research(d60): ship the G2/G5 script and report per-switch flips (issue #149)
The G2 precheck and the G5 noise floor had only a results file; the script that produces them is now in scripts/research/d60_two_way_g2_g5.py, with a test that reproduces the committed JSON byte for byte. G5 is now reported per switch, with the ayanamsa offset in arcminutes. Against the Raman baseline all three alternatives flip 77/77. Between the alternatives themselves the number follows the offset: Lahiri to KP 21/77 (27.3%, 5.81'), KP to True Chitra 19/77 (24.7%, 5.02'), Lahiri to True Chitra 2/77 (2.6%, 0.83'). The report says so and no longer claims that any frame change flips every chart; the registered gate is baseline relative and still triggers. The report also states that G2's "not evaluable" comes from the registered same-day answer policy, not from a measurement that D60 does not help. Co-Authored-By: Grok (agent loop)
This commit is contained in:
jesse-ux
co-authored by
Grok (agent loop)
Grok (agent loop)
parent
09620c6e0a
commit
3bc82acba6
+1
-1
@@ -20472,7 +20472,7 @@
|
||||
- 触发条件：用 D60 换格给候选分段打分，或对外宣称「到秒」。
|
||||
- 根因：未验证的技法被当成能力宣传；我方没有 D60 换格的评分实现，也没有对照实验。
|
||||
- 结论（研究单已做完）：G1 77/77 通过（BUG-1500 修好映射之后）。G2 按先登记口径**不可评估**：D60 换格题只认同日事件，v5 的 77 例里出生当天事件 0 件，B 组加不上任何判据、与 A 组相同。G3 平凡成立；G4 不适用。G5 触发上限：只换岁差（raman→lahiri / kp / true_citra），D60 上升格 77/77 全变（100%，阈值 20%）。因此**不做**：D60 不进校正判据，技法路由与文案不改；对外不出现「到秒 / 分钟级精确」，与 BUG-1240 同一口径。
|
||||
- 验证：报告 `docs/research/rectification_d60_two_way_2026_10_29.md`；结果 `docs/research/d60_two_way_results_2026_10_29.json`（G1）与 `docs/research/d60_two_way_g2_g5_results_2026_10_29.json`（G2 预检与 G5）；先登记 `docs/research/d60_two_way_preregistration_2026_10_28.json`。
|
||||
- 验证：报告 `docs/research/rectification_d60_two_way_2026_10_29.md`；结果 `docs/research/d60_two_way_results_2026_10_29.json`（G1）与 `docs/research/d60_two_way_g2_g5_results_2026_10_29.json`（G2 预检与 G5），分别由 `scripts/research/d60_two_way_run.py` 与 `scripts/research/d60_two_way_g2_g5.py` 生成，可逐字节复跑；先登记 `docs/research/d60_two_way_preregistration_2026_10_28.json`。
|
||||
- 防复发：研究单写死「先登记再跑、真值不可见、必须有安慰剂、负结果照写」。
|
||||
- 关联：BUG-1240、BUG-560、BUG-1090、BUG-1091。
|
||||
- 复发自：无。
|
||||
|
||||
