https://bensonlee5.github.io/tab-foundry/docs/reference/Recent content in Reference Index on tab-foundry DocsHugoentab-foundry contributorshttps://bensonlee5.github.io/tab-foundry/docs/reference/papers/Mon, 01 Jan 0001 00:00:00 +0000https://bensonlee5.github.io/tab-foundry/docs/reference/papers/<p>Start here when you want the reading list that actually informs architecture, training, and scaling decisions in <code>tab-foundry</code>.</p> <p>This list overlaps with <code>~/dev/dagzoo/reference</code>, but the lens here is <strong>architecture, training recipe, and scaling predictability</strong> rather than data generation.</p> <p>Use these alongside this reference set:</p> <ul> <li>Design decisions and repo structure: <code>docs/development/design-decisions.md</code></li> <li>Roadmap: <code>docs/development/roadmap.md</code></li> <li>Workflow runbooks: <code>docs/workflows.md</code></li> <li>Evidence mapping: <code>reference/evidence.md</code></li> </ul> <h2 id="adoption-tiers-for-this-repo">Adoption Tiers For This Repo</h2> <p>Use this directory to make explicit judgments, not just collect papers.</p>https://bensonlee5.github.io/tab-foundry/docs/reference/evidence/Mon, 01 Jan 0001 00:00:00 +0000https://bensonlee5.github.io/tab-foundry/docs/reference/evidence/<p>Use this map when you need to trace a roadmap claim back to papers and the most important repo-local evidence.</p> <p>Use these alongside this map:</p> <ul> <li>Canonical roadmap: <code>docs/development/roadmap.md</code></li> <li>Design decisions and repo structure: <code>docs/development/design-decisions.md</code></li> <li>Architecture reference: <code>docs/development/model-architecture.md</code></li> <li>Paper index: <code>reference/papers.md</code></li> <li>Detailed research-epic notes: <code>reference/roadmap_evidence/README.md</code></li> </ul> <p>Conventions:</p> <ul> <li>Confidence: <code>high</code>, <code>medium</code>, or <code>low</code></li> <li>Lane: roadmap milestone lane supported by this evidence (<code>Now</code>, <code>Next</code>, <code>Later</code>)</li> </ul> <h2 id="detailed-research-epic-notes">Detailed Research Epic Notes</h2> <p><code>reference/evidence.md</code> remains the compact cross-epic map. Canonical long-form evidence for TF-RD-018 and later research-oriented epics now lives under <a href="https://github.com/bensonlee5/tab-foundry/blob/main/reference/roadmap_evidence/README.md"><code>reference/roadmap_evidence/</code></a>.</p>https://bensonlee5.github.io/tab-foundry/docs/reference/tf-rd-013-support/Mon, 01 Jan 0001 00:00:00 +0000https://bensonlee5.github.io/tab-foundry/docs/reference/tf-rd-013-support/<p>Use this support bundle when you need the committed assumptions and comparison surfaces behind TF-RD-013 issue <code>#127</code>.</p> <p>It is a historical reference-only support surface for that shape-aware follow-up, not the main roadmap or sweep execution guide. Reconstructing these local artifacts is no longer a supported workflow.</p> <p>Environment assumptions:</p> <ul> <li><code>TAB_FOUNDRY_ROOT</code> is this repo root.</li> <li><code>DAGZOO_ROOT</code> defaults to the sibling checkout <code>../dagzoo</code>.</li> <li>The broader dagzoo follow-up uses three config-backed invocations: <ul> <li><code>../dagzoo/configs/benchmark_cpu.yaml</code> with <code>2048</code> datasets</li> <li><code>../dagzoo/configs/default.yaml</code> with <code>4096</code> datasets</li> <li><code>../dagzoo/configs/benchmark_cuda_h100_large_shape.yaml</code> with <code>128</code> datasets</li> </ul> </li> <li>The curated comparator baseline remains pinned to <code>data/manifests/bench/openml_classification_large_v1/manifest.parquet</code>.</li> </ul> <p>Historical support surface:</p>