse[fine_mask]
288
+
counts = Counter(coarse_for_fine)
289
+
fine_to_coarse[int(fine_id)] = int(counts.most_common(1)[0][0])
290
+
291
+
log(f" coarse: {len(coarse_labels)} labels")
292
+
log(f" fine: {len(fine_labels)} labels")
293
+
294
+
# --- step 5: build output ---
295
+
log("building output...")
296
+
297
+
points = []
298
+
for i, meta in enumerate(metadata):
299
+
points.append({
300
+
"x": round(float(X_2d[i, 0]), 4),
301
+
"y": round(float(X_2d[i, 1]), 4),
302
+
"uri": meta["uri"],
303
+
"title": meta["title"],
304
+
"platform": meta["platform"],
305
+
"basePath": meta["basePath"],
306
+
"clusterCoarse": int(labels_coarse[i]),
307
+
"clusterFine": int(labels_fine[i]),
308
+
})
309
+
310
+
coarse_clusters = []
311
+
for cid in sorted(coarse_centroids.keys()):
312
+
count = int((labels_coarse == cid).sum())
313
+
coarse_clusters.append({
314
+
"id": int(cid),
315
+
"label": coarse_labels.get(int(cid), f"cluster {cid}"),
316
+
"cx": round(float(coarse_centroids[cid][0]), 4),
317
+
"cy": round(float(coarse_centroids[cid][1]), 4),
318
+
"count": count,
319
+
})
320
+
321
+
fine_clusters = []
322
+
for cid in sorted(fine_centroids.keys()):
323
+
count = int((labels_fine == cid).sum())
324
+
fine_clusters.append({
325
+
"id": int(cid),
326
+
"label": fine_labels.get(int(cid), f"cluster {cid}"),
327
+
"cx": round(float(fine_centroids[cid][0]), 4),
328
+
"cy": round(float(fine_centroids[cid][1]), 4),
329
+
"count": count,
330
+
"parent": fine_to_coarse.get(int(cid), 0),
331
+
})
332
+
333
+
output = {
334
+
"points": points,
335
+
"clusters": {
336
+
"coarse": coarse_clusters,
337
+
"fine": fine_clusters,
338
+
},
339
+
"meta": {
340
+
"generatedAt": time.strftime("%Y-%m-%dT%H:%M:%SZ", time.gmtime()),
341
+
"nDocuments": len(points),
342
+
},
343
+
}
344
+
345
+
out_path = Path(args.output)
346
+
out_path.parent.mkdir(parents=True, exist_ok=True)
347
+
out_path.write_text(json.dumps(output, separators=(",", ":")))
348
+
349
+
size_kb = out_path.stat().st_size / 1024
350
+
elapsed = time.monotonic() - t_start
351
+
log(f"\nwrote {out_path} ({size_kb:.0f} KB, {len(points)} points)")
352
+
log(f"done in {elapsed:.1f}s")
353
+
354
+
355
+
if __name__ == "__main__":
356
+
main()
-356
scripts/build-constellation
-356
scripts/build-constellation
···
1
-
#!/usr/bin/env -S uv run --script --quiet
2
-
# /// script
3
-
# requires-python = ">=3.12"
4
-
# dependencies = ["httpx", "numpy", "scikit-learn", "umap-learn", "hdbscan", "pydantic-settings"]
5
-
# ///
6
-
"""
7
-
Build constellation.json — 2D semantic map of the document index.
8
-
9
-
Exports vectors from turbopuffer, projects to 2D via PCA+UMAP,
10
-
clusters with HDBSCAN at two granularities, labels via c-TF-IDF.
11
-
12
-
Usage:
13
-
./scripts/build-constellation # writes site/constellation.json
14
-
./scripts/build-constellation --output out.json
15
-
"""
16
-
17
-
import argparse
18
-
import json
19
-
import os
20
-
import re
21
-
import sys
22
-
import time
23
-
from collections import Counter
24
-
from pathlib import Path
25
-
26
-
import httpx
27
-
import numpy as np
28
-
from pydantic_settings import BaseSettings, SettingsConfigDict
29
-
from sklearn.decomposition import PCA
30
-
from sklearn.feature_extraction.text import TfidfVectorizer
31
-
32
-
33
-
def log(msg: str) -> None:
34
-
print(msg, flush=True)
35
-
36
-
37
-
class Settings(BaseSettings):
38
-
model_config = SettingsConfigDict(
39
-
env_file=os.environ.get("ENV_FILE", ".env"), extra="ignore"
40
-
)
41
-
42
-
turbopuffer_api_key: str
43
-
turbopuffer_namespace: str = "leaflet-search"
44
-
45
-
@classmethod
46
-
def settings_customise_sources(cls, settings_cls, **kwargs):
47
-
"""Dotenv file wins over environment variables."""