File diff suppressed because it is too large
Load Diff
@@ -1,51 +1,68 @@
|
||||
# D60 换格要不要进生时校正：双向对比报告（2026-10-29，issue #149）
|
||||
|
||||
- 任务书：`docs/tasks/TASK-rectification-d60-two-way-20261028.md`；修复单 `docs/tasks/TASK-rectification-d60-fix-20261029.md`
|
||||
- 任务书：`docs/tasks/TASK-rectification-d60-two-way-20261028.md`；修复单 `docs/tasks/TASK-rectification-d60-fix-20261029.md`、`docs/tasks/TASK-rectification-d60-t5-fix-20261029.md`
|
||||
- 先登记：`docs/research/d60_two_way_preregistration_2026_10_28.json`（commit 早于任何计分）
|
||||
- 数据：v5 77 例、964 件事（`minute_rectification_holdout_v5.json`，sha256 `ab365a3e…`）
|
||||
- 口径：ayanamsa raman、node mean、候选步长 2 分钟、半径 ±10 / ±30 / ±60
|
||||
- 脚本：`scripts/research/d60_two_way_run.py`（G1）、`scripts/research/d60_two_way_g2_g5.py`（G2 预检与 G5），输出排序、无时间戳，可逐字节复跑
|
||||
|
||||
## 结论表
|
||||
|
||||
| 门 | 结果 | 关键数字 | 对照 |
|
||||
| --- | --- | --- | --- |
|
||||
| G1 三引擎对账 | **过** | 77/77 与 PyJHora 一致 | 修映射前 33/77（BUG-1500） |
|
||||
| G2 找回记录分钟 | **不可评估**（按登记口径） | 可用 D60 换格题 **0 条**：77 例里没有一件事与出生同日 | B 组与 A 组完全相同，提升恒为 0，不是「测出来没有帮助」 |
|
||||
| G2 找回记录分钟 | **不可评估**（口径造成） | 可用 D60 换格题 **0 条**：77 例里出生当天事件 0 件 | B 组与 A 组完全相同，提升恒为 0 |
|
||||
| G3 不伤真值 | 平凡成立 | B 组不淘汰任何候选，区间与 A 相同 | A 基线 76/77 |
|
||||
| G4 安慰剂 | **不适用** | 没有可检验的提升 | — |
|
||||
| G5 噪声地板 | **触发上限** | 换岁差后 D60 上升格 **77/77 全变**（Lahiri / KP / True Chitra 都是 100%，阈值 20%） | — |
|
||||
| G5 噪声地板 | **触发上限** | 以 Raman 为基准换岁差：三种换法都是 **77/77（100%，阈值 20%）**；换格幅度见下表 | — |
|
||||
|
||||
**产品结论：不做。** 维持「D60 只作参考，不驱动结论」；对外文案不得出现「到秒」「分钟级精确」，与 BUG-1240 的纳迪结论同一口径。
|
||||
|
||||
## 为什么 G2 是「不可评估」而不是「没有帮助」
|
||||
|
||||
登记的口径（任务书 §3.3）是：每次 D60 换格当一题「事件在换格之前还是之后」，**只有事件日期与换格同一天才作答**，否则跳过、不淘汰。D60 换格发生在出生当天，因此只有出生当天的事件能回答这类题。v5 的 77 例、964 件事里，出生当天的事件是 **0** 件（逐例核对，见结果 JSON）。于是 B 组一条额外判据都加不上，A、B 两组逐字节相同。
|
||||
G2 的计分口径是上一轮修正单 §3.3 定的：每次 D60 换格当一题「事件在换格之前还是之后」，**只有事件日期与换格同一天才作答**。D60 换格发生在出生当天，因此只有出生当天的事件能回答这类题；v5 的 77 例、964 件事里出生当天的事件是 **0** 件（逐例核对，见结果 JSON 的 `g2_precheck.per_case`）。于是 B 组一条额外判据都加不上，A、B 两组逐字节相同。
|
||||
|
||||
这条限制来自证据格式，不是 D60 本身：v5 只记录日期（244 件日精度、其余到月或年），没有当天时刻。若要用 D60 换格判分钟，需要带时刻的事件，或另一套能给出时刻的证据；这批数据给不了。
|
||||
**这是口径设计造成的不可评估，不是把 D60 测出了「没用」。** 规划方（Claude）在本轮验收里也认领了这一点：这个口径注定 0 题。
|
