Files
threshing-atlas/analysis/stats/site_stats.py
T
d04b88e7b4 Site: show precomputed statistics; drop the upload and the click-to-select route map (#12)
Step 4a of the Python-first core.

- analysis/stats/measures.yaml (declarative): outcomes and denominators, route targets and
  comparison groups, encounter contexts, the route map's stages, six filter presets and the
  route thresholds (5, 10, 20, 30).
- analysis/stats/site_stats.py computes every number the page shows, for every preset, and
  writes data/public/stats.json: factor tables, models, chance events, encounters, the route map
  for every outcome, route rankings with q-values at every threshold, the valley map's place and
  step counts, and The Odds' summary cards.
- The page now only displays: app.js, ui.js and explorer.js read stats.json; engine.js is no
  longer loaded. The three filter controls become one "Runs" menu; minimum runs become fixed
  thresholds. Removed: "Load a newer route export", click-to-select and "Follow" on the Deep
  Dive route map, and the unused old valley renderer. The coded runs are a CSV download.
- tests/parity/compare_stats.py: every number matches what engine.js computed, for all six
  presets (factor tables, models, encounters, route rankings at every threshold, route map).


Claude-Session: https://claude.ai/code/session_01UCzTd31GphpMXDhQqWrCMQ

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-03 21:37:10 +00:00

531 lines
23 KiB
Python
Raw Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
"""Every number the Atlas page shows, precomputed: data/public/stats.json.
The page only displays these; it computes no statistics itself. What is measured (outcomes, route targets,
encounters, the route map's stages, the filter presets and route thresholds) is declared in measures.yaml.
python analysis/stats/site_stats.py [--out data/public/stats.json]
"""
import argparse
import json
import math
import re
import sys
from pathlib import Path
import yaml
HERE = Path(__file__).resolve().parent
ROOT = HERE.parents[1]
sys.path.insert(0, str(ROOT / "core"))
sys.path.insert(0, str(HERE))
from atlas import coding, model # noqa: E402
from stats_core import Z95, bh, logistic_ridge, mantel_haenszel, odds_ratio, p_from_z # noqa: E402
from stats_core import wilson as _wilson # noqa: E402
M = yaml.safe_load(open(HERE / "measures.yaml", encoding="utf-8"))
COLOURS = M["colours"]
is_missing = coding.is_missing
# ---------------------------------------------------------------- conditions from measures.yaml
def match(r, cond):
for k, want in cond.items():
if k == "any":
if not any(match(r, c) for c in want):
return False
continue
v = r.get(k)
if isinstance(want, dict):
if "in" in want and v not in want["in"]:
return False
if "not" in want and v == want["not"]:
return False
if "exists" in want and (v is not None and v != "") != want["exists"]:
return False
elif v != want:
return False
return True
def _fill(cond, colour):
return json.loads(json.dumps(cond).replace("{colour}", colour))
OUTCOMES = {k: v for k, v in M["outcomes"].items()}
for c in COLOURS:
b = M["bonded_outcome"]
OUTCOMES["bonded_" + c] = {"label": b["label"].replace("{colour}", c), "denom": b["denom"], "event": _fill(b["event"], c)}
TARGETS = M["targets"]
STAGES = M["stages"]
def W(k, n):
p, lo, hi = _wilson(k, n)
return {"k": int(k), "n": int(n), "p": p, "lo": lo, "hi": hi}
def rate(runs, pred):
return W(sum(1 for r in runs if pred(r)), len(runs))
def group_by(runs, key):
out = {}
for r in runs:
k = key(r) if callable(key) else r.get(key)
out.setdefault(_key(k), []).append(r)
return out
def _key(v):
"""Object keys as JavaScript writes them (true/false/null), so the page can look them up as before."""