48
-
return (
49
-
kwargs["init_settings"],
50
-
kwargs["dotenv_settings"],
51
-
kwargs["env_settings"],
52
-
kwargs["file_secret_settings"],
53
-
)
54
-
55
-
56
-
def tpuf_export(client: httpx.Client, settings: Settings) -> list[dict]:
57
-
"""Export all vectors from turbopuffer via paginated query."""
58
-
url = f"https://api.turbopuffer.com/v2/namespaces/{settings.turbopuffer_namespace}/query"
59
-
headers = {
60
-
"Authorization": f"Bearer {settings.turbopuffer_api_key}",
61
-
"Content-Type": "application/json",
62
-
}
63
-
attrs = ["uri", "title", "platform", "base_path", "did", "vector"]
64
-
all_rows = []
65
-
last_id = None
66
-
67
-
while True:
68
-
body: dict = {
69
-
"rank_by": ["id", "asc"],
70
-
"limit": 10000,
71
-
"include_attributes": attrs,
72
-
}
73
-
if last_id is not None:
74
-
body["filters"] = ["id", "Gt", last_id]
75
-
76
-
resp = client.post(url, headers=headers, json=body, timeout=120)
77
-
resp.raise_for_status()
78
-
rows = resp.json().get("rows", [])
79
-
if not rows:
80
-
break
81
-
82
-
all_rows.extend(rows)
83
-
last_id = rows[-1]["id"]
84
-
log(f" fetched {len(all_rows)} vectors so far...")
85
-
86
-
if len(rows) < 10000:
87
-
break
88
-
89
-
return all_rows
90
-
91
-
92
-
def extract_terms(title: str) -> list[str]:
93
-
"""Extract lowercase alphanumeric tokens from a title."""
94
-
if not title:
95
-
return []
96
-
return [t for t in re.findall(r"[a-z0-9]+", title.lower()) if len(t) > 1]
97
-
98
-
99
-
def cluster_labels(titles_per_cluster: dict[int, list[str]], n_terms: int = 3) -> dict[int, str]:
100
-
"""Compute c-TF-IDF labels for clusters from document titles."""
101
-
cluster_ids = sorted(titles_per_cluster.keys())
102
-
if not cluster_ids:
103
-
return {}
104
-
105
-
# build one "document" per cluster: concatenated titles
106
-
docs = []
107
-
for cid in cluster_ids:
108
-
docs.append(" ".join(titles_per_cluster[cid]))
109
-
110
-
# TF-IDF across cluster pseudo-documents
111
-
vectorizer = TfidfVectorizer(
112
-
max_features=5000,
113
-
stop_words="english",
114
-
token_pattern=r"[a-z0-9]{2,}",
115
-
lowercase=True,
116
-
)
117
-
try:
118
-
tfidf = vectorizer.fit_transform(docs)
119
-
except ValueError:
120
-
return {cid: f"cluster {cid}" for cid in cluster_ids}
121
-
122
-
feature_names = vectorizer.get_feature_names_out()
123
-
labels = {}
124
-
for i, cid in enumerate(cluster_ids):
125
-
row = tfidf[i].toarray().flatten()
126
-
top_idx = row.argsort()[-n_terms:][::-1]
127
-
top_terms = [feature_names[j] for j in top_idx if row[j] > 0]
128
-
labels[cid] = " ".join(top_terms) if top_terms else f"cluster {cid}"
129
-
130
-
return labels
131
-
132
-
133
-
def assign_outliers(coords_2d: np.ndarray, labels: np.ndarray, centroids: dict[int, np.ndarray]) -> np.ndarray:
134
-
"""Assign outlier points (label == -1) to nearest cluster centroid."""
135
-
result = labels.copy()
136
-
outlier_mask = result == -1
137
-
if not outlier_mask.any() or not centroids:
138
-
return result
139
-
140
-
centroid_ids = sorted(centroids.keys())
141
-
centroid_arr = np.array([centroids[c] for c in centroid_ids])
142
-
outlier_coords = coords_2d[outlier_mask]
143
-
144
-
# nearest centroid for each outlier
145
-
dists = np.linalg.norm(outlier_coords[:, None] - centroid_arr[None, :], axis=2)
146
-
nearest = dists.argmin(axis=1)
147
-
result[outlier_mask] = np.array(centroid_ids)[nearest]
148
-
149
-
return result
150
-
151
-
152
-
def main():
153
-
parser = argparse.ArgumentParser(description="Build constellation.json")
154
-
parser.add_argument(
155
-
"--output", "-o",
156
-
default=str(Path(__file__).resolve().parent.parent / "site" / "constellation.json"),
157
-
help="Output path (default: site/constellation.json)",
158
-
)
159
-
args = parser.parse_args()
160
-
161
-
try:
162
-
settings = Settings() # type: ignore
163
-
except Exception as e:
164
-
print(f"error loading settings: {e}", file=sys.stderr)
165
-
print("required env vars: TURBOPUFFER_API_KEY", file=sys.stderr)
166
-
sys.exit(1)
167
-
168
-
t_start = time.monotonic()
169
-
170
-
# --- step 1: export vectors from turbopuffer ---
171
-
log("exporting vectors from turbopuffer...")
