Append E6-cost v1.1 budget correction and timing caveat

This commit is contained in:
Frank14f committed 2026-11-14 12:03:52 +08:00
1 parent ffa9bdc931
commit 7673a2150f
3 files changed
+81

No files matched your search

@@ -0,0 +1,21 @@
# E6-cost 预算更正与计时说明 v1.1
本补充更新当前执行规则,取代冻结提交 `ffa9bdc9314fbe3d5e3ed037cdea1c4899639a31` 中 README 和 `conditional_budget.json` 所引用的 11:48:55 预算提议。原提交、17份文件、ZIP及其校验清单全部保留;原文中的通用 **2.5e9 次访问上限已撤销**,只作为历史记录。
## 当前规则
- 每场通用约束为 **30分钟墙钟、1.25 GiB内存**;每个新实件先做邻居计数预检。**5e8 次访问仅是预检预警,不是运行上限**,不再采用通用2.5e9上限。
- **778是具名的一次性例外**:既有计数1,891,156,963次访问,预留30分钟、约1 GiB。只有在 O2-rep v1 四项修改完成并经复审冻结、成本包已实读审核后,才可运行。这个例外不推广到其他实件或全库。
- 既定较小预留继续适用:850为30秒/512 MiB,666为2分钟/1.25 GiB。**超出预留即停止**,不能借通用30分钟约束自动放宽它们。778约1 GiB保留来源中的近似表达,不在此补充中悄悄改为更大预留。
- Type A旋转重算按**第二场**独立计时和预留,不能包含在第一场名义下。
- 表示规范冻结前继续保留778已有2000次抽样;R和完整四点源点不变。全库前须在分层panel先做计数预检;三件结果不能外推624件。源点稀疏化仍是需另行精度核验的新估计器。
依据:[Claude 11:51:46 撤回旧上限](https://arcticnimbus.slack.com/archives/C0C78UFKGRM/p1794628306196729)、[11:54:46 要求追加更正](https://arcticnimbus.slack.com/archives/C0C78UFKGRM/p1794628486205699)。时间均为2026-11-14,Asia/Shanghai。
## 778计时的不确定性
现有记录的内部 stratum 0 包含1300个节点,两个样点分别耗时 **0.147831471秒、0.010348175秒**。该层的均值外推贡献 **102.8168秒**;快/慢情景分别为13.4526秒和192.1809秒。这个样点差异显著影响总体快/慢情景的宽度;不能把约14分钟解读为精确确定的全量耗时。
Grok在[11:58:02实读审核](https://arcticnimbus.slack.com/archives/C0C78UFKGRM/p1794628682875319)中指出,较慢样点可能含冷启动影响;原因没有另行验证。保留原始样点及原有约14分钟规划值、约16分钟慢情景,**不剔除样点、不重拟合、不新增校准**。这些是规划情景,既不是置信区间,也不是保证上下界。
本次只追加规则与解释,不表示规范已冻结或计算已获即时启动;没有全节点运行、新表示实现、训练或其他试验。机器可读版本见 `conditional_budget.v1.1.json`,补充文件校验见 `SUPPLEMENT_v1.1_SHA256SUMS`。原ZIP仍为2,134,345字节,SHA256为 `76be2bdb8afbf9566e6a4d2ffa832deafa853f954f09e39d36887bf037b61b08`。
@@ -0,0 +1,2 @@
2e17ba1cd8c5b2a42608c042e6ab92b07c96f15891dac34d2d217b057e9ad4d8 BUDGET_SUPPLEMENT_v1.1.md
9896ef05930ca261099ac732b3d53638b318168553ce2dfa13fe1a33ac0d0ef4 conditional_budget.v1.1.json
@@ -0,0 +1,58 @@
{
"version": "1.1",
"status": "CONDITIONAL_ONLY_NO_RUN_STARTED",
"supersedes_budget_text_at_commit": "ffa9bdc9314fbe3d5e3ed037cdea1c4899639a31",
"preserves_original_files_and_zip": true,
"sources": {
"replacement_rule": "https://arcticnimbus.slack.com/archives/C0C78UFKGRM/p1794628306196729",
"append_only_correction_request": "https://arcticnimbus.slack.com/archives/C0C78UFKGRM/p1794628486205699",
"independent_cost_audit": "https://arcticnimbus.slack.com/archives/C0C78UFKGRM/p1794628682875319"
},
"withdrawn_general_visit_cap": 2500000000,
"general_gate": {
"per_field_wall_seconds": 1800,
"per_field_memory_gib": 1.25,
"per_field_memory_bytes": 1342177280,
"neighbor_count_precheck_required_for_each_new_part": true,
"neighbor_visit_alarm": 500000000,
"neighbor_visit_alarm_is_cap": false,
"general_neighbor_visit_cap": null
},
"case_reservations": {
"00000850": {"wall_seconds": 30, "memory_mib": 512},
"00000666": {"wall_seconds": 120, "memory_gib": 1.25},
"00000778": {
"named_one_off_exception": true,
"observed_neighbor_visits": 1891156963,
"wall_seconds": 1800,
"memory_reservation_source_text": "approximately 1 GiB",
"memory_gib_approximate": 1.0,
"does_not_authorize_other_parts": true
}
},
"stop_rule": "Stop on an exceeded case reservation or general gate; smaller reservations remain binding, including 666 at 120 seconds.",
"type_A_rotation": "A separate second field with its own reservation and timing.",
"required_before_execution": [
"O2-rep v1 four requested edits completed",
"Revised specification re-reviewed and frozen",
"Calibration package read and audited"
],
"this_supplement_confirms_spec_freeze": false,
"until_spec_freeze": "Keep existing 2000-draw 778 result; unchanged R and all four-point sources; no full-node run.",
"corpus_rule": "Count-only precheck on a stratified panel before corpus-scale work; no three-case extrapolation to 624.",
"source_subsampling": "New estimator requiring its own accuracy validation.",
"timing_caveat_778": {
"group": "interior",
"stratum": 0,
"nodes": 1300,
"sample_query_seconds": [0.14783147102571093, 0.010348174982937053],
"estimated_integral_contribution_seconds": 102.81676990562119,
"interpretation": "Large sample disparity materially affects the scenario width; a cold-sample explanation is unverified. Retain original approximately 14-minute estimate and approximately 16-minute slow scenario; neither is a confidence interval or guaranteed bound.",
"samples_removed": false,
"timing_refitted": false,
"new_calibration": false
},
"new_full_node_runs": 0,
"new_representation_implementation": false,
"new_training_or_other_experiments": false
}