||||
|
||||
## G5：D60 在分钟尺度上完全取决于岁差
|
||||
## G5：分口径的数字
|
||||
|
||||
同一批盘、同一个出生时刻，只换岁差（raman → lahiri / kp / true_citra），上升的 D60 星座 **77 例全部改变**（100%，阈值 20%）。换句话说，D60 上升格在 2 分钟量级上本来就会随口径翻格；任何「D60 把关可以把时间收到秒」的说法，在这个噪声地板上站不住。
|
||||
基准是数据声明的 Raman。换岁差后记录时刻的 D60 上升格变化：
|
||||
|
||||
外部资料的说法与此一致：D60 每格 0.5°、上升约每 2 分钟换一格，记录时间按 5 分钟取整就跨两三个 D60 上升格（VedaCharts、Nyovah，见 MEVG）。他们给的也是「出生时间要准到分钟」，不是「用 D60 反推能到秒」。
|
||||
| 换法 | 变了多少例 | 比例 | 岁差差值（角分，中位） |
|
||||
| --- | ---: | ---: | ---: |
|
||||
| Raman → Lahiri | 77 / 77 | 100% | 86.78′ |
|
||||
| Raman → KP | 77 / 77 | 100% | 80.97′ |
|
||||
| Raman → True Chitra | 77 / 77 | 100% | 85.95′（85.51–86.27） |
|
||||
|
||||
替代岁差两两之间（同一批盘、同一时刻）：
|
||||
|
||||
| 换法 | 变了多少例 | 比例 | 岁差差值（角分，中位） |
|
||||
| --- | ---: | ---: | ---: |
|
||||
| Lahiri → KP | 21 / 77 | 27.3% | 5.81′ |
|
||||
| KP → True Chitra | 19 / 77 | 24.7% | 5.02′ |
|
||||
| Lahiri → True Chitra | 2 / 77 | **2.6%** | 0.83′ |
|
||||
|
||||
读法：换格比例跟着两组岁差的差走。差值接近或超过一个 D60 格（30′）时必然大面积翻格（Raman 与三种替代都差 80′ 以上，100%）；差值只有约 0.8′ 时（Lahiri 与 True Chitra）只有 2.6%。登记的门是「相对基准 Raman 的换法」，三种都 100%，超过 20% 阈值，所以触发「不得出现到秒 / 分钟级精确」。
|
||||
|
||||
**对「低于 20% 就重写结论」的处理**：Lahiri→True Chitra 2.6% 确实低于 20%。这一对是两套几乎重合的岁差（差 0.83′），它说明的是「岁差接近时 D60 不会翻」，而不是「D60 对口径不敏感」。因此我把 G5 的措辞改成分口径报数，不再说「换任何岁差都全变」；产品建议不变——主流岁差之间相差 5–87′，而哪一套岁差才是"对"的本身没有共识，用 D60 去把出生时间收到秒，仍然是用一个比答案更不确定的东西去定答案。若验收方认为这仍算「结论要重写」，请以验收意见为准，我按新的措辞改。
|
||||
|
||||
## 模型参与的端到端对比
|
||||
|
||||
`not_started`。按任务书的让步顺序在 G4 之后，而 G2/G4 没有可评估的提升；跑 40 场完整对话只会得到 A、B 逐字节相同的结果，不再消耗模型预算。
|
||||
`not_started`。按任务书的让步顺序在 G4 之后，而 G2/G4 没有可评估的提升；跑 40 场完整对话只会得到 A、B 逐字节相同的结果。
|
||||
|
||||
## MEVG（AGENTS.md §3）
|
||||
|
||||
- **MEVG / Global Web Evidence**：`used`。
|
||||
- **A（经典原文 / 可复现数据）**：BPHS 的 Shashtyamsa 规则（每星座 60 份、从本星座起算）与 Jataka Parijata / Bhavartha Ratnakara 的分盘权重，是可核对的原典；本仓的 12×60 段枚举与 PyJHora 对账属于可复现数据。原典只定义了几何与命名，**没有**「D60 可把出生时间收到秒」的判据。
|
||||
- **B（有案例的专业说明）**：[VedaCharts 的 D60 指南](https://vedacharts.com/learn/articles/shastiamsa-d60-subtle-karma) 明确「每格 0.5°、约 2 分钟换格，5 分钟取整的出生时间不可靠」；[Nyovah 的 D60 页](https://www.nyovah.com/learn/shashtiamsa-d60.html) 用自家引擎做了 24 小时逐分钟扫描，量到上升每分钟 0.25°、约 2 分钟换格，并明确说**不**把 D60 喂给问答模型，因为两分钟误差就会翻盘。两者都把「需要准到分钟」当成前提，没有给出「能收到秒」的证据。
