release: quantifact 0.2.0 candidate - #1
Closed
leoncuhk wants to merge 8 commits into
Closed
Conversation
… by construction An investment-research agent that has to prove its numbers. A question is clarified into a typed plan with a knowledge date; every task compiles in parallel into one pandas function; static analysis derives the dependency graph and cross-checks it against the plan; layered contracts run as ordinary Python; and a caching harness executes the code bound to the knowledge date, so look-ahead is structurally impossible rather than discouraged. 41 tests, two adapters (synthetic, DuckDB), reproducible benchmarks.
- report header said 'Pocket Analyst'; it now says quantifact and carries the knowledge date, which belongs at the top of the page rather than an appendix - 'five methods' -> six; '15 classes of compile error' -> 30 checks in 9 categories, both now matching what the code actually does - source docstrings no longer assume the reader watched a talk - README states the prior art in the words people search for, plus a FAQ
Thirteen passes, a real compiled plan with its generated code, the benchmarks, and a new point-in-time section showing the same question answered as of two knowledge dates — 3 episodes and 167 return cells versus 4 and 223, with different cache keys. The page embeds nothing private: no credentials, no vendor data, no local paths and no endpoint identity, only the model name, which is the part that makes a grading result mean anything. The endpoint was scrubbed from the committed grading JSONs for the same reason.
The action's default mode diffs <commit>^..HEAD, which has no parent on an initial push and fails for a reason unrelated to secrets. Scanning history and the working tree separately is both more honest and works from commit one.
Root now holds only what a visitor opens: README, CHANGELOG, LICENSE, NOTICE, CITATION. Community-health files moved to .github/ (GitHub reads them there), the disclaimer to docs/, and the duplicated markdown is gone — the API table folded into docs/index.md, two thin example READMEs into examples/README.md, and the byo-data example into its guide, which already carried the same DDL. Also: cache keys in the exported page are truncated to the ten characters it actually displays, and .gitleaks.toml allowlists content-addressed digests by field name rather than by file, so a real credential in those artefacts would still fail the scan.
Add fail-closed model planning, point-in-time document retrieval, complete run receipts, evidence-backed maturity gates, and installable artifact verification. Formal release remains blocked until traceable operating evidence passes the strict audit.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Outcome
Prepares an auditable 0.2.0 release candidate without claiming PAT-level maturity before operating evidence exists.
What changed
Verification
Release gate
Formal tag/PyPI publication intentionally remains blocked until a traceable
benchmarks/quality-evidence.jsonpassesqf audit --strict. Current repository-only audit is 40.6/100 and correctly reports missing real-task accuracy, diverse planner evaluation, tool coverage, learning outcomes, sandbox evidence, service SLOs, and expert adoption.