172
-
client = httpx.Client()
173
-
rows = tpuf_export(client, settings)
174
-
client.close()
175
-
log(f" got {len(rows)} documents")
176
-
177
-
if len(rows) < 10:
178
-
print("too few documents to build constellation", file=sys.stderr)
179
-
sys.exit(1)
180
-
181
-
# parse into arrays
182
-
vectors = []
183
-
metadata = []
184
-
for row in rows:
185
-
vec = row.get("vector")
186
-
if not vec:
187
-
continue
188
-
vectors.append(vec)
189
-
metadata.append({
190
-
"uri": row.get("uri", row.get("id", "")),
191
-
"title": row.get("title", ""),
192
-
"platform": row.get("platform", "other"),
193
-
"basePath": row.get("base_path", ""),
194
-
"did": row.get("did", ""),
195
-
})
196
-
197
-
X = np.array(vectors, dtype=np.float32)
198
-
log(f" {X.shape[0]} vectors, {X.shape[1]} dims")
199
-
200
-
# --- step 2: dimensionality reduction ---
201
-
log("PCA 1024 → 50...")
202
-
n_components_pca = min(50, X.shape[0] - 1, X.shape[1])
203
-
pca = PCA(n_components=n_components_pca, random_state=42)
204
-
X_pca = pca.fit_transform(X)
205
-
log(f" variance explained: {pca.explained_variance_ratio_.sum():.2%}")
206
-
207
-
log("UMAP 50 → 2...")
208
-
import umap
209
-
reducer = umap.UMAP(
210
-
n_components=2,
211
-
metric="cosine",
212
-
n_neighbors=15,
213
-
min_dist=0.1,
214
-
random_state=42,
215
-
)
216
-
X_2d = reducer.fit_transform(X_pca)
217
-
218
-
# normalize to [-1, 1] range
219
-
for dim in range(2):
220
-
lo, hi = X_2d[:, dim].min(), X_2d[:, dim].max()
221
-
if hi > lo:
222
-
X_2d[:, dim] = 2 * (X_2d[:, dim] - lo) / (hi - lo) - 1
223
-
224
-
log(f" 2D range: x=[{X_2d[:,0].min():.2f}, {X_2d[:,0].max():.2f}] y=[{X_2d[:,1].min():.2f}, {X_2d[:,1].max():.2f}]")
225
-
226
-
# --- step 3: clustering ---
227
-
import hdbscan
228
-
229
-
log("HDBSCAN coarse (min_cluster_size=100)...")
230
-
clusterer_coarse = hdbscan.HDBSCAN(min_cluster_size=100, min_samples=5)
231
-
labels_coarse_raw = clusterer_coarse.fit_predict(X_2d)
232
-
n_coarse = len(set(labels_coarse_raw)) - (1 if -1 in labels_coarse_raw else 0)
233
-
n_outliers_coarse = (labels_coarse_raw == -1).sum()
234
-
log(f" {n_coarse} clusters, {n_outliers_coarse} outliers")
235
-
236
-
# compute centroids for coarse
237
-
coarse_centroids = {}
238
-
for cid in set(labels_coarse_raw):
239
-
if cid == -1:
240
-
continue
241
-
mask = labels_coarse_raw == cid
242
-
coarse_centroids[cid] = X_2d[mask].mean(axis=0)
243
-
244
-
labels_coarse = assign_outliers(X_2d, labels_coarse_raw, coarse_centroids)
245
-
# recompute centroids after outlier assignment
246
-
for cid in set(labels_coarse):
247
-
mask = labels_coarse == cid
248
-
coarse_centroids[cid] = X_2d[mask].mean(axis=0)
249
-
250
-
log("HDBSCAN fine (min_cluster_size=20)...")
251
-
clusterer_fine = hdbscan.HDBSCAN(min_cluster_size=20, min_samples=3)
252
-
labels_fine_raw = clusterer_fine.fit_predict(X_2d)
253
-
n_fine = len(set(labels_fine_raw)) - (1 if -1 in labels_fine_raw else 0)
254
-
n_outliers_fine = (labels_fine_raw == -1).sum()
255
-
log(f" {n_fine} clusters, {n_outliers_fine} outliers")
256
-
257
-
fine_centroids = {}
258
-
for cid in set(labels_fine_raw):
259
-
if cid == -1:
260
-
continue
261
-
mask = labels_fine_raw == cid
262
-
fine_centroids[cid] = X_2d[mask].mean(axis=0)
263
-
264
-
labels_fine = assign_outliers(X_2d, labels_fine_raw, fine_centroids)
265
-
for cid in set(labels_fine):
266
-
mask = labels_fine == cid
267
-
fine_centroids[cid] = X_2d[mask].mean(axis=0)
268
-
269
-
# --- step 4: labels via c-TF-IDF ---
270
-
log("computing cluster labels (c-TF-IDF on titles)...")
271
-
272
-
coarse_titles: dict[int, list[str]] = {}
273
-
fine_titles: dict[int, list[str]] = {}
274
-
for i, meta in enumerate(metadata):
275
-
title = meta.get("title", "")
276
-
if title:
277
-
coarse_titles.setdefault(int(labels_coarse[i]), []).append(title)
278
-
fine_titles.setdefault(int(labels_fine[i]), []).append(title)
279
-
280
-
coarse_labels = cluster_labels(coarse_titles)
281
-
fine_labels = cluster_labels(fine_titles)
282
-
283
-
# map fine clusters to their parent coarse cluster (majority vote)
284
-
fine_to_coarse = {}
285
-
for fine_id in set(labels_fine):
286
-
fine_mask = labels_fine == fine_id
287
-
coarse_for_fine = labels_coarse[fine_mask]
288
-
counts = Counter(coarse_for_fine)
289
-
fine_to_coarse[int(fine_id)] = int(counts.most_common(1)[0][0])
290
-
291
-
log(f" coarse: {len(coarse_labels)} labels")
292
-
log(f" fine: {len(fine_labels)} labels")
293
-
294
-
# --- step 5: build output ---
295
-
log("building output...")
296
-
297
-
points = []
298
-
for i, meta in enumerate(metadata):
299
-
points.append({
300
-
"x": round(float(X_2d[i, 0]), 4),
301
-
"y": round(float(X_2d[i, 1]), 4),
302
-
"uri": meta["uri"],
303
-
"title": meta["title"],
304
-
"platform": meta["platform"],
305
-
"basePath": meta["basePath"],
306
-
"clusterCoarse": int(labels_coarse[i]),
307
-
"clusterFine": int(labels_fine[i]),
308
-
})
309
-
310
-
coarse_clusters = []
311
-
for cid in sorted(coarse_centroids.keys()):
312
-
count = int((labels_coarse == cid).sum())
313
-
coarse_clusters.append({
314
-
"id": int(cid),
315
-
"label": coarse_labels.get(int(cid), f"cluster {cid}"),
316
-
"cx": round(float(coarse_centroids[cid][0]), 4),
317
-
"cy": round(float(coarse_centroids[cid][1]), 4),
318
-
"count": count,
319
-
})
320
-
321
-
fine_clusters = []
322
-
for cid in sorted(fine_centroids.keys()):
323
-
count = int((labels_fine == cid).sum())
324
-
fine_clusters.append({
325
-
"id": int(cid),
326
-
"label": fine_labels.get(int(cid), f"cluster {cid}"),
327
-
"cx": round(float(fine_centroids[cid][0]), 4),
328
-
"cy": round(float(fine_centroids[cid][1]), 4),
329
-
"count": count,
330
-
"parent": fine_to_coarse.get(int(cid), 0),
331
-
})
332
-
333
-
output = {
334
-
"points": points,
335
-
"clusters": {
336
-
"coarse": coarse_clusters,
337
-
"fine": fine_clusters,
338
-
},
339
-
"meta": {
340
-
"generatedAt": time.strftime("%Y-%m-%dT%H:%M:%SZ", time.gmtime()),
341
-
"nDocuments": len(points),
342
-
},
343
-
}
344
-
345
-
out_path = Path(args.output)
346
-
out_path.parent.mkdir(parents=True, exist_ok=True)
347
-
out_path.write_text(json.dumps(output, separators=(",", ":")))
348
-
349
-
size_kb = out_path.stat().st_size / 1024
350
-
elapsed = time.monotonic() - t_start
351
-
log(f"\nwrote {out_path} ({size_kb:.0f} KB, {len(points)} points)")
352
-
log(f"done in {elapsed:.1f}s")
353
-
354
-
355
-
if __name__ == "__main__":
356
-
main()
+34
site/constellation.css
site/atlas.css
+34
site/constellation.css
site/atlas.css
···
194
194
border-radius: 50%;
195
195
}
196
196
197
+
/* search */
198
+
