feat(collect): 纯离线生成打通 1ms 签到文字点选 —— errorCode=0 收口

做了什么: - 新增 tcaptcha/tdc_values.py: 37 个语义名的取值生成器（静态画像 / 会话量 / 每次采集量三层）+ build_plain 四段组装 + generate_collect 纯离线出包 - crypto: 新增 generate_telemetry(mode="python_values")；轨迹生成统一到 trajectory_points()（首点绝对坐标 + 增量 + 固定收尾 [1,1,12]） - crypto.split_collect_segments 重写: pad 只有 1 个空格时旧 ` {2,}` 判据会漏段，且 UA 串里的普通空格会假阳性 —— 改为「模 24 算术 + 不落在 JSON 字符串内部 + 结构校验」 - tools/tdc_modules: trajectory 指纹原先指向只做 JSON.stringify 的模块，改为 mouseMoveEvent/touchStartEvent（真轨迹模块），37/37 不变 - client 默认 telemetry_mode 改为 python_values - 新增 scripts/verify/{align_session,audit_names,verify_values,verify_offline,send_test}.py 与 tools/{cdp_collect,cdp_dump_scripts,tdc_profile}.py 为什么: - 账本 #40「谁在发 HTTP 是 12 的剩余自变量」被推翻。同 aid/同客户端/同求解器、答案都对，只换 collect 构造方式: 本次会话构建生成 -> 0 (4/4)，旧模板回填 -> 12 (2/2)。真因是 collect 必须与本次会话构建同源。怎么验证: - .venv/bin/python -m tcaptcha.selftest 全绿（含新增四段切分与纯离线闭环回归） - scripts/verify/verify_values.py 4 份真机明文 152/152 逐槽全等 - scripts/verify/verify_offline.py 逐槽全等 + 分段几何一致 - scripts/verify/send_test.py --mode python_values 1ms 签到页文字点选 errorCode=0 x4
2026-10-12 03:38:18 +08:00
parent 80a122245a
commit 51dff870cf
18 changed files with 2102 additions and 91 deletions
@@ -0,0 +1,138 @@
+# -*- coding: utf-8 -*-
+"""把真机明文的值，与「语义名」精确配对。
+
+难点：真机明文的槽顺序 = **它自己那次构建**的 require 顺序，
+而我手上只有 har/online/ 等若干构建；不知道明文属于哪一份。
+
+解法（自洽性打分）：把明文的值**按每个候选构建的语义名顺序**标注一遍，
+再看标注结果是否语义自洽（platform 的位置该是 "MacIntel"、user_agent 该含
+"Mozilla"、charset 该是 "UTF-8" …）。自洽度最高者即为它所属的构建；
+若都不匹配，也能从"哪些语义名被正确喂到"反推出每个名字的真实取值。
+
+用法:
+    .venv/bin/python scripts/verify/label_plaintext.py <明文文件> <若干 tdc.js>
+"""
+import glob
+import json
+import re
+import sys
+from pathlib import Path
+
+ROOT = Path(__file__).resolve().parents[2]
+sys.path.insert(0, str(ROOT / "tools"))
+import tdc_modules as tm  # noqa: E402
+
+# ---- 语义名 -> 期望值特征（这就是"画像"的定义雏形）----
+def likes(name: str, v) -> str:
+    """值是否与语义名相容。返回 OK / NO / ??（?? = 该名字无法从值判断）"""
+    s = json.dumps(v, ensure_ascii=False)
+    if name.startswith("user_agent"):
+        return "OK" if isinstance(v, str) and "Mozilla" in v else "NO"
+    if name == "platform":
+        return "OK" if isinstance(v, str) and v and " " not in v else "NO"
+    if name == "languages":
+        return "OK" if isinstance(v, list) and v and isinstance(v[0], str) else "NO"
+    if name == "hardware_concurrency":
+        return "OK" if isinstance(v, int) and 2 <= v <= 64 else "NO"
+    if name == "screen_width":
+        return "OK" if isinstance(v, int) and v > 500 else "NO"
+    if name == "screen_height":
+        return "OK" if isinstance(v, int) and v > 300 else "NO"
+    if name == "screen_color_depth":
+        return "OK" if isinstance(v, int) and 16 <= v <= 48 else "NO"
+    if name == "screen_fingerprint":
+        return "OK" if isinstance(v, str) and "-" in v and "*" in v else "NO"
+    if name == "timezone_offset":
+        return "OK" if isinstance(v, str) and re.fullmatch(r"[+-]\d\d", v) else "NO"
+    if name == "charset":
+        return "OK" if isinstance(v, str) and "UTF" in v.upper() else "NO"