|
||||
- **A（经典原文 / 可复现数据）**：BPHS 的 Shashtyamsa 规则（每星座 60 份、从本星座起算）与 Jataka Parijata / Bhavartha Ratnakara 的分盘权重是可核对的原典；本仓的 12×60 段枚举与 PyJHora 对账是可复现数据。原典只定义几何与命名，**没有**「D60 可把出生时间收到秒」的判据。
|
||||
- **B（有案例的专业说明）**：[VedaCharts 的 D60 指南](https://vedacharts.com/learn/articles/shastiamsa-d60-subtle-karma) 明确「每格 0.5°、约 2 分钟换格，5 分钟取整的出生时间不可靠」；[Nyovah 的 D60 页](https://www.nyovah.com/learn/shashtiamsa-d60.html) 用自家引擎做了 24 小时逐分钟扫描，量到上升每分钟 0.25°、约 2 分钟换格，并明确说**不**把 D60 喂给问答模型，因为两分钟误差就会翻盘。两者都把「需要准到分钟」当前提，没有给出「能收到秒」的证据。
|
||||
- **C（无依据的宣传）**：竞品「证据够密时 D60 与纳迪反向校准、收束可到秒」——无可复现数据、无案例、无算法说明，归 C。纳迪那一半已在 BUG-1240 证伪（N3 找回记录分钟 5/77、N4 换岁差 73/73 换段）。
|
||||
- 冲突仲裁：B 类来源给的是「D60 需要准到分钟」而 C 类结论是「D60 反推能到秒」；前者有测量，后者没有，按有测量的一方取舍，并降级 C 类结论。
|
||||
- **Real Case Calibration**：`used`。v5 77 例、964 件事，G1 77/77；G2 因证据格式不可评估；G5 100% 翻格。
|
||||
- 冲突仲裁：B 类给的是「D60 需要准到分钟」，C 类主张「D60 反推能到秒」；前者有测量，后者没有，按有测量的一方取舍，并降级 C 类结论。
|
||||
- **Real Case Calibration**：`used`。v5 77 例、964 件事；G1 77/77；G2 因口径不可评估；G5 见上表。
|
||||
|
||||
## 事后（post-hoc，不参与判门）
|
||||
|
||||
- 查了「如果放宽成『事件日期与换格日期至少差一天就作答』会怎样」：那样每条换格题都会被答成「事件在之后」，把窗口切掉一半且与真值无关，属于一边倒淘汰，因此不把它当结果。这一条写在先登记文件之后，只作解释。
|
||||
- 前端 `report-density-fictional-engine.json` 里仍有旧 D60 定义字符串的同类问题（与映射无关），留给前端轮次。
|
||||
- 「如果放宽成『事件日期与换格日期至少差一天就作答』」：每条换格题都会被答成「事件在之后」，把窗口切掉一半且与真值无关，属于一边倒淘汰，因此不当结果。此条写在先登记文件之后，只作解释。
|
||||
- 前端 `report-density-fictional-engine.json` 里仍有旧的 D60 定义字符串（与本次映射无关），留给前端轮次。
|
||||
|
||||
## 建议
|
||||
|
||||
@@ -57,8 +74,7 @@
|
||||
|
||||
```bash
|
||||
export PYTHONHASHSEED=0
|
||||
python scripts/research/d60_two_way_run.py # G1，写 docs/research/d60_two_way_results_2026_10_29.json
|
||||
python - <<'PY' # G2 预检 + G5，写 docs/research/d60_two_way_g2_g5_results_2026_10_29.json
|
||||
# 见 PROGRESS「第 2 次回报」一节的完整命令
|
||||
PY
|
||||
python scripts/research/d60_two_way_run.py # G1 → docs/research/d60_two_way_results_2026_10_29.json
|
||||
python scripts/research/d60_two_way_g2_g5.py # G2 预检 + G5 → docs/research/d60_two_way_g2_g5_results_2026_10_29.json
|
||||
python -m pytest tests/test_d60_two_way_g2_g5.py tests/test_d60_two_way_research.py tests/test_d60_varga_mapping.py -q
|
||||
```
|
||||
|
||||
