后续观察登记记录
本次于 2026 年 10 月 22 日复核既有规则，并补充本机登记工具。工具不会采集行情、启动定时任务或下单。当前没有新增已完成的未来收益。
低换手率
沿用 microcap-prospective-20261017/v1。主要指标固定为加入贝塔和估值控制后的低换手毛收益差，同一共同样本上的原始信号作对照。沿用 60 日窗口、至少 45 个观测和十组排序。首批 24 个合格且完成的新月份形成一次主要评价，HAC 滞后为 3 阶。月度读数保持描述性。
现有公开收据保留十月月末作为候选形成日。日期须由官方日历确认，输入须在当时封存。旧八月窗口仍缺可核验的事前时间凭据，继续保留为后续观察。
估值分组
沿用 冻结的向前验证协议。首次形成不早于 2026 年 11 月，并且晚于实际冻结时间。保留全部五组和既定对照，观察 12 个计划批次，最后检查 20 个交易日的退出窗口。当前尚未开始。
怎样留下登记记录
形成日收盘后，准备冻结规则、输入清单和官方日历文件。输入清单应绑定实际数据的内容哈希与首次取得时间，保存在仓库外。下一次入场前，运行 scripts/register_research_window.py，传入研究编号、规则文件、输入清单、日历文件、带时区的形成时间与入场时间，以及仓库外的新收据路径。
参数为 --study、--protocol、--inputs、--calendar、--formation-at、--entry-at、--output。研究编号使用 low-turnover 或 valuation。时间须含时区。输出目录须事先存在。工具记录当前 UTC 时间与三份文件的 SHA-256，不复制输入内容或私有路径。收据只允许新建，已有文件不会覆盖。
工具拒绝晚于入场才登记的批次，也拒绝在形成前把尚未取得的输入登记为现成数据。拒绝之后应留下错过冻结的记录，不能改动系统时钟或把新文件补写成旧日期。规则实质改变时建立新批次，保留旧记录。
本机时钟只能证明本机声称的记录时间。生成的收据固定为 pending_external_time_proof，接纳标记为假，已完成窗口为零。须另取得可独立核验、早于入场且绑定收据哈希的公开时间凭据，再人工复核规则、官方日历、首次可见数据和排名代码，才能接纳。普通 Git 提交的作者时间可以自行设定，不能单独充当这份证明。缺少独立凭据时继续待定。
后续月度结果应追加新收据，保留缺价、缺失月份和未决资产。没有结果时不填零，也不把登记收据改称收益证据。
Future-observation registration record
Reviewed on 22 October 2026. This update adds a local receipt tool and preserves the existing rules. It collects no market data, installs no scheduler and submits no orders. No new completed future return is reported.
Low turnover
Keep microcap-prospective-20261017/v1. The primary is the gross spread after beta and valuation controls, paired with the raw signal on identical support. Keep the 60-session feature, 45-observation minimum and ten groups. Evaluate the first 24 eligible completed new months once, with HAC lag 3. Monthly readings remain descriptive.
The retained protocol names the final October session as a candidate. Verify it against an official calendar and archive contemporaneous inputs before entry. The older August window still lacks verified pre-period time evidence and remains a later observation.
Valuation groups
Keep the frozen forward protocol. First formation must be no earlier than November 2026 and later than the actual freeze timestamp. Keep all five groups, fixed contrasts and execution rules. Observe 12 planned monthly cohorts followed by the 20-session exit window. This has not started.
Recording a window
After formation close, prepare the frozen protocol, input manifest and official calendar. The input manifest must bind actual data hashes and first-seen timestamps. Keep source data and receipts outside the repository. Before entry, run scripts/register_research_window.py with --study, --protocol, --inputs, --calendar, --formation-at, --entry-at and --output. Study IDs are low-turnover and valuation. Timestamps require explicit timezones; the output directory must already exist.
The tool records current UTC time and the SHA-256 of the three files. It includes neither input contents nor private paths. It exclusively creates a new receipt and refuses to overwrite an existing one. It rejects registration at or after entry and rejects formation inputs recorded before formation. Preserve missed-freeze records instead of changing the clock or backdating a file. Material rule changes require a new cohort and retention of old records.
A local clock is not independent time proof. Every generated receipt remains pending_external_time_proof, with admission false and zero completed windows. Before admission, obtain independently verifiable public time evidence that binds the receipt hash before entry. Review the rules, official calendar, first-seen data and ranking implementation separately. Ordinary Git author timestamps can be set by the author and cannot establish this proof alone. Without independent evidence, keep the window pending.
Append later outcome receipts and retain missing prices, absent months and unresolved assets. Missing outcomes are not zero returns, and a registration receipt is not performance evidence.