Add evidence.csv: flag records by number of distinct public sources

Most enriched records carry 2-10 verifier-checked sources (median ~4),
but sites_seed.csv keeps only the single strongest evidence_url. The new
normalized table restores the full set — one row per (record, distinct
URL) — reconstructed from the verified enrichment JSON.

- pipeline/build_evidence.py -> data/evidence.csv (291 source rows;
  53/72 sites carry >=2 distinct sources, max 10)
- build_map.py: joins per-record counts, adds a 'Sources (#)' tooltip
  field on pins and arcs (defaults to 1 — every record has >=1 source)
- compute_sites.csv gains an evidence_count column (Renaissance and
  Citadel show as single-source; Verne 8, HRT Lefdal 6)
- 6 new tests: evidence schema, referential integrity, url hygiene,
  per-record url uniqueness, every-site-covered, compute-count sync

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

Author: whiteskylondon

README.md

@@ -5,17 +5,28 @@ documented quant/HFT infrastructure, with an evidence grade on every record.
 
 ## Structure
 
-- `data/sites_seed.csv` — the site inventory (target ~80-120 records)
+- `data/sites_seed.csv` — the site inventory (72 records; target ~80-120)
 - `data/paths.csv` — Tier-2 arcs (origin/dest site_ids, operator, medium),
   rendered as a kepler arc layer
+- `data/compute_sites.csv` — the megawatts sheet: research-tier compute
+  centres only, with GPU counts, power (MW), and build status (generated by
+  `pipeline/build_compute_sheet.py`)
+- `data/evidence.csv` — normalized multi-source table: one row per (record,
+  distinct source URL), so each pin can show how many independent sources
+  back it (generated by `pipeline/build_evidence.py`)
 - `build_map.py` — CSVs → standalone kepler.gl HTML
-  (`python build_map.py --out docs/index.html`)
+  (`python build_map.py --out docs/index.html`); joins evidence.csv to show a
+  source count per pin
 - `pipeline/geocode.py` — Nominatim geocoder (rate-limited, cached) for
   verifying coordinates against documented addresses
-- `docs/index.html` — output (tracked; servable via GitHub Pages); embeddable
-  in Ghost via iframe or direct upload
+- `docs/index.html` — output (tracked; auto-deployed to GitHub Pages via
+  `.github/workflows/pages.yml`); embeddable in Ghost via iframe
 - `ENRICHMENT.md` — candidate records awaiting research/verification
 
+Regenerate derived files after editing `sites_seed.csv`:
+`python pipeline/build_evidence.py && python pipeline/build_compute_sheet.py
+&& python build_map.py --out docs/index.html`
+
 ## Schema
 
 | column | meaning |

build_map.py

@@ -70,6 +70,7 @@
     "status": "Status",
     "coord_precision": "Pin precision",
     "confidence": "Confidence",
+    "evidence_count": "Sources (#)",
     "evidence_note": "Evidence",
     "evidence_url": "Source URL",
     "gmaps_url": "Google Maps",
@@ -80,6 +81,7 @@
     "medium": "Medium",
     "status": "Status",
     "confidence": "Confidence",
+    "evidence_count": "Sources (#)",
     "evidence_note": "Evidence",
     "evidence_url": "Source URL",
 }
@@ -95,14 +97,31 @@
 PATHS_COLOR = [250, 190, 60]      # amber, matching the network layer
 
 
-def load_sites(csv_path: Path) -> pd.DataFrame:
+def evidence_counts(evidence_csv: Path) -> pd.Series:
+    """ref_id -> number of distinct public sources, from evidence.csv.
+
+    Empty Series when the file is absent (evidence_count then defaults to 1
+    since every mapped record carries at least its own evidence_url)."""
+    if not evidence_csv.exists():
+        return pd.Series(dtype="int64")
+    ev = pd.read_csv(evidence_csv, dtype=str)
+    return ev.groupby("ref_id").size()
+
+
+def load_sites(csv_path: Path, counts: pd.Series | None = None) -> pd.DataFrame:
     df = pd.read_csv(csv_path)
 
     required = {"site_id", "site_name", "tier", "lat", "lon", "confidence"}
     missing = required - set(df.columns)
     if missing:
         raise ValueError(f"CSV missing required columns: {missing}")
 
+    # How many distinct public sources back each pin (defaults to 1: every
+    # record carries at least its own evidence_url).
+    counts = pd.Series(dtype="int64") if counts is None else counts
+    df["evidence_count"] = (
+        df["site_id"].map(counts).fillna(1).astype(int).astype(str))
+
     # Google Maps handoff, honest about pin precision: coordinate deep link
     # only where the pin is placed (exact/approximate); a name+city text
     # search for city-level pins (the coords are a centroid, not the site);
@@ -130,7 +149,8 @@ def gmaps(row):
     return df.fillna("")
 
 
-def load_paths(paths_csv: Path, sites: pd.DataFrame) -> pd.DataFrame | None:
+def load_paths(paths_csv: Path, sites: pd.DataFrame,
+               counts: pd.Series | None = None) -> pd.DataFrame | None:
     """Load Tier-2 arcs and join origin/dest coordinates from the sites table.
 
     Returns None when the paths file is absent or has no rows."""
@@ -151,6 +171,10 @@ def load_paths(paths_csv: Path, sites: pd.DataFrame) -> pd.DataFrame | None:
         paths[f"{end}_lon"] = joined["lon"]
         paths[f"{end}_name"] = joined["site_name"]
     paths["route"] = paths["origin_name"] + " → " + paths["dest_name"]
+
+    counts = pd.Series(dtype="int64") if counts is None else counts
+    paths["evidence_count"] = (
+        paths["path_id"].map(counts).fillna(1).astype(int).astype(str))
     return paths
 
 
@@ -233,9 +257,11 @@ def kepler_config(df: pd.DataFrame, with_paths: bool = False) -> dict:
     }
 
 
-def build(csv_path: Path, out_path: Path, paths_csv: Path | None = None) -> None:
-    df = load_sites(csv_path)
-    paths = load_paths(paths_csv, df) if paths_csv else None
+def build(csv_path: Path, out_path: Path, paths_csv: Path | None = None,
+          evidence_csv: Path | None = None) -> None:
+    counts = evidence_counts(evidence_csv) if evidence_csv else pd.Series(dtype="int64")
+    df = load_sites(csv_path, counts=counts)
+    paths = load_paths(paths_csv, df, counts=counts) if paths_csv else None
 
     m = KeplerGl(config=kepler_config(df, with_paths=paths is not None))
     # One dataset per tier so each gets its own layer + legend entry + toggle.
@@ -267,6 +293,7 @@ def build(csv_path: Path, out_path: Path, paths_csv: Path | None = None) -> None
     parser = argparse.ArgumentParser()
     parser.add_argument("--csv", default="data/sites_seed.csv", type=Path)
     parser.add_argument("--paths", default="data/paths.csv", type=Path)
+    parser.add_argument("--evidence", default="data/evidence.csv", type=Path)
     parser.add_argument("--out", default="quant_dc_map.html", type=Path)
     args = parser.parse_args()
-    build(args.csv, args.out, paths_csv=args.paths)
+    build(args.csv, args.out, paths_csv=args.paths, evidence_csv=args.evidence)