.search-form {
199
+
display: flex;
200
+
align-items: center;
201
+
}
202
+
203
+
#search-input {
204
+
background: var(--bg-subtle);
205
+
border: 1px solid var(--tooltip-border);
206
+
color: var(--text);
207
+
font-family: monospace;
208
+
font-size: 11px;
209
+
padding: 3px 8px;
210
+
width: 140px;
211
+
outline: none;
212
+
transition: width 0.2s, border-color 0.2s;
213
+
}
214
+
215
+
#search-input:focus {
216
+
width: 200px;
217
+
border-color: var(--text-dim);
218
+
}
219
+
220
+
#search-input::placeholder {
221
+
color: var(--text-muted);
222
+
}
223
+
224
+
.search-status {
225
+
font-size: 10px;
226
+
color: var(--text-dim);
227
+
margin-left: 6px;
228
+
white-space: nowrap;
229
+
}
230
+
197
231
/* stats */
198
232
.stats {
199
233
position: fixed;
+13
-10
site/constellation.html
site/atlas.html
+13
-10
site/constellation.html
site/atlas.html
···
3
3
<head>
4
4
<meta charset="UTF-8">
5
5
<meta name="viewport" content="width=device-width, initial-scale=1.0, maximum-scale=1.0, user-scalable=no">
6
-
<title>pub search / constellation</title>
6
+
<title>pub search / atlas</title>
7
7
<meta name="description" content="2d semantic map of atproto publishing platforms">
8
-
<meta property="og:title" content="pub search / constellation">
8
+
<meta property="og:title" content="pub search / atlas">
9
9
<meta property="og:description" content="2d semantic map of atproto publishing platforms">
10
10
<meta property="og:type" content="website">
11
-
<meta property="og:image" content="https://pub-search.waow.tech/og-image?page=constellation">
11
+
<meta property="og:image" content="https://pub-search.waow.tech/og-image?page=atlas">
12
12
<meta property="og:image:width" content="1200">
13
13
<meta property="og:image:height" content="630">
14
14
<meta name="twitter:card" content="summary_large_image">
15
-
<meta name="twitter:title" content="pub search / constellation">
15
+
<meta name="twitter:title" content="pub search / atlas">
16
16
<meta name="twitter:description" content="2d semantic map of atproto publishing platforms">
17
-
<meta name="twitter:image" content="https://pub-search.waow.tech/og-image?page=constellation">
17
+
<meta name="twitter:image" content="https://pub-search.waow.tech/og-image?page=atlas">
18
18
<link rel="icon" href="data:image/svg+xml,<svg xmlns='http://www.w3.org/2000/svg' viewBox='0 0 32 32'><circle cx='16' cy='16' r='2' fill='%231B7340'/><circle cx='8' cy='10' r='1.5' fill='%231B7340' opacity='.6'/><circle cx='24' cy='8' r='1' fill='%231B7340' opacity='.4'/><circle cx='22' cy='22' r='1.5' fill='%231B7340' opacity='.5'/><circle cx='6' cy='24' r='1' fill='%231B7340' opacity='.3'/></svg>">
19
19
<script>
20
20
(function() {
···
23
23
document.documentElement.setAttribute('data-theme', t);
24
24
})();
25
25
</script>
26
-
<link rel="stylesheet" href="constellation.css">
26
+
<link rel="stylesheet" href="atlas.css">
27
27
</head>
28
28
<body>
29
29
<div class="loading-overlay" id="loading">
30
-
<div class="spinner">loading constellation...</div>
30
+
<div class="spinner">loading atlas...</div>
31
31
</div>
32
32
33
33
<canvas id="canvas"></canvas>
34
34
35
35
<div class="header">
36
-
<h1><a href="/">pub search</a> <span class="dim">/ constellation</span></h1>
36
+
<h1><a href="/">pub search</a> <span class="dim">/ atlas</span></h1>
37
37
<div class="nav">
38
+
<form class="search-form" id="search-form">
39
+
<input type="text" id="search-input" placeholder="search..." autocomplete="off" spellcheck="false">
40
+
</form>
38
41
<a href="/dashboard.html">stats</a>
39
42
<span class="theme-toggle" id="theme-toggle"></span>
40
43
</div>
···
58
61
if (theme === 'system') resolved = matchMedia('(prefers-color-scheme: light)').matches ? 'light' : 'dark';
59
62
document.documentElement.setAttribute('data-theme', resolved);
60
63
renderThemeToggle();
61
-
if (window.constellation) window.constellation.setDirty();