if v is True:
return "true"
if v is False:
return "false"
if v is None:
return "null"
if isinstance(v, float) and v.is_integer():
return str(int(v))
return str(v)
def is_rare(goal):
return goal in M["rare_goals"]
def stratum(r):
if r["source"] != "Form":
return "Bulk logs (imports)"
g = r["goal"]
return {"Blue": "Form: chasing Blue", "Black": "Form: chasing Black", "Blue or Black": "Form: chasing Blue or Black"}.get(g, "Form: other/blank goal")
def goal2(r):
return r["goal"] if is_rare(r["goal"]) else "Blank/not asked" if r["goal"] == "(not asked/blank)" else "Other colour/any"
def apply_filters(runs, f):
def ok(r):
if f.get("source") == "form" and r["source"] != "Form":
return False
if f.get("source") == "bulk" and r["source"] == "Form":
return False
if f.get("goal") == "exclude_rare" and is_rare(r["goal"]):
return False
if f.get("goal") == "only_rare" and not is_rare(r["goal"]):
return False
if f.get("excludeFlagged") and r["data_flags"]:
return False
return True
return [r for r in runs if ok(r)]
# ---------------------------------------------------------------- tables and models
def odds(a, b, c, d):
or_, lo, hi, p, corr = odds_ratio(a, b, c, d)
return {"or": or_, "or_lo": lo, "or_hi": hi, "fisher_p": p, "corrected": corr}
def factor_table(runs, key, min_n=5):
O = OUTCOMES[key]
D = [r for r in runs if match(r, O["denom"])]
ev = lambda r: match(r, O["event"]) # noqa: E731
rows, missing = [], {}
for f in M["factors"]:
known = [r for r in D if not is_missing(r.get(f))]
missing[f] = len(D) - len(known)
groups = group_by(known, f)
nf, ef = len(known), sum(1 for r in known if ev(r))
for lv in sorted(groups):
g = groups[lv]
k, n = sum(1 for r in g if ev(r)), len(g)
row = {"factor": f, "level": lv, **W(k, n)}
rk, rn = ef - k, nf - n
if n >= min_n and rn >= min_n and 0 < ef < nf:
row.update(odds(k, n - k, rk, rn - rk))
rows.append(row)
tested = [r for r in rows if "fisher_p" in r]
for r, q in zip(tested, bh([r["fisher_p"] for r in tested])):
r["q"] = q
return {"outcome": key, "label": O["label"], "overall": W(sum(1 for r in D if ev(r)), len(D)), "rows": rows, "missing": missing}
def fit_model(runs, key, factors, lam=1.0):
import numpy as np
O = OUTCOMES[key]
D = []
for r in runs:
if not match(r, O["denom"]):
continue
o = {"y": 1 if match(r, O["event"]) else 0, **{name: fn(r) for name, fn in factors.items()}}
if all(o[f] is not None for f in factors):
D.append(o)
if len(D) < 20:
return None
cols, names, refs = [np.ones(len(D))], [], {}
for f in factors:
counts = group_by(D, f)
levels = sorted(counts, key=lambda lv: (-len(counts[lv]), lv))
refs[f] = levels[0]
for lv in levels[1:]:
cols.append(np.array([1.0 if _key(o[f]) == lv else 0.0 for o in D]))
names.append((f, lv, len(counts[lv])))
y = np.array([o["y"] for o in D])
b, se = logistic_ridge(np.column_stack(cols), y, lam)
terms = [{"term": f"{f} = {lv}", "factor": f, "level": lv, "n": n, "or": math.exp(b[i + 1]),
"lo": math.exp(b[i + 1] - Z95 * se[i + 1]), "hi": math.exp(b[i + 1] + Z95 * se[i + 1]),
"p": float(p_from_z(b[i + 1] / se[i + 1]))} for i, (f, lv, n) in enumerate(names)]
return {"outcome": key, "label": O["label"], "n": len(D), "events": int(y.sum()), "refs": refs, "lambda": lam, "terms": terms}
def route_key(r):
return [r["company"], r["direction"], r["first_sighting"], r["scream"], r["resolution"], r.get("post_path") or "—", r.get("hour") or "—"]
def analyse(all_runs, filters):
runs = apply_filters(all_runs, filters)
res = {"filters": filters, "n": len(runs)}
flag_counts = {}
for r in runs:
for f in (r["data_flags"].split("; ") if r["data_flags"] else []):
k = "first sighting outside documented pool" if f.startswith("first sighting") else f