@@ -104,3 +104,16 @@ python scripts/research/d60_two_way_run.py
|
||||
**结论**：不做。D60 不进校正判据，技法路由与对外文案不改；对外不出现「到秒 / 分钟级精确」。BUG-1497 记 `closed_by_design`。
|
||||
|
||||
**偏离**：任务书要求五门都跑；G2/G4 在登记口径下不可评估，按先登记文件的 stop rule 如实记录，不用放宽口径去凑数字；端到端对比按让步顺序未跑。用户可见代码本轮零改动。
|
||||
|
||||
## T5 第 1 次打回：补脚本与分口径报数（2026-10-29）
|
||||
|
||||
修复单 `docs/tasks/TASK-rectification-d60-t5-fix-20261029.md`。
|
||||
|
||||
1. **补脚本**：`scripts/research/d60_two_way_g2_g5.py`（G2 预检 + G5），输出排序、无时间戳；`tests/test_d60_two_way_g2_g5.py` 4 项（同日预检一致、汇总与逐例一致、无出生时刻泄漏、**结果文件逐字节复跑**）。G1 脚本已在上一轮提交。
|
||||
2. **G5 分口径**：
|
||||
- 以 Raman 为基准：Lahiri 77/77（86.78′）、KP 77/77（80.97′）、True Chitra 77/77（85.95′）。
|
||||
- 替代岁差两两：Lahiri→KP 21/77（27.3%，5.81′）、KP→True Chitra 19/77（24.7%，5.02′）、**Lahiri→True Chitra 2/77（2.6%，0.83′）**。
|
||||
- 结论措辞改成分口径：不再说「换任何岁差都全变」；登记门（相对基准）仍三种都 100%。2.6% 这一对是两套几乎重合的岁差，已在报告里写明，并说明若验收方认定「低于 20% 就要重写结论」，以验收意见为准。
|
||||
3. **报告补口径说明**：G2 不可评估是「只认同日事件」的口径造成的（Claude 已在验收里认领），不是 D60 被测出没帮助。
|
||||
|
||||
跑过：`pytest tests/test_d60_two_way_g2_g5.py` 4 passed；`pytest tests/test_d60_two_way_research.py tests/test_d60_varga_mapping.py` 9 passed。
|
||||
|
||||
@@ -0,0 +1,197 @@
|
||||
#!/usr/bin/env python3
|
||||
"""G2 precheck and G5 noise floor for the D60 two-way study (issue #149).
|
||||
|
||||
Two measurements, both deterministic (no timestamps, sorted keys, fixed
|
||||
formatting), so a rerun can be compared byte for byte:
|
||||
|
||||
* G2 precheck: how many v5 cases have an event dated on the birth day. The
|
||||
registered answer policy (`docs/research/d60_two_way_preregistration_2026_10_28.json`)
|
||||
only answers a D60 boundary probe when an event shares the boundary's date,
|
||||
so zero same-day events means group B adds no question at all.
|
||||
* G5: how often the ascendant's D60 sign at the recorded time changes when the
|
||||
ayanamsa is switched away from Raman, per switch, with the ayanamsa offset in
|
||||
arcminutes.
|
||||
|
||||
Usage:
|
||||
python scripts/research/d60_two_way_g2_g5.py # writes the results JSON
|
||||
python scripts/research/d60_two_way_g2_g5.py --json # prints it
|
||||
|
||||
PyJHora is not needed here; only the production ephemeris and the D60 mapping.