64
+
if (window.atlas) window.atlas.setDirty();
62
65
}
63
66
var THEME_ICONS = { dark: '\u263E', light: '\u263C', system: '\u25D1' };
64
67
var THEME_CYCLE = ['dark', 'light', 'system'];
···
76
79
});
77
80
renderThemeToggle();
78
81
</script>
79
-
<script src="constellation.js"></script>
82
+
<script src="atlas.js"></script>
80
83
</body>
81
84
</html>
+208
-3
site/constellation.js
site/atlas.js
+208
-3
site/constellation.js
site/atlas.js
···
39
39
var pointsY = null;
40
40
var platformIdx = null;
41
41
var gridIndex = null;
42
+
var uriToIndex = null; // Map<uri, index> for search matching
43
+
44
+
// --- search state ---
45
+
var searchMatches = null; // Set of point indices matching current search
46
+
var searchCenter = null; // {x, y} weighted centroid of matches
47
+
var searchQuery = '';
48
+
49
+
// --- animation state ---
50
+
var animating = false;
51
+
var animFrom = null;
52
+
var animTo = null;
53
+
var animStart = 0;
54
+
var ANIM_DURATION = 600; // ms
42
55
43
56
// --- canvas ---
44
57
var canvas = document.getElementById('canvas');
···
316
329
}
317
330
}
318
331
332
+
// --- search highlights ---
333
+
if (searchMatches && searchMatches.size > 0) {
334
+
// dim non-matching points by drawing a semi-transparent overlay
335
+
ctx.globalAlpha = dark ? 0.6 : 0.5;
336
+
ctx.fillStyle = dark ? '#050505' : '#f5f5f0';
337
+
ctx.fillRect(0, 0, W, H);
338
+
339
+
// redraw matched points brighter
340
+
ctx.globalAlpha = 1;
341
+
searchMatches.forEach(function(i) {
342
+
var px = pointsX[i], py = pointsY[i];
343
+
if (px < xMin || px > xMax || py < yMin || py > yMax) return;
344
+
var sx = cx + px * scale, sy = cy + py * scale;
345
+
var pi = platformIdx[i];
346
+
if (useGlow) {
347
+
var spr = sprites[pi].hover;
348
+
ctx.drawImage(spr, sx - spr.width / (2 * dpr), sy - spr.height / (2 * dpr), spr.width / dpr, spr.height / dpr);
349
+
} else {
350
+
var dot = dotSprites[pi];
351
+
ctx.drawImage(dot, sx - dot.width / (2 * dpr), sy - dot.height / (2 * dpr), dot.width / dpr, dot.height / dpr);
352
+
}
353
+
});
354
+
355
+
// draw search centroid marker
356
+
if (searchCenter) {
357
+
var mx = cx + searchCenter.x * scale, my = cy + searchCenter.y * scale;
358
+
// crosshair
359
+
ctx.strokeStyle = dark ? 'rgba(255,255,255,0.5)' : 'rgba(0,0,0,0.4)';
360
+
ctx.lineWidth = 1;
361
+
ctx.beginPath();
362
+
ctx.moveTo(mx - 12, my); ctx.lineTo(mx + 12, my);
363
+
ctx.moveTo(mx, my - 12); ctx.lineTo(mx, my + 12);
364
+
ctx.stroke();
365
+
// ring
366
+
ctx.beginPath();
367
+
ctx.arc(mx, my, 6, 0, Math.PI * 2);
368
+
ctx.strokeStyle = dark ? 'rgba(255,255,255,0.7)' : 'rgba(0,0,0,0.5)';
369
+
ctx.lineWidth = 1.5;
370
+
ctx.stroke();
371
+
// label
372
+
ctx.font = '10px monospace';
373
+
ctx.fillStyle = dark ? 'rgba(255,255,255,0.6)' : 'rgba(0,0,0,0.5)';
374
+
ctx.textAlign = 'left';
375
+
ctx.textBaseline = 'top';
376
+
ctx.fillText('"' + searchQuery + '"', mx + 12, my - 6);
377
+
}
378
+
}
379
+
319
380
// --- labels (no shadowBlur — uses strokeText outline instead) ---
320
381
ctx.globalAlpha = 1;
321
382
ctx.fillStyle = dark ? 'rgba(255,255,255,0.75)' : 'rgba(0,0,0,0.65)';
···
365
426
}
366
427
367
428
// --- animation loop ---
429
+
function easeOutCubic(t) { return 1 - Math.pow(1 - t, 3); }
430
+
431
+
function tickAnimation() {
432
+
if (!animating) return;
433
+
var t = Math.min(1, (Date.now() - animStart) / ANIM_DURATION);
434
+
var e = easeOutCubic(t);
435
+
view.zoom = animFrom.zoom + (animTo.zoom - animFrom.zoom) * e;
436
+
view.panX = animFrom.panX + (animTo.panX - animFrom.panX) * e;
437
+
view.panY = animFrom.panY + (animTo.panY - animFrom.panY) * e;
438
+
view.dirty = true;
439
+
if (t >= 1) animating = false;
440
+
}
441
+
442
+
function animateTo(targetX, targetY, targetZoom) {
443
+
animFrom = { zoom: view.zoom, panX: view.panX, panY: view.panY };
444
+
