"""Compute every figure used in the report. Writes numbers.json.""" import json, os, collections, statistics from urllib.parse import urlparse HERE = os.path.dirname(os.path.abspath(__file__)) J = lambda n: json.load(open(os.path.join(HERE, n))) def reg(host): p = host.split(".") return ".".join(p[-2:]) if len(p) > 2 else host recs = [json.loads(l) for l in open(os.path.join(HERE, "main_raw.jsonl")) if l.strip()] cited = J("cited_domains.json") vendors = J("vendor_domains_merged.json") CD = {c["domain"]: c for c in cited} N = {} MODELS = sorted({r["model"] for r in recs}) CATS = sorted({r["cat"] for r in recs}) N["run_date"] = "2026-09-02" N["models"] = MODELS N["n_categories"] = len(CATS) N["n_calls"] = len(recs) N["n_errors"] = sum(1 for r in recs if r.get("error")) N["n_answers_parsed"] = sum(1 for r in recs if r.get("picks")) N["tranco_list_id"] = "K9QPW" N["tranco_date"] = "2026-09-01" # ---- citations ---- allc = collections.Counter() per_model = {} for m in MODELS: c = collections.Counter() for r in recs: if r["model"] != m: continue for u in r.get("citations", []): h = urlparse(u).netloc.lower().replace("www.", "") if h: c[reg(h)] += 1 allc += c tot = sum(c.values()) unr = sum(n for d, n in c.items() if CD.get(d, {}).get("tranco_rank") is None) over = sum(n for d, n in c.items() if (CD.get(d, {}).get("tranco_rank") or 10 ** 9) > 100000) ranks = [] for d, n in c.items(): rk = CD.get(d, {}).get("tranco_rank") if rk: ranks += [rk] * n per_model[m] = { "citations": tot, "unique_domains": len(c), "pct_unranked": round(100 * unr / tot, 1), "pct_worse_than_100k": round(100 * over / tot, 1), "median_rank_of_ranked": int(statistics.median(ranks)), "top10": c.most_common(10), } N["per_model"] = per_model tot = sum(allc.values()) N["total_citations"] = tot N["unique_cited_domains"] = len(allc) unr = sum(n for d, n in allc.items() if CD.get(d, {}).get("tranco_rank") is None) over = sum(n for d, n in allc.items() if (CD.get(d, {}).get("tranco_rank") or 10 ** 9) > 100000) N["pct_citations_unranked"] = round(100 * unr / tot, 1) N["pct_citations_worse_than_100k"] = round(100 * over / tot, 1) N["n_unranked_domains"] = sum(1 for d in allc if CD.get(d, {}).get("tranco_rank") is None) N["pct_domains_unranked"] = round(100 * N["n_unranked_domains"] / len(allc), 1) N["top25_domains"] = [[d, n, round(100 * n / tot, 2), CD.get(d, {}).get("tranco_rank")] for d, n in allc.most_common(25)] N["top10_share_pct"] = round(100 * sum(n for _, n in allc.most_common(10)) / tot, 1) _r = [] for d, n in allc.items(): rk = CD.get(d, {}).get("tranco_rank") if rk: _r += [rk] * n N["pooled_median_rank_of_ranked_citations"] = int(statistics.median(_r)) N["n_citations_to_ranked_domains"] = len(_r) # ---- unranked leaders ---- unranked = [(d, n) for d, n in allc.most_common() if CD.get(d, {}).get("tranco_rank") is None] N["top_unranked"] = [{"domain": d, "citations": n, "categories": CD[d]["categories"], "wayback_first": CD[d]["wayback_first"], "wayback_pages": CD[d]["wayback_pages"]} for d, n in unranked[:30]] # ---- wayback ---- def year(ts): return int(str(ts)[:4]) if ts else None unr_doms = [d for d, n in unranked] rk_doms = [d for d, n in allc.most_common() if CD.get(d, {}).get("tranco_rank")] u_yrs = [year(CD[d]["wayback_first"]) for d in unr_doms if CD[d]["wayback_first"]] r_yrs = [year(CD[d]["wayback_first"]) for d in rk_doms if CD[d]["wayback_first"]] N["wayback"] = { "unranked_domains": len(unr_doms), "unranked_never_archived": sum(1 for d in unr_doms if not CD[d]["wayback_first"]), "unranked_never_archived_pct": round( 100 * sum(1 for d in unr_doms if not CD[d]["wayback_first"]) / len(unr_doms), 1), "ranked_domains": len(rk_doms), "ranked_never_archived": sum(1 for d in rk_doms if not CD[d]["wayback_first"]), "ranked_never_archived_pct": round( 100 * sum(1 for d in rk_doms if not CD[d]["wayback_first"]) / len(rk_doms), 1), "median_first_year_unranked": statistics.median(u_yrs) if u_yrs else None, "median_first_year_ranked": statistics.median(r_yrs) if r_yrs else None, "unranked_first_seen_2025_or_later_pct": round( 100 * sum(1 for y in u_yrs if y >= 2025) / len(u_yrs), 1) if u_yrs else None, "ranked_first_seen_2025_or_later_pct": round( 100 * sum(1 for y in r_yrs if y >= 2025) / len(r_yrs), 1) if r_yrs else None, } # ---- citation-weighted: share of citations to domains never archived ---- never = sum(n for d, n in allc.items() if not CD.get(d, {}).get("wayback_first")) N["pct_citations_never_archived"] = round(100 * never / tot, 1) # ---- model agreement on #1 ---- top1 = collections.defaultdict(dict) for r in recs: ps = r.get("picks") or [] if ps: top1[r["cat"]][r["model"]] = ps[0]["name"].lower().strip() both = [c for c, d in top1.items() if len(d) == len(MODELS)] agree = sum(1 for c in both if len(set(top1[c].values())) == 1) N["agreement"] = {"categories_with_both": len(both), "same_top1": agree, "same_top1_pct": round(100 * agree / len(both), 1)} # top-5 set Jaccard across models jac = [] sets = collections.defaultdict(dict) for r in recs: if r.get("picks"): sets[r["cat"]][r["model"]] = {p["name"].lower().strip() for p in r["picks"]} for c, d in sets.items(): if len(d) == 2: a, b = list(d.values()) jac.append(len(a & b) / len(a | b)) N["agreement"]["mean_top5_jaccard"] = round(sum(jac) / len(jac), 3) N["agreement"]["n_jaccard_pairs"] = len(jac) # ---- vendor domains (zombie section) ---- V = {v["domain"]: v for v in vendors} N["vendors"] = {"unique": len(vendors)} no_dns = [v["domain"] for v in vendors if not v.get("ip")] unreach = [v["domain"] for v in vendors if v.get("ip") and v.get("status") is None] err4xx = [(v["domain"], v["status"]) for v in vendors if isinstance(v.get("status"), int) and v["status"] >= 400] def host_of(url): h = urlparse(url).netloc.lower().split("@")[-1].split(":")[0] return h[4:] if h.startswith("www.") else h def offdomain(v): """True only if the final URL sits on a different registrable domain.""" if not v.get("final"): return False return reg(host_of(v["final"])) != reg(v["domain"]) redir = [(v["domain"], v["final"]) for v in vendors if offdomain(v)] # 403/429/417/409 are bot-blocking or rate limits, not dead sites: excluded # from the wrong-or-stale count. Only 404/410 and 5xx count as broken. broken_http = [(d, s) for d, s in err4xx if s in (404, 410) or s >= 500] N["vendors"].update({ "no_dns": sorted(no_dns), "n_no_dns": len(no_dns), "dns_but_unreachable": sorted(unreach), "n_unreachable": len(unreach), "http_status_counts": sorted(collections.Counter(s for _, s in