|
||||
"""
|
||||
|
||||
from __future__ import annotations
|
||||
|
||||
import argparse
|
||||
import hashlib
|
||||
import json
|
||||
import sys
|
||||
from pathlib import Path
|
||||
|
||||
ROOT = Path(__file__).resolve().parents[2]
|
||||
if str(ROOT) not in sys.path:
|
||||
sys.path.insert(0, str(ROOT))
|
||||
SCRIPTS = ROOT / "scripts"
|
||||
if str(SCRIPTS) not in sys.path:
|
||||
sys.path.insert(0, str(SCRIPTS))
|
||||
|
||||
from scripts.jyotish_engine import compute_chart_data # noqa: E402
|
||||
|
||||
DATASET = ROOT / "references/real_case_calibration/minute_rectification_holdout_v5.json"
|
||||
DEFAULT_OUT = ROOT / "docs/research/d60_two_way_g2_g5_results_2026_10_29.json"
|
||||
BASELINE = "raman"
|
||||
ALTERNATIVES = ("lahiri", "kp", "true_citra")
|
||||
G5_THRESHOLD = 0.20
|
||||
|
||||
|
||||
def _cases() -> list[dict]:
|
||||
return json.loads(DATASET.read_text(encoding="utf-8"))["cases"]
|
||||
|
||||
|
||||
def _ascendant(case: dict, ayanamsa: str) -> tuple[float, int]:
|
||||
import varga
|
||||
|
||||
birth = case["birth"]
|
||||
year, month, day = (int(part) for part in birth["date"].split("-"))
|
||||
hour, minute = (int(part) for part in birth["time"].split(":"))
|
||||
chart, _asc_idx, _jd, ayanamsa_degrees = compute_chart_data(
|
||||
year,
|
||||
month,
|
||||
day,
|
||||
hour,
|
||||
minute,
|
||||
birth["latitude"],
|
||||
birth["longitude"],
|
||||
birth["timezone_offset"],
|
||||
node_mode="mean",
|
||||
ayanamsa_name=ayanamsa,
|
||||
)
|
||||
longitude = float((chart.get("ascendant") or {})["degree_raw"])
|
||||
return float(ayanamsa_degrees), int(varga.calc_varga(longitude, 60)["sign_idx"])
|
||||
|
||||
|
||||
def same_day_precheck(cases: list[dict] | None = None) -> dict:
|
||||
cases = _cases() if cases is None else cases
|
||||
per_case = []
|
||||
for case in cases:
|
||||
birth_date = case["birth"]["date"]
|
||||
same_day = [e for e in case.get("events", []) if str(e.get("date", "")).startswith(birth_date)]
|
||||
per_case.append({"case_id": case["case_id"], "events_same_day": len(same_day)})
|
||||
return {
|
||||
"cases_with_same_day_events": sum(1 for row in per_case if row["events_same_day"]),
|
||||
"same_day_events": sum(row["events_same_day"] for row in per_case),
|
||||
"usable_d60_boundary_probes": 0,
|
||||
"verdict": "not_evaluable_by_construction",
|
||||
"reason": (
|
||||
"the registered answer policy needs an event dated on the same day as the D60 boundary "
|
||||
"(the birth date); the v5 holdout has none, so group B adds no question and is identical "
|
||||
"to group A. This is a design consequence of the registered rule, not a measurement that "
|
||||
"D60 does not help."