+    if name == "webgl_renderer":
+        return "OK" if isinstance(v, str) and ("ANGLE" in v or "Renderer" in v) else "NO"
+    if name == "webgl_vendor":
+        return "OK" if isinstance(v, str) and ("Inc." in v or "Google" in v) else "NO"
+    if name == "canvas_fingerprint":
+        return "OK" if isinstance(v, str) and len(v) > 40 else "NO"
+    if name == "constant_98k":
+        return "OK" if v == "98k" else "NO"
+    if name == "template_url":
+        return "OK" if isinstance(v, str) and "captcha" in v and "rand=" in v else "NO"
+    if name in ("location_href", "referrer"):
+        return "OK" if isinstance(v, str) and v.startswith("http") else "??"
+    if name == "trajectory":
+        return "OK" if (isinstance(v, list) and v
+                        and all(isinstance(x, list) and len(x) == 3 for x in v)) else "NO"
+    return "??"
+
+
+def main() -> int:
+    if len(sys.argv) < 3:
+        print(__doc__)
+        return 2
+    plain = open(sys.argv[1], encoding="latin1").read().rstrip("\x00")
+    logical = re.sub(r" {2,}", "", plain)
+    cd = json.loads(logical)["cd"]
+    print("明文 cd 槽数: %d" % len(cd))
+    print()
+
+    cands = []
+    for a in sys.argv[2:]:
+        cands += sorted(glob.glob(a))
+    if not cands:
+        print("没有候选构建")
+        return 1
+
+    def assign(names, nslots):
+        """按语义名顺序把 nslots 个槽分配给各名字（交错推进）。"""
+        out = {}
+        k = 0
+        for nm in names:
+            if k >= nslots:
+                break
+            out.setdefault(nm, []).append(cd[k])
+            k += 1
+        return out
+
+    best = None
+    for f in cands:
+        rows = tm.analyze(f)
+        names = [r["name"] for r in rows]
+        # 假设除 itoken（占 2 槽）外都是 1 槽
+        itok = [i for i, n in enumerate(names) if n.startswith("itoken")]
+        n_extra = 1 if itok else 0
+        # 逐个候选：把 cd 的槽按名字顺序切（itoken 那两个槽合并看第一个）
+        mapping = {}
+        k = 0
+        for i, nm in enumerate(names):
+            if k >= len(cd):
+                break
+            if i in itok:
+                mapping[nm] = cd[k:k + 2]
+                k += 2
+            else:
+                mapping[nm] = [cd[k]]
+                k += 1
+        ok = no = unk = 0
+        for nm, vals in mapping.items():
+            st = likes(nm, vals[0])
+            ok += st == "OK"
+            no += st == "NO"
+            unk += st == "??"
+        score = ok - no
+        print("  %-46s OK=%2d NO=%2d ??=%2d  分=%d"
+              % (Path(f).name, ok, no, unk, score))
+        if best is None or score > best[0]:
+            best = (score, f, mapping, ok, no)
+
+    print()
+    print(">>> 最自洽的构建: %s (OK=%d NO=%d)" % (Path(best[1]).name, best[3], best[4]))
+    print()
+    print("该假设下的 name -> 值 标注:")
+    for nm, vals in best[2].items():
+        st = likes(nm, vals[0])
+        mark = {"OK": " ", "NO": "!", "??": "?"}[st]
+        print("  %s %-40s %s" % (mark, nm, json.dumps(vals, ensure_ascii=False)[:70]))
+    return 0
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())