Skip to content

Run frozen continuous capital studies and immutable revision pairs - #24

Merged
PureSaber merged 2 commits into
mainfrom
codex/continuous-capital-revision
Oct 12, 2026
Merged

PureSaber merged 2 commits into
mainfrom
codex/continuous-capital-revision

Conversation

@PureSaber

@PureSaber PureSaber commented Oct 11, 2026 •

Copy link
Copy Markdown
Owner

Summary

The portfolio CLI now freezes an actual executable/control identity, runs a continuous QExec shared account from a prespecified prior-observation notional policy, and reruns the same frozen plan against immutable old/new data. The revision report preserves both results and shows affected downstream objects plus signal, order, cash, NAV and return changes. Features/forecasts are explicitly unavailable in this finite adapter.

Reports emit the common quant.measurement/v1 contract with a strictly increasing UTC observation grid, explicit flow/cost/valuation policies and actual hashes. Same-timestamp events retain their sequence and compound into one observation. Ordinary comparisons intentionally reject old/new input revisions; their differences are reported only as an explicit data intervention.

Dependencies are pinned in metadata, lock and governance tests to QDK f85128307dd0e3407d24350282fe7b679b591309 (QDK PR) and QExec 49ef53d2c502ad6faad6818f853dadd1f8b8b182 (QExec PR).

Verification

  • Full Windows/Python 3.12 suite: 295 passed, 1 skipped; 83.90% coverage and existing cross_asset pure branch gate passed.
  • Final dependency/governance and continuous tests: 5 passed; Ruff check/format and pip check passed.
  • Existing synthetic demo passed. New real subprocess CLI acceptance independently expects old/new NAV 1034/1026 USD, PnL +34/+26 and delta -8; observed values matched. Added to CI on all Python versions.
  • A's actual validate_measurement and validate_observations accepted both reports; actual cost/valuation hash checks passed; identical-input comparison succeeds while revision pair comparison is rejected.
  • B's 9 produced forecast execution/settlement events were consumed by QExec: independent expected/actual PnL 17.60 USD, final cash 1017.60. Consumer margin was explicitly configured differently, so no margin or comparison equivalence is claimed.
  • Negative cases cover changed code/controls, future observations, same-time signal leakage, corrupt inputs and output overwrite refusal. Exact final-head CI and CodeQL passed on Python 3.10–3.12, including the real CLI acceptance (CI); no merge/deployment requested.

The workflow is offline synthetic or retrospective sensitivity. It establishes no market validity, forecast skill or live readiness. No secrets, paid data, generated run artifacts or real orders are added. Revert the additive workflow and coordinated dependency pins to roll back, retaining published history. Implements authorized batch items 4 and 5; no separate issue required.

@PureSaber
PureSaber merged commit 8b657e2 into main Oct 12, 2026
5 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant