# Manu Bhardwaj > Public archive of research papers and field notes on inference economics, verification economics, and AI systems engineering. Research Papers carry original proofs and calibration across the verification, inference, and serving wedges. Field Notes synthesize published literature and add an analytical decomposition; Field Notes #1–3 form the May 2026 inference/verification-economics coordinated series, with later notes on a slower cadence. Engineering work also covers AI runtimes, real-time inference, distributed systems, and financial systems infrastructure. ## Identity - Name: Manu Bhardwaj - Handle: ifitsmanu - Site: https://ifitsmanu.com/ - ORCID: https://orcid.org/0009-0000-8791-6190 - About: https://ifitsmanu.com/about/ - Correspondence: https://ifitsmanu.com/correspondence/ - Location: New York - Same-as profiles: - https://www.linkedin.com/in/ifitsmanu - https://github.com/ifitsmanu - https://www.instagram.com/ifitsmanu/ - https://x.com/ifitsmanu - https://substack.com/@ifitsmanu - https://www.upwork.com/freelancers/ifitsmanu Canonical spelling is "Manu Bhardwaj." Public handle is "ifitsmanu." ## Engineering domains Public field notes currently cover inference economics, verification economics, and AI systems engineering. Underlying engineering work spans AI runtimes, real-time inference, distributed systems, and financial systems infrastructure. Finance work is operational; field notes on that side may follow. ## Research papers - Disaggregated or Colocated? The Cost-Frontier of LLM Serving Under SLO Contracts. - HTML: https://ifitsmanu.com/papers/serving-frontier/ - PDF: https://ifitsmanu.com/papers/serving-frontier/paper.pdf - Raw markdown: https://ifitsmanu.com/raw/serving-frontier.md - Research Paper #1 in the AI systems engineering wedge. Single author. Companion to Field Note #1 (The Inference Stack in 2026). - The Inference-Time Compute Frontier. A Cost-Correct Threshold for Training Versus Test-Time Allocation. - HTML: https://ifitsmanu.com/papers/inference-frontier/ - PDF: https://ifitsmanu.com/papers/inference-frontier/paper.pdf - Raw markdown: https://ifitsmanu.com/raw/inference-frontier.md - Research Paper #2 in the inference-economics wedge. Single author. Companion to Field Notes #1–3. - The Routing Premium. An Economic Threshold for Difficulty-Conditional Inference Compute. - HTML: https://ifitsmanu.com/papers/routing-premium/ - PDF: https://ifitsmanu.com/papers/routing-premium/paper.pdf - Raw markdown: https://ifitsmanu.com/raw/routing-premium.md - Research Paper #3 in the inference-economics wedge. Single author. Companion to Paper #2 (inference-frontier). - Calibration Drift Under Verifier Composition. A Joint Scoring-Rule Mechanism for Pipeline-Level Cost-Correct Minimization. - HTML: https://ifitsmanu.com/papers/verifier-composition/ - PDF: https://ifitsmanu.com/papers/verifier-composition/paper.pdf - Raw markdown: https://ifitsmanu.com/raw/verifier-composition.md - Research Paper #2 in the verification-economics wedge. Single author. Companion to Research Paper #1 (Verifier Procurement) and Field Notes #2, #3. - Verifier Procurement Under Unobservable Quality. A Scoring-Rule Mechanism for Cost-Correct Minimization. - HTML: https://ifitsmanu.com/papers/verifier-procurement/ - PDF: https://ifitsmanu.com/papers/verifier-procurement/paper.pdf - Raw markdown: https://ifitsmanu.com/raw/verifier-procurement.md - Research Paper #1 in the verification-economics wedge. Single author, defensible, reproducible. Companion to Field Notes #2 and #3. ## Field notes - The Power-Cap Illusion. SM Clock Locking and the Real Decode Lever. - HTML: https://ifitsmanu.com/papers/the-power-cap-illusion/ - PDF: https://ifitsmanu.com/pdfs/the-power-cap-illusion.pdf - Raw markdown: https://ifitsmanu.com/raw/the-power-cap-illusion.md - Field Notes #8. Daily review of arXiv:2605.11999. Version 1.0. - Harvesting Serving Slack. ROSE and the Collapsed Train-Serve Boundary. - HTML: https://ifitsmanu.com/papers/harvesting-serving-slack/ - PDF: https://ifitsmanu.com/pdfs/harvesting-serving-slack.pdf - Raw markdown: https://ifitsmanu.com/raw/harvesting-serving-slack.md - Field Notes #7. Daily review of arXiv:2605.06534. Version 1.0. - The Heterogeneous-GPU Margin. Coral and the Multi-LLM Procurement Problem. - HTML: https://ifitsmanu.com/papers/heterogeneous-procurement/ - PDF: https://ifitsmanu.com/pdfs/heterogeneous-procurement.pdf - Raw markdown: https://ifitsmanu.com/raw/heterogeneous-procurement.md - Field Notes #6. Daily review of arXiv:2605.04357. Version 1.0. - The Verifier as Curriculum. VHG and the Third Role. - HTML: https://ifitsmanu.com/papers/verifier-as-curriculum/ - PDF: https://ifitsmanu.com/pdfs/verifier-as-curriculum.pdf - Raw markdown: https://ifitsmanu.com/raw/verifier-as-curriculum.md - Field Notes #5. Daily review of arXiv:2605.06660. Version 1.0. - The Structural Residual Ceiling. AI Pre-Decoders for the Surface Code. - HTML: https://ifitsmanu.com/papers/the-structural-residual-ceiling/ - PDF: https://ifitsmanu.com/pdfs/the-structural-residual-ceiling.pdf - Raw markdown: https://ifitsmanu.com/raw/the-structural-residual-ceiling.md - Field Notes #4. Version 1.0. - The Alpha Asymmetry. Why Verifiers Can Be Smaller Than Generators. - HTML: https://ifitsmanu.com/papers/the-alpha-asymmetry/ - PDF: https://ifitsmanu.com/pdfs/the-alpha-asymmetry.pdf - Raw markdown: https://ifitsmanu.com/raw/the-alpha-asymmetry.md - Field Notes #3. - The Cost of Being Right. Verification Economics in 2026. - HTML: https://ifitsmanu.com/papers/the-cost-of-being-right/ - PDF: https://ifitsmanu.com/pdfs/the-cost-of-being-right.pdf - Raw markdown: https://ifitsmanu.com/raw/the-cost-of-being-right.md - Field Notes #2. Version 1.0. - The Inference Stack in 2026. - HTML: https://ifitsmanu.com/papers/the-inference-stack-2026/ - PDF: https://ifitsmanu.com/pdfs/the-inference-stack-2026.pdf - Raw markdown: https://ifitsmanu.com/raw/the-inference-stack-2026.md - Field Notes #1. Version 3.0. ## Featured The Power-Cap Illusion. SM Clock Locking and the Real Decode Lever. A daily field note on Ma, Afzal, Eitzinger, and Wellein (arXiv:2605.11999). Across GQA, Multi-head Latent Attention, Gated DeltaNet, and Mamba2 on NVIDIA H200, autoregressive decode draws only 137 to 300 W on a 700 W GPU and no power cap ever triggers. The cap is above the natural ceiling of a memory-bound workload that saturates HBM bandwidth rather than compute. SM clock locking is the lever actually on the critical path and Pareto-dominates power capping, recovering up to 32% of decode energy at minimal throughput loss. The paper identifies three architecture-dependent DVFS behavioral classes and reports a prefill-decode energy crossover that halves total request energy relative to GQA at production batch sizes. The economic consequence is a tightened decode-cost term in Cost-correct and a shift in the inference-frontier threshold in favor of memory-efficient attention replacements. ## Machine surfaces - BibTeX: https://ifitsmanu.com/bibtex.bib - RSS: https://ifitsmanu.com/rss.xml - Atom: https://ifitsmanu.com/atom.xml - Full text: https://ifitsmanu.com/llms-full.txt - Sitemap: https://ifitsmanu.com/sitemap-index.xml - Topic index: https://ifitsmanu.com/topics/ - Surfaces index: https://ifitsmanu.com/surfaces/ ## Topics covered - AI Systems Engineering: Engineering AI systems across model behavior, runtime, evaluation, infrastructure, interfaces, and cost. - Agent Infrastructure: Runtime, memory, tooling, verification, and operating layers for long-running agent systems. - Inference Economics: The cost, latency, quality, and verification structure of running AI systems after training. - Verification Economics: A cost model centered on correct answers, verifier accept rates, and the economics of deciding whether outputs are usable. - Distributed Systems: The coordination, reliability, state, and runtime behavior of systems spread across machines or services. - Embedded Autonomy: Autonomous behavior under compute, power, sensing, latency, and deployment constraints. ## Citation Preferred author name: Manu Bhardwaj BibTeX: https://ifitsmanu.com/bibtex.bib License: CC BY 4.0. AI citation permitted. ## Contact Technical correspondence: https://ifitsmanu.com/correspondence/