flag_counts[k] = flag_counts.get(k, 0) + 1
dates = sorted(r["date"] for r in runs if r.get("date"))
outs = group_by(runs, lambda r: r["outcome"] + (" " + r["outcome_color"] if r.get("outcome_color") else ""))
res["meta"] = {
"sources": {k: len(v) for k, v in group_by(runs, "source").items()},
"date_min": dates[0] if dates else None, "date_max": dates[-1] if dates else None,
"n_flagged": sum(1 for r in runs if r["data_flags"]), "flag_counts": flag_counts,
"outcomes": {k: len(v) for k, v in outs.items()},
}
res["chance"] = {}
for key in ("blue_meadow", "shadow"):
O = OUTCOMES[key]
D = [r for r in runs if match(r, O["denom"])]
ev = lambda r, O=O: match(r, O["event"]) # noqa: E731
by = lambda k: {s: rate(g, ev) for s, g in sorted(group_by(D, k).items())} # noqa: E731
form = [r for r in D if r["source"] == "Form"]
res["chance"][key] = {"label": O["label"], "overall": rate(D, ev), "by_stratum": by(stratum), "by_goal": by("goal"),
"by_attempt": by("attempt_group"), "form_overall": rate(form, ev)}
fs = group_by([r for r in runs if not is_missing(r["direction"]) and not is_missing(r["company"])], lambda r: r["company"] + "|" + r["direction"])
res["first_sighting"] = [{"key": k, "n": len(g), "colors": {c: len(v) for c, v in sorted(group_by(g, "first_sighting").items())}}
for k, g in sorted(fs.items())]
res["encounters"] = []
for e in M["encounters"]:
col, colour = e["column"], e["colour"]
A = [r for r in runs if match(r, e["where"]) and r.get(col) in ("Bonded!", "Incinerated")]
win = lambda r, col=col: r.get(col) == "Bonded!" # noqa: E731
intent = [r for r in A if r["goal"] == colour or r["goal"].startswith("Any") or (r["goal"] == "Blue or Black" and colour in ("Blue", "Black"))]
res["encounters"].append({"context": e["context"], "color": colour, "overall": rate(A, win), "intent_matched": rate(intent, win),
"by_stratum": {s: rate(g, win) for s, g in sorted(group_by(A, stratum).items())}})
greens_where = next(e["where"] for e in M["encounters"] if e["colour"] == "Green")
greens = [r for r in runs if match(r, greens_where) and r.get("enc_Green")]
g = {"green_encounters": len(greens),
"bonded_green": sum(1 for r in greens if r["enc_Green"] == "Bonded!"),
"rescued_orange": sum(1 for r in greens if r["rescued"] and r.get("outcome_color") == "Orange"),
"rescued_brown": sum(1 for r in greens if r["rescued"] and r.get("outcome_color") == "Brown"),
"incinerated": sum(1 for r in greens if r["outcome"] == "Incinerated")}
g["no_outcome"] = g["green_encounters"] - g["bonded_green"] - g["rescued_orange"] - g["rescued_brown"] - g["incinerated"]
g["rates"] = {k: W(g[k], g["green_encounters"]) for k in ("bonded_green", "rescued_orange", "rescued_brown", "incinerated", "no_outcome")}
res["green_rescue"] = g
res["factors"] = {k: factor_table(runs, k) for k in OUTCOMES}
first2 = lambda r: r["first_sighting"] if r["first_sighting"] in ("Red", "Brown", "Orange", "Green") else None # noqa: E731
known = lambda f: (lambda r: None if is_missing(r.get(f)) else r[f]) # noqa: E731
src2 = lambda r: "Form submission" if r["source"] == "Form" else "Bulk log (import)" # noqa: E731
res["models"] = {
"blue_meadow": fit_model(runs, "blue_meadow", {"source": src2, "goal": goal2, "direction": known("direction"),
"first_sighting": first2, "resolution": known("resolution"), "company": known("company")}),
"shadow": fit_model(runs, "shadow", {"source": src2, "goal": goal2, "attempt": known("attempt_group"), "direction": known("direction")}),
"bond_blue_meadow": fit_model(runs, "bond_blue_meadow", {"source": src2, "goal": goal2, "direction": known("direction"), "resolution": known("resolution")}),