animTo = { zoom: targetZoom, panX: -targetX, panY: -targetY };
445
+
animStart = Date.now();
446
+
animating = true;
447
+
}
448
+
368
449
function loop() {
450
+
tickAnimation();
369
451
render();
370
452
requestAnimationFrame(loop);
371
453
}
···
528
610
}
529
611
530
612
function loadData() {
531
-
fetch('constellation.json')
613
+
fetch('atlas.json')
532
614
.then(function(r) {
533
-
if (!r.ok) throw new Error('failed to load constellation.json: ' + r.status);
615
+
if (!r.ok) throw new Error('failed to load atlas.json: ' + r.status);
534
616
return r.json();
535
617
})
536
618
.then(function(d) {
···
547
629
pointsY[i] = d.points[i].y;
548
630
platformIdx[i] = platMap[d.points[i].platform] !== undefined ? platMap[d.points[i].platform] : otherIdx;
549
631
}
632
+
// build URI → index map for search matching
633
+
uriToIndex = new Map();
634
+
for (var i = 0; i < n; i++) {
635
+
uriToIndex.set(d.points[i].uri, i);
636
+
}
550
637
buildSpatialIndex();
551
638
renderLegend();
552
639
document.getElementById('stats').textContent =
···
567
654
loadData();
568
655
loop();
569
656
570
-
window.constellation = {
657
+
// --- search ---
658
+
var API_URL = 'https://leaflet-search-backend.fly.dev';
659
+
var searchInput = document.getElementById('search-input');
660
+
var searchForm = document.getElementById('search-form');
661
+
var searchStatusEl = null;
662
+
663
+
function setSearchStatus(msg) {
664
+
if (!searchStatusEl) {
665
+
searchStatusEl = document.createElement('span');
666
+
searchStatusEl.className = 'search-status';
667
+
searchForm.appendChild(searchStatusEl);
668
+
}
669
+
searchStatusEl.textContent = msg;
670
+
}
671
+
672
+
function clearSearch() {
673
+
searchMatches = null;
674
+
searchCenter = null;
675
+
searchQuery = '';
676
+
setSearchStatus('');
677
+
view.dirty = true;
678
+
}
679
+
680
+
function doSearch(query) {
681
+
if (!query || !data || !uriToIndex) return;
682
+
searchQuery = query;
683
+
setSearchStatus('searching...');
684
+
685
+
fetch(API_URL + '/search?mode=semantic&limit=20&format=v2&q=' + encodeURIComponent(query))
686
+
.then(function(r) {
687
+
if (!r.ok) throw new Error('search failed: ' + r.status);
688
+
return r.json();
689
+
})
690
+
.then(function(resp) {
691
+
var results = resp.results || [];
692
+
if (results.length === 0) {
693
+
setSearchStatus('no results');
694
+
searchMatches = null;
695
+
searchCenter = null;
696
+
view.dirty = true;
697
+
return;
698
+
}
699
+
700
+
// match result URIs to atlas points
701
+
var matches = new Set();
702
+
var weightedX = 0, weightedY = 0, totalWeight = 0;
703
+
for (var i = 0; i < results.length; i++) {
704
+
var uri = results[i].uri;
705
+
if (uriToIndex.has(uri)) {
706
+
var idx = uriToIndex.get(uri);
707
+
matches.add(idx);
708
+
// weight by rank (higher rank = more weight)
709
+
var w = results.length - i;
710
+
weightedX += pointsX[idx] * w;
711
+
weightedY += pointsY[idx] * w;
712
+
totalWeight += w;
713
+
}
714
+
}
715
+
716
+
if (matches.size === 0) {
717
+
setSearchStatus(results.length + ' results, 0 on map');
718
+
searchMatches = null;
719
+
searchCenter = null;
720
+
view.dirty = true;
721
+
return;
722
+
}
723
+
724
+
searchMatches = matches;
725
+
searchCenter = { x: weightedX / totalWeight, y: weightedY / totalWeight };
726
+
setSearchStatus(matches.size + ' of ' + results.length + ' on map');
727
+
728
+
// compute spread to determine zoom level
729
+
var maxDist = 0;
730
+
matches.forEach(function(idx) {
731
+
var dx = pointsX[idx] - searchCenter.x;
732
+
var dy = pointsY[idx] - searchCenter.y;
733
+
var d = Math.sqrt(dx * dx + dy * dy);
734
+
if (d > maxDist) maxDist = d;
735
+
});
736
+
737
+
// zoom to fit the spread with some padding
738
+
// at zoom=1, visible radius in data coords is ~1.0 (since range is [-1,1])
739
+
// we want maxDist to fit in ~40% of the viewport
740
+
