Skip to content

Rebuild README around measured results, with evidence and an architecture diagram - #37

Merged
Mark007-R merged 1 commit into
devfrom
feat/readme-with-evidence
Aug 5, 2026
Merged

Mark007-R merged 1 commit into
devfrom
feat/readme-with-evidence

Conversation

@Mark007-R

Copy link
Copy Markdown
Owner

Rebuilds the README so the claims are backed by files in the repo, and adds an architecture diagram.

Structure

Live link → what it is (plain prose) → architecture image → measured results → how it works → infrastructure → quick start → datasets → licence.

Evidence restored (8 files)

Only the artifacts the README actually cites, recovered from history:

baseline_metrics.json · day05/day05_leaderboard.csv · day06/ablation_modelling.csv · day06/frontier_comparison.csv · registry_rollback_times.csv · drift_replay_summary.json · drift_retrain_events.csv · throughput_speedup.csv

Every metric table links to the file it came from. Nothing else from the old results/ tree came back.

Numbers

Read directly out of the artifacts and re-verified after writing — not recalled:

Claim Value Source
Leakage fix 0.9210 → 0.7949 baseline_metrics.json
Champion vs AutoGluon 0.95196 vs 0.952 (−0.00004) day05_leaderboard.csv
Ablation L0→L3 0.5467 → 0.6574 → 0.6962 → 0.9480 (+0.4013) ablation_modelling.csv
Champion vs LLM 0.9156 vs 0.6225 · 60 µs vs 1.82 s · $0.43 vs $1,250,691 frontier_comparison.csv
Rollback 3.9 ms alias flip · 11.9 ms audited registry_rollback_times.csv
Drift precision 1.0 · recall 1.0 · 0-day lag drift_replay_summary.json
Retrain median 6.85 s · 3/3 promoted drift_retrain_events.csv
Dask determinism bit-exact to 5.5e-12 throughput_speedup.csv

Architecture diagram

assets/architecture.png — PNG, no SVG and no mermaid. Generated by assets/make_architecture.py (Pillow, drawn at 2× and downsampled), committed alongside it so the diagram is reproducible rather than a binary drop.

It uses a dark card with light text so it stays readable on both GitHub themes — an embedded SVG using currentColor would have rendered black-on-black in dark mode, which is why this is a raster.

Verification

Every link in the README was checked to resolve, and each headline figure was re-read from its artifact after the file was written.

Restores the eight metrics artifacts the README cites so every number is
verifiable in-repo, adds a PNG architecture diagram with the script that
generates it, and rewrites the README as: live link, what it is,
architecture, measured results, how it works, infrastructure, quick start.

Numbers were read from the artifacts, not from memory: leakage 0.9210 ->
0.7949, champion 0.95196 vs AutoGluon 0.952, ablation +0.4013 L0->L3,
frontier 0.9156 vs 0.6225, rollback 3.9 ms, drift precision/recall 1.0 at
0-day lag, retrain median 6.85 s.
Copilot AI lite review requested due to automatic review settings August 5, 2026 12:46
@Mark007-R
Mark007-R merged commit 8fb90e5 into dev Aug 5, 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 was unable to review this pull request because the user who requested the review has reached their quota limit.

@Mark007-R
Mark007-R deleted the feat/readme-with-evidence branch August 5, 2026 12:46
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.

2 participants