}
res["daily"] = []
for d, gr in sorted(group_by([r for r in runs if r.get("date")], "date").items()):
m = [r for r in gr if match(r, OUTCOMES["blue_meadow"]["denom"])]
t = [r for r in gr if match(r, OUTCOMES["shadow"]["denom"])]
bk, sk = sum(1 for r in m if r["saw_blue_meadow"]), sum(1 for r in t if r["saw_shadow"])
res["daily"].append({"date": d, "runs": len(gr), "meadow_n": len(m), "blue_k": bk, "tree_n": len(t), "shadow_k": sk,
"blue": W(bk, len(m)), "shadow": W(sk, len(t))})
res["source_rates"] = {key: {s: rate([r for r in g if match(r, OUTCOMES[key]["denom"])], lambda r, k=key: match(r, OUTCOMES[k]["event"]))
for s, g in group_by(runs, "source").items()} for key in ("blue_meadow", "shadow")}
return res
# ---------------------------------------------------------------- route map and route rankings
def stage_value(r, key):
if key in ("company", "direction", "post_path"):
return None if is_missing(r.get(key)) else r[key]
if key == "first_sighting":
return None if r["scream"] == "DIED at first sighting" or is_missing(r["first_sighting"]) else r["first_sighting"]
if key == "cadet":
if r["scream"].startswith("DIED") or is_missing(r["scream"]) or is_missing(r["resolution"]):
return None
return "Help → " + r["resolution"] if r["scream"] == "Investigate & help" else r["scream"] + " → " + r["resolution"]
if key == "hour":
return "Climb the tree" if r.get("hour") == "Tree" else "Valley centre" if r.get("hour") == "Valley" else None
if key == "encounter":
if not r.get("branch"):
return None
if r["hour"] == "Tree":
if r["branch"] == "Shadow (rare)":
return "Tree: shadow → Blue/Black"
if r["branch"] == "Descend quickly: Red":
return "Tree: Red below"
return "Tree: two Greens"
if r["branch"] == "Valley Red":
return "Valley: bond the Red"
for cond, label in ((r.get("enc_Blue"), "Meadow: Blue"), (r.get("enc_Orange") and r.get("enc_Brown"), "Meadow: Orange/Brown"),
(r.get("enc_Orange"), "Meadow: Orange"), (r.get("enc_Brown"), "Meadow: Brown")):
if cond:
return label
return "Meadow (no attempt recorded)"
if key == "end":
if r["outcome"] == "Bonded":
return "Bonded " + r["outcome_color"]
if r["outcome"] in ("Incinerated", "Died before the hour"):
return r["outcome"]
return None
def route_map(runs):
"""The route map: nodes and links with run counts, and for every target the hits and their Wilson interval."""
nodes, links = {}, {}
hits = {t: {} for t in TARGETS}
for r in runs:
prev = None
hit = {t: match(r, T["event"]) for t, T in TARGETS.items()}
for si, s in enumerate(STAGES):
v = stage_value(r, s["key"])
if v is None:
continue
nid = f"{si}|{v}"
nd = nodes.setdefault(nid, {"id": nid, "stage": si, "stageKey": s["key"], "value": v, "n": 0})
nd["n"] += 1
for t in TARGETS:
hits[t][nid] = hits[t].get(nid, 0) + hit[t]
if prev:
lid = prev + "→" + nid
ln = links.setdefault(lid, {"id": lid, "source": prev, "target": nid, "n": 0, "chance": bool(s.get("chance"))})
ln["n"] += 1
for t in TARGETS:
hits[t][lid] = hits[t].get(lid, 0) + hit[t]
prev = nid
nl, ll = list(nodes.values()), list(links.values())
by_target = {t: {"nodes": [W(hits[t][n["id"]], n["n"]) for n in nl], "links": [W(hits[t][l["id"]], l["n"]) for l in ll]} for t in TARGETS}
return {"nodes": nl, "links": ll, "by_target": by_target}
def rank_routes(runs, target, thresholds):
"""Full routes (every choice up to the hour), ranked for one target. Each route with at least min_n runs is
compared with all other runs that could reach the target (Fisher), and within data-source / chase-goal strata
(Mantel-Haenszel). q-values are corrected within each threshold. Ranked by the lower end of the 95% interval."""
T = TARGETS[target]
ev = lambda r: match(r, T["event"]) # noqa: E731
key_stages = [s["key"] for s in STAGES[:6]]
pool = [r for r in runs if match(r, T["pool"])]
tot_k, tot_n = sum(1 for r in pool if ev(r)), len(pool)
full = [r for r in pool if all(stage_value(r, k) or (k == "post_path" and r.get("post_path") is None) for k in key_stages)]
groups = group_by(full, lambda r: " › ".join(stage_value(r, k) or "—" for k in key_stages))
strata = list(group_by(pool, stratum).values())
rows = []
for key, g in groups.items():
if len(g) < min(thresholds):
continue
k, n = sum(1 for r in g if ev(r)), len(g)
o = {"route": key, "values": [stage_value(g[0], s) for s in key_stages], **W(k, n)}
o.update(odds(k, n - k, tot_k - k, tot_n - n - (tot_k - k)))
ids = {id(r) for r in g}
tabs = []
for sg in strata:
a = sum(1 for r in sg if id(r) in ids and ev(r))
nin = sum(1 for r in sg if id(r) in ids)
c = sum(1 for r in sg if id(r) not in ids and ev(r))
tabs.append([a, nin - a, c, len(sg) - nin - c])
mh = mantel_haenszel(tabs)
o.update(dict(zip(("mh_or", "mh_lo", "mh_hi", "mh_p"), mh)) if mh else {"mh_or": None, "mh_lo": None, "mh_hi": None, "mh_p": None})
rows.append(o)
rows.sort(key=lambda r: (-r["lo"], -r["n"]))
out = {"target": target, "overall": W(tot_k, tot_n), "rows": rows, "by_threshold": {}}
for m in thresholds:
idx = [i for i, r in enumerate(rows) if r["n"] >= m]
q = bh([rows[i]["fisher_p"] for i in idx])
mi = [i for i in idx if rows[i]["mh_p"] is not None]
mq = bh([rows[i]["mh_p"] for i in mi])
out["by_threshold"][str(m)] = {"n_routes_tested": len(idx), "q": {str(i): v for i, v in zip(idx, q)},
"mh_q": {str(i): v for i, v in zip(mi, mq)}}
return out
# ---------------------------------------------------------------- the valley map and the summary cards
def place_ids():
"""The places drawn on the valley map (the keys of L in site/atlas/valley.js)."""
src = (ROOT / "site/atlas/valley.js").read_text(encoding="utf-8")
body = src[src.index("const L = {"):]
body = body[:body.index("\n };")]
return set(re.findall(r'^\s{4}"?([^":{}\n]+(?::[^":{}\n]+)*)"?:\s*\{\s*x:', body, re.M))
PLACES = None
def place_path(r):
"""The places a run passed through, as drawn on the valley map (mirrors path() in valley.js)."""
L = PLACES
s = []
if "comp:" + str(r["company"]) in L:
s.append("comp:" + r["company"])
if r["direction"] not in ("East", "West", "Center"):
return s
s.append("dir:" + r["direction"])
sight = f"see:{r['direction']}:{r['first_sighting']}"
if sight in L:
s.append(sight)
if r["scream"] in ("DIED at first sighting", "Unknown"):
return s
s.append("scream")
if r["scream"] == "Investigate & help":
s.append("bully")
if "res:" + r["resolution"] in L:
s.append("res:" + r["resolution"])
if r["resolution"].startswith("DIED"):
return s
if r.get("post_path") in ("Forest", "West", "East"):
s.append("fork:" + r["post_path"])
elif r["scream"] == "Ignore the scream":
s.append("wounded")
if "res:" + r["resolution"] in L:
s.append("res:" + r["resolution"])
elif r["scream"] == "Investigate & leave":
s.append("away")
if r.get("hour") not in ("Tree", "Valley"):
return s
s.append("hour")
b = r.get("branch")
if r["hour"] == "Tree":
s.append("tree")
if b == "Descend quickly: Red":
s.append("treeRed")
elif b == "Descend quickly: let the Red pass":
s += ["treeRed", "greens"]
elif b == "Descend slowly: two Greens":
s.append("greens")
elif b == "Shadow (rare)":
s.append("stream")
if r.get("enc_Blue") or r.get("enc_Black"):
s.append("redwood")
if r.get("enc_Black"):
s.append("black")
else:
s.append("valley")
if b == "Valley Red":
s.append("valleyRed")
elif b == "Passed the Red: meadow":
s.append("meadow")
oc = r.get("outcome_color")
tried = "Blue" if r.get("enc_Blue") else oc if oc in ("Orange", "Brown") else "Orange" if r.get("enc_Orange") else "Brown" if r.get("enc_Brown") else None
if tried:
s.append("meadow:" + tried)
return s
def valley(runs):
counts, edges = {}, {}
for r in runs:
seq = place_path(r)
for i, p in enumerate(seq):
counts[p] = counts.get(p, 0) + 1
if i:
k = seq[i - 1] + ">" + p
edges[k] = edges.get(k, 0) + 1
return {"counts": counts, "edges": [[*k.split(">"), n] for k, n in sorted(edges.items(), key=lambda kv: -kv[1])]}
def overview(runs):
"""The Odds' summary cards: cadets who went in and either bonded or died. Deliberate deaths (burned at an
encounter the rider did not genuinely attempt) and runs with no recorded outcome are left out."""