|
||||
),
|
||||
"per_case": per_case,
|
||||
}
|
||||
|
||||
|
||||
def g5(cases: list[dict] | None = None) -> dict:
|
||||
cases = _cases() if cases is None else cases
|
||||
modes = (BASELINE,) + ALTERNATIVES
|
||||
rows = []
|
||||
for case in cases:
|
||||
base_ayanamsa, base_sign = _ascendant(case, BASELINE)
|
||||
row = {
|
||||
"case_id": case["case_id"],
|
||||
f"d60_{BASELINE}": base_sign,
|
||||
f"ayanamsa_offset_arcminutes_{BASELINE}": 0.0,
|
||||
}
|
||||
for mode in ALTERNATIVES:
|
||||
ayanamsa, sign = _ascendant(case, mode)
|
||||
row[f"d60_{mode}"] = sign
|
||||
row[f"ayanamsa_offset_arcminutes_{mode}"] = round((ayanamsa - base_ayanamsa) * 60.0, 2)
|
||||
for mode in ALTERNATIVES:
|
||||
row[f"changes_{mode}"] = row[f"d60_{mode}"] != row[f"d60_{BASELINE}"]
|
||||
rows.append(row)
|
||||
count = len(rows)
|
||||
switches = {}
|
||||
for mode in ALTERNATIVES:
|
||||
offsets = sorted(row[f"ayanamsa_offset_arcminutes_{mode}"] for row in rows)
|
||||
changed = sum(1 for row in rows if row[f"changes_{mode}"])
|
||||
switches[f"{BASELINE}_to_{mode}"] = {
|
||||
"changed_cases": changed,
|
||||
"share": round(changed / count, 4) if count else None,
|
||||
"ayanamsa_offset_arcminutes": {
|
||||
"min": offsets[0] if offsets else None,
|
||||
"median": offsets[len(offsets) // 2] if offsets else None,
|
||||
"max": offsets[-1] if offsets else None,
|
||||
},
|
||||
}
|
||||
pairwise = {}
|
||||
for index, left in enumerate(modes):
|
||||
for right in modes[index + 1:]:
|
||||
changed = sum(1 for row in rows if row[f"d60_{left}"] != row[f"d60_{right}"])
|
||||
deltas = sorted(
|
||||
round(row[f"ayanamsa_offset_arcminutes_{right}"] - row[f"ayanamsa_offset_arcminutes_{left}"], 2)
|
||||
for row in rows
|
||||
)
|
||||
pairwise[f"{left}_to_{right}"] = {
|
||||
"changed_cases": changed,
|
||||
"share": round(changed / count, 4) if count else None,
|
||||
"ayanamsa_offset_arcminutes": {
|
||||
"min": deltas[0] if deltas else None,
|
||||
"median": deltas[len(deltas) // 2] if deltas else None,
|
||||
"max": deltas[-1] if deltas else None,
|
||||
},
|
||||
}
|
||||
return {
|
||||
"baseline": BASELINE,
|
||||
"switches": switches,
|
||||
"pairwise": pairwise,
|
||||
"threshold": G5_THRESHOLD,
|
||||
"any_above_threshold": any(item["share"] > G5_THRESHOLD for item in switches.values())
|
||||
or any(item["share"] > G5_THRESHOLD for item in pairwise.values()),
|
||||
"cases": rows,
|
||||
}
|
||||
|
||||
|
||||
def run() -> dict:
|
||||
cases = _cases()
|
||||
precheck = same_day_precheck(cases)
|
||||
noise = g5(cases)
|
||||
return {
|
||||
"scope": "d60_two_way_g2_g5",
|
||||
"task": "docs/tasks/TASK-rectification-d60-two-way-20261028.md",
|
||||
"fix_brief": "docs/tasks/TASK-rectification-d60-t5-fix-20261029.md",
|
||||
"issue": 149,
|
||||
"bug": "BUG-1497",
|
||||
"preregistration": "docs/research/d60_two_way_preregistration_2026_10_28.json",
|
||||
"dataset": {
|
||||
"path": DATASET.relative_to(ROOT).as_posix(),
|
||||
"sha256": hashlib.sha256(DATASET.read_bytes()).hexdigest(),
|
||||
"cases": len(cases),
|
||||
},
|
||||
"g2_precheck": precheck,
|
||||
"g3": {
|
||||
"verdict": "trivially_equal",
|
||||
"reason": "B adds no elimination, so the delivered range equals A and truth-in-range cannot drop",
|
||||
},
|
||||
"g4": {"verdict": "not_applicable", "reason": "no B-minus-A lift exists to test against placebos"},
|
||||
"g5": noise,
|
||||
}
|
||||
|
||||
|
||||
def main() -> int:
|
||||
parser = argparse.ArgumentParser()
|
||||
parser.add_argument("--json", action="store_true")
|
||||
parser.add_argument("--out", default=None)
|
||||
args = parser.parse_args()
|
||||
payload = run()
|
||||
text = json.dumps(payload, ensure_ascii=False, indent=2, sort_keys=True) + "\n"
|
||||
if args.json:
|
||||
print(text, end="")
|
||||
return 0
|
||||
out = Path(args.out) if args.out else DEFAULT_OUT
|
||||
out.write_text(text, encoding="utf-8", newline="\n")
|
||||
print(out.relative_to(ROOT).as_posix())
|
||||
return 0
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
raise SystemExit(main())
|
||||
@@ -0,0 +1,65 @@
|
||||
"""G2 precheck and G5 noise floor for the D60 study (issue #149).
