# Studied projects and research | PostTrainLLM > Canonical, source-linked evidence dossiers for every external product, model, platform, technical writeup, and industry case study retained by PostTrainLLM. Canonical: https://posttrainllm.com/studies Tracked external evidence # What we studied, what we kept, and what we rejected. Every external product, model, platform, writeup, and case study that changed the PostTrainLLM learning map gets one canonical record. Each dossier keeps the local translation beside its evidence limits so a useful idea never turns into an unsupported claim. Canonical studies: 20 Distinct dispositions: 11 Unresolved ownership: 0 ## All retained studies Indexed from the tracked review ledger and industry learning map. ### [TrainLoop AI](https://posttrainllm.com/studies/trainloop-ai) product review partially adopted Failed-attempt accounting, slice metrics, trace review, candidate selection, batch-first post-training, policy lag, and LoRA update geometry make autoresearch inspectable instead of hiding the search process behind a final score. ### [TrainLoop NollaMD case study](https://posttrainllm.com/studies/trainloop-nollamd) case-study review scaffolded Candidate-selection framing can make a sparse task measurable when unconstrained generation is too difficult or too ambiguous to grade reliably. ### [TrainLoop OAPL writeup](https://posttrainllm.com/studies/trainloop-oapl) technical writeup scaffolded Batch rollouts, offline scoring, and compact updates can make a post-training loop easier to inspect, while policy lag can stabilize learning under some reward surfaces. ### [TrainLoop LoRA geometry writeup](https://posttrainllm.com/studies/trainloop-lora-geometry) technical writeup scaffolded Effective update geometry, layer placement, and controlled rank analysis can explain why two adapters with similar parameter counts behave differently. ### [TrainLoop Mercor coding-agent post](https://posttrainllm.com/studies/trainloop-mercor-coding-agent) product writeup parked File-native and code-native harnesses may fit knowledge-work agents better than bespoke tool APIs because the environment already exposes durable state and familiar operations. ### [Baseten post-training positioning](https://posttrainllm.com/studies/baseten-post-training) product positioning adopted Post-training is a system of custom data, reward shaping, performance work, infrastructure, evaluation, and packaging rather than a generic fine-tuning screen. ### [Hugging Face model hub and SQL specialists](https://posttrainllm.com/studies/hugging-face-sql-specialists) model landscape tried Small public SQL models provide realistic baselines and expose how much benchmark, model size, and serving format affect apparently similar claims. ### [Defog SQLCoder and Arctic Text2SQL class models](https://posttrainllm.com/studies/defog-sqlcoder-arctic-text2sql) model landscape adopted as evaluation guidance Strong SQL systems report execution accuracy on substantial SQL benchmarks rather than relying on exact string match alone. ### [Apple on-device Foundation Models](https://posttrainllm.com/studies/apple-on-device-foundation-models) platform review rejected for core capability Apple's system model is a free and private routing floor, but its context and action-grounding limits make it unsuitable as the core capability dependency for a realistic tool catalog. ### [Castform RL fine-tune platform](https://posttrainllm.com/studies/castform-rl-finetune) product review partially adopted Composite rewards, trace-driven data loops, reasoning-depth classification, and explicit environment contracts make reinforcement-style fine-tuning more inspectable. ### [Cline and agent context hierarchy research](https://posttrainllm.com/studies/cline-context-hierarchy) agent architecture review partially adopted Structured-output enforcement and an explicit context hierarchy can reduce ambiguous agent actions and make tool boundaries easier to inspect. ### [Gigatoken](https://posttrainllm.com/studies/gigatoken) tokenization system review parked SIMD and cache-heavy bulk BPE can make first-pass tokenization much faster on large corpora, including Qwen tokenizers on Apple Silicon. ### [Needle 2](https://posttrainllm.com/studies/needle-2) model review and local smoke reviewed and rejected A compact call-only model combines top-five tool retrieval, constrained decoding, confidence escalation, bounded context, quantization, and a small packaged Mac artifact. ### [parakeet.wgsl](https://posttrainllm.com/studies/parakeet-wgsl) runtime review and local smoke validated proof Raw WebGPU plus SIMD-WASM can run Parakeet TDT browser inference with public kernels, a package format, cache behavior, a converter, and a benchmark harness. ### [Savante and Aryabhata](https://posttrainllm.com/studies/savante-aryabhata) industry case study study only Curated domain data, filtered reasoning traces, supervised fine-tuning, model merging, and verifiable-reward reinforcement learning combine in a JEE mathematics specialist. ### [Bonsai 2 27B](https://posttrainllm.com/studies/bonsai-2-27b) industry case study study only Ternary representation, packing overhead, activation transforms, kernel support, artifact size, runtime memory, and capability retention can rank a model differently depending on the deployment constraint. ### [QORL](https://posttrainllm.com/studies/qorl) industry case study learning queued Parameter-aware query optimization needs equal search budgets, fresh held-out measurement, SQL-equivalence checks, and a native-planner baseline before an offline plan search can claim value. ### [Inside vLLM](https://posttrainllm.com/studies/inside-vllm) industry case study study only Admission control, scheduling, paged KV allocation, prefill, decode, continuous batching, caching, and distributed execution explain why a serving engine behaves differently from a single-request model loop. ### [Splash model-specific Mac inference](https://posttrainllm.com/studies/splash-mac-inference) industry case study study only Shape-specific Metal kernels, a model-specific speculative draft, packed weights, cache reuse, and a startup memory plan trade generic model support for Mac-local serving performance. ### [Halo / White Circle](https://posttrainllm.com/studies/halo) training-framework review partially adopted HF-native training and portable checkpoints, independent expert/context/tensor/data parallelism, faithful multi-turn trajectories, operation-specific model support, and explicit resume/export/serving boundaries make the single-machine-to-cluster transition inspectable.