Add TwoTower scratch training-loop run records (E540–E548) and update model card/README

[Tyler-R-Kendrick]

Motivation

• Record local CPU-only TwoTower training-loop continuity evidence for several short scratch/resume runs (E540–E548) so experiment history and model-card tables reflect recent local activity.
• Surface those runs in the public-facing documentation (README.md, docs/MODEL_CARD.md, and docs/design/checkpoint-bucket.md) so the checkpoint ledger and measured-results table stay current.
• Persist per-run metadata as reproducible JSON artifacts under docs/design/ for auditability and future analysis.

Description

• Added JSON run artifacts docs/design/e540-...json through docs/design/e548-...json capturing telemetry and recipe metadata for runs e540_training_loop_scratch_20260719 through e548_training_loop_twotower_scratch_20260720.
• Updated README.md and docs/MODEL_CARD.md to insert table rows documenting the new E540–E548 entries with brief run summaries and local checkpoint references.
• Updated docs/design/checkpoint-bucket.md measured-results table to include the new run rows and to mark these runs as local scratch evidence with no eval or HF sync.
• No model code, training logic, or tests were changed; this is documentation and artifact addition only.

Testing

• No automated unit or integration tests were required or modified for these documentation/artifact changes.
• The changes are limited to documentation and JSON metadata files and therefore no CI test failures are expected from code paths.
• Manual inspection was used to verify that the new table rows and JSON files are well-formed and consistent with the existing run-record schema.

---

Codex Task

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
slm-training	Ready	Preview, Comment	Jul 20, 2026 3:04am

[coderabbitai]

Warning

Review limit reached

@Tyler-R-Kendrick, you've reached your PR review limit, so we couldn't start this review.

Next review available in: 50 minutes

Enable usage-based reviews in Billing to review now. Otherwise, wait until the next included review is available.
You're only billed for reviews past your plan's rate limits ($0.25/file).

How can I continue?

After more reviews become available, a review can be triggered using the @coderabbitai review command as a PR comment. Alternatively, push new commits to this PR.

To avoid repeated limits, reduce automatic review volume by pausing incremental auto-reviews earlier, using label-based review opt-in, excluding WIP or generated PR titles, or requesting reviews manually when the PR is ready. If your team needs uninterrupted high-volume reviews, an organization admin can enable usage-based reviews.

How do review limits work?

CodeRabbit enforces per-developer PR review limits for each organization. Most developers receive the normal plan review availability.

For paid Pro and Pro+ PR reviews, CodeRabbit uses adaptive limits for sustained high-volume activity. When a developer's recent PR review activity reaches the 95th percentile or higher among CodeRabbit users, additional reviews become available more gradually as earlier reviews age out of the rolling window.

Please refer docs for additional details.

Review details

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro Plus

Run ID: ddf9334d-7d1b-4de9-9a95-cff328530852

📥 Commits

Reviewing files that changed from the base of the PR and between 7c71031 and 674d79a.

📒 Files selected for processing (24)

• README.md
• docs/MODEL_CARD.md
• docs/design/checkpoint-bucket.md
• docs/design/e540-training-loop-scratch-20260719.json
• docs/design/e541-training-loop-twotower-scratch-20260719.json
• docs/design/e542-training-loop-twotower-resume-scratch-20260719.json
• docs/design/e543-training-loop-twotower-resume-scratch-20260719.json
• docs/design/e544-training-loop-twotower-scratch-20260719.json
• docs/design/e545-training-loop-twotower-scratch-20260719.json
• docs/design/e546-training-loop-twotower-scratch-20260720.json
• docs/design/e547-training-loop-twotower-scratch-20260720.json
• docs/design/e548-training-loop-twotower-scratch-20260720.json
• docs/design/iter-ldi4-02-sae-decision-state-diagnostic-20260720.md
• docs/design/iter-ldi4-03-intervention-unification-20260720.md
• docs/design/iter-sde4-02-min-controller-capacity-20260720.md
• docs/design/iter-slm127-candidate-selector-20260719.md
• docs/design/iter-slm130-canonical-ast-dedup-20260719.md
• docs/design/iter-slm135-trailed-assumptions-20260720.md
• docs/design/iter-slm138-recursive-denoiser-20260720.md
• docs/design/iter-slm144-plan-predictor-20260720.md
• docs/design/iter-spv0-03-semantic-regret-20260719.md
• scripts/run_semantic_regret_fixture.py
• src/slm_training/resources/versions.json
• tests/test_scripts/test_run_twotower_adapter_subspace_geometry.py

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch codex/run-continuous-training-loop

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}