|
||||
|
||||
The committed results file must be reproducible byte for byte from the script;
|
||||
the fast checks below verify the summary against the per-case rows without
|
||||
touching the ephemeris.
|
||||
"""
|
||||
|
||||
from __future__ import annotations
|
||||
|
||||
import json
|
||||
from pathlib import Path
|
||||
|
||||
from scripts.research.d60_two_way_g2_g5 import (
|
||||
DATASET,
|
||||
DEFAULT_OUT,
|
||||
ALTERNATIVES,
|
||||
same_day_precheck,
|
||||
)
|
||||
|
||||
ROOT = Path(__file__).resolve().parents[1]
|
||||
|
||||
|
||||
def _committed() -> dict:
|
||||
return json.loads(Path(DEFAULT_OUT).read_text(encoding="utf-8"))
|
||||
|
||||
|
||||
def test_same_day_precheck_matches_the_committed_file() -> None:
|
||||
committed = _committed()
|
||||
live = same_day_precheck()
|
||||
assert live["cases_with_same_day_events"] == committed["g2_precheck"]["cases_with_same_day_events"] == 0
|
||||
assert live["same_day_events"] == committed["g2_precheck"]["same_day_events"] == 0
|
||||
assert committed["g2_precheck"]["verdict"] == "not_evaluable_by_construction"
|
||||
|
||||
|
||||
def test_g5_summary_matches_the_committed_rows() -> None:
|
||||
committed = _committed()
|
||||
rows = committed["g5"]["cases"]
|
||||
assert len(rows) == committed["dataset"]["cases"] == 77
|
||||
for mode in ALTERNATIVES:
|
||||
changed = sum(1 for row in rows if row[f"d60_{mode}"] != row[f"d60_raman"])
|
||||
assert committed["g5"]["switches"][f"raman_to_{mode}"]["changed_cases"] == changed
|
||||
for key, switch in committed["g5"]["pairwise"].items():
|
||||
left, right = key.split("_to_")
|
||||
changed = sum(1 for row in rows if row[f"d60_{left}"] != row[f"d60_{right}"])
|
||||
assert switch["changed_cases"] == changed, key
|
||||
assert committed["g5"]["any_above_threshold"] is True
|
||||
|
||||
|
||||
def test_payload_has_no_recorded_birth_times() -> None:
|
||||
payload = _committed()
|
||||
assert "true_minute" not in json.dumps(payload["g5"]["cases"])
|
||||
for row in payload["g5"]["cases"]:
|
||||
assert set(row) == {"case_id", "d60_raman", "d60_lahiri", "d60_kp", "d60_true_citra",
|
||||
"ayanamsa_offset_arcminutes_raman", "ayanamsa_offset_arcminutes_lahiri",
|
||||
"ayanamsa_offset_arcminutes_kp", "ayanamsa_offset_arcminutes_true_citra",
|
||||
"changes_lahiri", "changes_kp", "changes_true_citra"}
|
||||
|
||||
|
||||
def test_results_file_is_reproducible_byte_for_byte() -> None:
|
||||
from scripts.research import d60_two_way_g2_g5 as module
|
||||
|
||||
payload = module.run()
|
||||
text = json.dumps(payload, ensure_ascii=False, indent=2, sort_keys=True) + "\n"
|
||||
assert text == Path(DEFAULT_OUT).read_text(encoding="utf-8")
|
||||
assert Path(DATASET).is_file()
|
||||
Reference in New Issue
Block a user