err4xx).items()), "http_broken": sorted(broken_http), "n_http_broken": len(broken_http), "n_bot_blocked_403": sum(1 for _, s in err4xx if s == 403), "redirect_off_domain": sorted(redir), "n_redirect_off_domain": len(redir), }) bad = (set(no_dns) | set(unreach) | {d for d, _ in broken_http} | {d for d, _ in redir}) N["vendors"]["n_wrong_or_stale"] = len(bad) N["vendors"]["pct_wrong_or_stale"] = round(100 * len(bad) / len(vendors), 1) N["vendors"]["n_gone_or_unreachable"] = len( set(no_dns) | set(unreach) | {d for d, _ in broken_http}) N["vendors"]["pct_gone_or_unreachable"] = round( 100 * N["vendors"]["n_gone_or_unreachable"] / len(vendors), 1) # ---- brands ---- bc = collections.Counter() for r in recs: for p in r.get("picks", []): bc[p["name"].lower().strip()] += 1 N["n_distinct_brands"] = len(bc) N["n_brand_slots"] = sum(bc.values()) N["top_brands"] = bc.most_common(15) # ---- shared retrieval between the two tiers ---- by_cat = collections.defaultdict(dict) url_sets = collections.defaultdict(set) for r in recs: by_cat[r["cat"]][r["model"]] = tuple(r.get("citations", [])) url_sets[r["model"]] |= set(r.get("citations", [])) ident = sum(1 for c, d in by_cat.items() if len(d) == len(MODELS) and len(set(d.values())) == 1) a, b = MODELS N["shared_retrieval"] = { "categories_identical_citation_list": ident, "categories_identical_pct": round(100 * ident / len(by_cat), 1), "unique_urls_per_model": {m: len(url_sets[m]) for m in MODELS}, "url_overlap": len(url_sets[a] & url_sets[b]), "url_jaccard": round(len(url_sets[a] & url_sets[b]) / len(url_sets[a] | url_sets[b]), 3), "citations_per_answer_mean": round(tot / len(recs), 2), } # ---- the named actors ---- def dom_stats(d): c = collections.Counter() cats = set() urls = set() for r in recs: for u in r.get("citations", []): h = urlparse(u).netloc.lower().replace("www.", "") if reg(h) == d: c[u] += 1 cats.add(r["cat"]) urls.add(u) n = sum(c.values()) return {"domain": d, "citations": n, "pct_of_all": round(100 * n / tot, 2), "categories": len(cats), "unique_urls": len(urls), "tranco_rank": CD.get(d, {}).get("tranco_rank"), "rank_among_domains": [x for x, _ in allc.most_common()].index(d) + 1 if d in allc else None} FARM = ["worldmetrics.org", "gitnux.org", "wifitalents.com"] N["named"] = {d: dom_stats(d) for d in ["guideflow.com"] + FARM + ["gartner.com", "g2.com", "reddit.com", "capterra.com", "wikipedia.org"]} fn = sum(N["named"][d]["citations"] for d in FARM) fc = set() for r in recs: for u in r.get("citations", []): if reg(urlparse(u).netloc.lower().replace("www.", "")) in FARM: fc.add(r["cat"]) N["farm_network"] = {"domains": FARM, "citations": fn, "pct_of_all": round(100 * fn / tot, 2), "categories_touched": len(fc), "categories_touched_pct": round(100 * len(fc) / len(CATS), 1)} json.dump(N, open(os.path.join(HERE, "numbers.json"), "w"), indent=1) print(json.dumps({k: v for k, v in N.items() if k not in ("top25_domains", "top_unranked", "per_model", "top_brands", "vendors")}, indent=1)) print("\nper model:") for m, d in per_model.items(): print(" ", m, {k: v for k, v in d.items() if k != "top10"}) print("\ntop unranked:") for u in N["top_unranked"][:20]: print(" ", u) print("\nvendors:", {k: v for k, v in N["vendors"].items() if not isinstance(v, list)})