Skip to content

docs: full README rewrite — temporal-split fix + MLOps capabilities + frontier guard-rail - #32

Merged
Mark007-R merged 1 commit into
devfrom
docs/readme-rewrite
May 31, 2026
Merged

docs: full README rewrite — temporal-split fix + MLOps capabilities + frontier guard-rail#32
Mark007-R merged 1 commit into
devfrom
docs/readme-rewrite

Conversation

@Mark007-R

@Mark007-R Mark007-R commented May 31, 2026

Copy link
Copy Markdown
Owner

What this is

A fully new README reflecting the 7-day MLOps sprint, replacing the legacy version.

Highlights

  • The leakage-fix headline: honest Sparkov OOT AUC 0.9210 (leaked) → 0.7949 (real) after replacing a random split with a per-source temporal split; then Optuna + source-balanced weights earn it back to 0.9520 (ties AutoGluon 0.952).
  • Four MLOps capabilities with source files: Dask bit-exact features (5.5e-12), 3.9 ms registry rollback, 0-day-lag KS+PSI drift detection, 6.85 s drift→promote auto-retrain.
  • Day-6 frontier guard-rail: champion 0.916 vs Claude LLM-judged 0.622 at 30,400× latency / 2.9M× cost; naive notebook XGBoost ties the LLM — the discipline is the model.
  • L0→L3 MLOps ablation (+0.401 OOT AUC), 31-test suite, docker-compose stack (Postgres + MLflow + Redis + FastAPI), CI gate.

Scope

Docs only — Readme.md rewrite (+127 / −563). No code touched. All numbers cite committed files under results/, results/day05/, results/day06/, and docs/.

@Mark007-R
docs: full README rewrite — temporal-split-fix story, MLOps capabilit…
2d32b2e 
…ies, frontier guard-rail

Replaces the legacy README with a results-driven one reflecting the 7-day MLOps sprint:
- The leakage-fix headline: honest Sparkov OOT AUC 0.9210 (leaked) -> 0.7949 (real),
  then Optuna + source-balanced weights earn it back to 0.9520 (ties AutoGluon)
- Four MLOps capabilities with source files: Dask bit-exact features, 3.9ms registry
  rollback, 0-day-lag KS+PSI drift detection, 6.85s drift->promote auto-retrain
- Day-6 frontier guard-rail: champion 0.916 vs Claude LLM 0.622 at 30,400x latency /
  2.9M x cost; naive notebook ties the LLM (discipline is the model)
- L0->L3 MLOps ablation (+0.401 OOT AUC), 31-test suite, docker-compose stack, CI gate
Copilot AI review requested due to automatic review settings May 31, 2026 17:35
@Mark007-R
Mark007-R merged commit 72013d3 into dev May 31, 2026
1 check passed
@Mark007-R
Mark007-R deleted the docs/readme-rewrite branch May 31, 2026 17:35
Copilot AI reviewed May 31, 2026

Copilot AI left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Copilot encountered an error and was unable to review this pull request. You can try again by re-requesting a review.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

Copilot code review Copilot
Copilot review effort, defaults to Lite
Applies to this pull request for everyone.Learn more about Copilot code review.
Copilot left review comments

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

2 participants

@Mark007-R