var targetZoom = maxDist > 0 ? Math.min(view.maxZoom, 0.4 / maxDist) : 4;
741
+
targetZoom = Math.max(2, Math.min(8, targetZoom)); // clamp to reasonable range
742
+
743
+
animateTo(searchCenter.x, searchCenter.y, targetZoom);
744
+
})
745
+
.catch(function(err) {
746
+
setSearchStatus('error');
747
+
console.error(err);
748
+
});
749
+
}
750
+
751
+
searchForm.addEventListener('submit', function(e) {
752
+
e.preventDefault();
753
+
var q = searchInput.value.trim();
754
+
if (q) doSearch(q);
755
+
else clearSearch();
756
+
});
757
+
758
+
searchInput.addEventListener('keydown', function(e) {
759
+
if (e.key === 'Escape') {
760
+
searchInput.value = '';
761
+
searchInput.blur();
762
+
clearSearch();
763
+
}
764
+
});
765
+
766
+
// cmd+k / ctrl+k to focus search
767
+
window.addEventListener('keydown', function(e) {
768
+
if ((e.metaKey || e.ctrlKey) && e.key === 'k') {
769
+
e.preventDefault();
770
+
searchInput.focus();
771
+
searchInput.select();
772
+
}
773
+
});
774
+
775
+
window.atlas = {
571
776
setDirty: function() {
572
777
sprites = null;
573
778
dotSprites = null;
+1
-1
site/dashboard.html
+1
-1
site/dashboard.html
···
94
94
</section>
95
95
96
96
<footer>
97
-
<a href="/">back</a> · <a href="/constellation.html">constellation</a> · source on <a href="https://tangled.sh/@zzstoatzz.io/leaflet-search">tangled</a> · <span class="theme-toggle" id="theme-toggle"></span>
97
+
<a href="/">back</a> · <a href="/atlas.html">atlas</a> · source on <a href="https://tangled.sh/@zzstoatzz.io/leaflet-search">tangled</a> · <span class="theme-toggle" id="theme-toggle"></span>
98
98
</footer>
99
99
</div>
100
100
+7
-7
site/functions/og-image.js
+7
-7
site/functions/og-image.js
···
101
101
}
102
102
}
103
103
104
-
// platform colors for constellation OG image (core colors from the canvas)
104
+
// platform colors for atlas OG image (core colors from the canvas)
105
105
const PLATFORM_DOTS = [
106
106
{ color: "#4ade80", label: "leaflet" },
107
107
{ color: "#60a5fa", label: "whitewind" },
···
111
111
{ color: "#9ca3af", label: "other" },
112
112
];
113
113
114
-
// deterministic "random" positions for constellation dots
115
-
function constellationDots() {
114
+
// deterministic "random" positions for atlas dots
115
+
function atlasDots() {
116
116
const dots = [];
117
117
const positions = [
118
118
[180, 200], [340, 150], [520, 280], [700, 180], [850, 250],
···
149
149
const children = [];
150
150
151
151
// scattered dots as background decoration
152
-
children.push(...constellationDots());
152
+
children.push(...atlasDots());
153
153
154
154
// header
155
155
children.push({
···
175
175
fontFamily: '"JetBrains Mono", monospace',
176
176
marginTop: "16px",
177
177
},
178
-
children: "constellation",
178
+
children: "atlas",
179
179
},
180
180
});
181
181
···
284
284
const since = url.searchParams.get("since");
285
285
const mode = url.searchParams.get("mode");
286
286
287
-
// constellation page
288
-
if (page === "constellation") {
287
+
// atlas page
288
+
if (page === "atlas") {
289
289
const stats = await fetchStats();
290
290
const html = buildConstellationImage(stats ? stats.documents : null);
291
291
return new ImageResponse(html, {
+1
-1
site/index.html
+1
-1
site/index.html
···
1594
1594
fetch(`${API_URL}/stats`)
1595
1595
.then(r => r.json())
1596
1596
.then(data => {
1597
-
statsDiv.innerHTML = `${data.documents} documents, ${data.publications} publications · <a href="/dashboard.html">stats</a> · <a href="/constellation.html">constellation</a>`;
1597
+
statsDiv.innerHTML = `${data.documents} documents, ${data.publications} publications · <a href="/dashboard.html">stats</a> · <a href="/atlas.html">atlas</a>`;
1598
1598
const headerStats = document.getElementById('header-stats');
1599
1599
headerStats.href = '/dashboard.html';
1600
1600
headerStats.textContent = `[${data.documents.toLocaleString()} docs]`;