ended = [r for r in runs if r["outcome"] in ("Bonded", "Incinerated", "Died before the hour")]
def deliberate(r):
if r["outcome"] != "Incinerated":
return False
v = model.prepare_run(r)
return bool(v.get("encCtx")) and not model.genuine_attempt(r, v["encCtx"])
cadets = [r for r in ended if not deliberate(r)]
bonded = sum(1 for r in cadets if r["outcome"] == "Bonded")
return {"cadets": len(cadets), "bonded": bonded, "dead": len(cadets) - bonded, "deliberate": len(ended) - len(cadets),
"early_deaths": sum(1 for r in cadets if r["outcome"] == "Died before the hour")}
# ---------------------------------------------------------------- everything
def clean(o):
if isinstance(o, dict):
return {str(k): clean(v) for k, v in o.items()}
if isinstance(o, (list, tuple)):
return [clean(v) for v in o]
if hasattr(o, "item"):
o = o.item()
if isinstance(o, float) and (math.isinf(o) or math.isnan(o)):
return None
return o
def build(runs, rows, exported=None):
global PLACES
PLACES = place_ids()
th = M["route_thresholds"]
out = {
"schema": 1,
"data": {"form_rows": rows, "runs": len(runs), "exported": exported},
"meta": {
"presets": {k: v["label"] for k, v in M["presets"].items()}, "route_thresholds": th,
"targets": {k: {"label": t["label"], "color": t["colour"]} for k, t in TARGETS.items()},
"outcomes": {k: {"label": o["label"]} for k, o in OUTCOMES.items()},
"factors": M["factors"], "stages": [{"key": s["key"], "label": s["label"], "chance": bool(s.get("chance"))} for s in STAGES],
"missing_values": sorted(coding.MISSING),
},
"presets": {},
}
for pid, p in M["presets"].items():
sub = apply_filters(runs, p["filters"])
out["presets"][pid] = {
"label": p["label"], "analyse": analyse(runs, p["filters"]),
"routes": {t: rank_routes(sub, t, th) for t in TARGETS},
"map": route_map(sub), "valley": valley(sub), "overview": overview(sub),
}
return clean(out)
def coded_runs(path):
df, _ = coding.load(path)
df = df.drop(columns=["run_id"])
return df.astype(object).where(df.notna(), None).to_dict("records")
def main(argv=None):
ap = argparse.ArgumentParser(description=__doc__)
ap.add_argument("--data", default=str(ROOT / "data/public/form_responses.json"))
ap.add_argument("--out", default=str(ROOT / "data/public/stats.json"))
a = ap.parse_args(argv)
raw = json.load(open(a.data, encoding="utf-8"))
runs = coded_runs(a.data)
out = build(runs, len(raw["rows"]), raw.get("exported"))
Path(a.out).write_text(json.dumps(out, separators=(",", ":"), ensure_ascii=False), encoding="utf-8")
print(f"{a.out}: {len(runs)} runs, {len(out['presets'])} presets, {Path(a.out).stat().st_size / 1e6:.1f} MB")
if __name__ == "__main__":
main()