第 88 天：独立掌握度检索——无 AI 应证日

{
"declared_condition": "independent",
"time_cap_minutes": 105,
"tasks": [
{"domain":"PY","task_id":"PY-1","recovery_day":"day-09","declared_condition":"independent"},
{"domain":"PY","task_id":"PY-2","recovery_day":"day-11","declared_condition":"independent"},
{"domain":"PY","task_id":"PY-3","recovery_day":"day-12","declared_condition":"independent"},
{"domain":"GIT","task_id":"GIT-1","recovery_day":"day-14","declared_condition":"independent"},
{"domain":"GIT","task_id":"GIT-2","recovery_day":"day-80","declared_condition":"independent"},
{"domain":"GIT","task_id":"GIT-3","recovery_day":"day-81","declared_condition":"independent"},
{"domain":"MATH","task_id":"M-1","recovery_day":"day-19","declared_condition":"independent"},
{"domain":"MATH","task_id":"M-2","recovery_day":"day-25","declared_condition":"independent"},
{"domain":"MATH","task_id":"M-3","recovery_day":"day-27","declared_condition":"independent"},
{"domain":"GRAD","task_id":"G-1","recovery_day":"day-31","declared_condition":"independent"},
{"domain":"GRAD","task_id":"G-2","recovery_day":"day-32","declared_condition":"independent"},
{"domain":"GRAD","task_id":"G-3","recovery_day":"day-33","declared_condition":"independent"},
{"domain":"MODEL","task_id":"MR-1","recovery_day":"day-50","declared_condition":"independent"},
{"domain":"MODEL","task_id":"MR-2","recovery_day":"day-65","declared_condition":"independent"},
{"domain":"MODEL","task_id":"MR-3","recovery_day":"day-83","declared_condition":"independent"}
]
}
assert len(tasks) == 15 # 实际应列出全部 15 行

PY-1:
from collections import Counter
import re, unittest

def top_k_words(text, k):
words = re.findall(r"[a-z0-9']+", text.lower())
counts = Counter(words)
ranked = sorted(counts.items(), key=lambda kv: (-kv[1], kv[0]))
return [w for w, _ in ranked[:k]]

class TestTopK(unittest.TestCase):
def test_counts(self): self.assertEqual(top_k_words('b a a b c', 1), ['a'])
def test_ties(self): self.assertEqual(top_k_words('b a', 2), ['a', 'b'])
def test_empty(self): self.assertEqual(top_k_words('', 3), [])
def test_k_large(self): self.assertEqual(top_k_words('x', 5), ['x'])

GIT-1（一次性仓库）:
git branch fix/login # 提交保留到新分支
git checkout main
git reset --hard origin/main # 恢复 main；仅一次性克隆上安全
git checkout fix/login
git reflog # 定位丢失尖端: git branch rescued

DBG-1: KeyError 'total' → 最可能是键未初始化；首个命令: python -m unittest -v 或 print/pdb 断点。7 对 6 的差一错误 → 很可能是 range/slice 边界包含问题；首个命令: 打印索引轨迹检查 window 边界。

M-1: df/da=b+1=4.0，df/db=a=2.0。验证: (f(2.00001,3)−f(1.99999,3))/(2e-5)≈4.0±1e-6。
M-2: 1040+256+6144=7440——与 run_artifact.json 的 parameter_ledger.total_parameters 一致。
M-3: 位置 t 读到 t+1 会泄漏未来 token，使 next-token 损失可被平凡解出并破坏自回归生成；上三角掩码在 softmax 前把 j>t 的 logit 置为 -inf，使其注意力权重归零。
M-4: 可训练 32768 vs 稠密 1,048,576（约 3.1%）；适配器输出在叠加到冻结基权重输出前先按 alpha/r=32/16=2.0 缩放。
M-5: exp(4.0372)≈56.7——对 tinyshakespeare 的 10 步 CPU smoke 而言偏高但健全；它验证的是管线而非收敛。证伪项：训练损失上升、NaN/Inf 梯度、检查点重载不一致。

{
"honesty_rules": "Independent checks evidence unaided ability only; assisted results never substitute; unmet criteria remain visibly unmet; no official certification is implied.",
"scores": {"PY-1": 2, "PY-2": 1, "GIT-1": 2, "M-1": 2, "MR-1": 1},
"domain_subtotals": {"PY": 5, "GIT": 5, "MATH": 5, "GRAD": 6, "MODEL": 5},
"domain_pass": {"PY": true, "GIT": true, "MATH": true, "GRAD": true, "MODEL": true},
"ledger": [
{"task_id":"PY-2","miss_type":"slip","recovery_day":"day-11","next_attempt_date":"2026-11-01"},
{"task_id":"MR-1","miss_type":"concept_gap","recovery_day":"day-50","next_attempt_date":"2026-11-02"}
]
}