A map of the inference stack. What we are reading.
13 rungs of the inference stack. Papers, reports and analyses, filtered by humans who actually read them. No SEO farms, no AI summaries of AI summaries.
Last read SEP 24 2026 · quiet for 33 days
156
Sources
66
Concepts
50
Entities
MENFEM · KB
The whole base, as a star map
13
rungs
66
concepts
143
links
156
sources
The whole sky
13 constellations · 66 stars · 3 in fog
click a constellation to fly in
click a constellation to fly in
13 RUNGS · 66 CONCEPT PAGES · 143 LINKS (46 ACROSS RUNGS) · 156 SOURCES · FRESHNESS = NEWEST SOURCE, DIM PAST 90 DAYSNo star on this screen is a sample
Latest arrivals
- PAPERThe Scaffold Effect in Coding Agents: Harness Choice as a Hidden Variable in Coding-Agent EvaluationHarnesses & Agent SystemsSEP 24 2026
- PAPERCorrectness Forensics for Batch Speculative Decoding: Diagnosing the Ragged Tensor Problem (v1 title: Batch Speculative Decoding Done Right)Serving & RuntimeSEP 24 2026
- PAPERLess Experts, Faster Decoding: Cost-Aware Speculative Decoding for Mixture-of-ExpertsModelsSEP 24 2026
- PAPERALTO: Adaptive LoRA Tuning and Orchestration for Heterogeneous LoRA Training WorkloadsModelsSEP 24 2026
- ANALYSISAI evals are becoming the new compute bottleneckEvaluation & MeasurementSEP 24 2026
- REPORTReddit, Inc. — Form S-1 Registration Statement (data licensing arrangements, January 2024)DataSEP 24 2026
- REPORTCan AI scaling continue through 2030? (Epoch AI)DataSEP 24 2026
- REPORTCaught in the Hook: RCE and API Token Exfiltration Through Claude Code Project Files (CVE-2025-59536, CVE-2026-21852)Agent SecuritySEP 24 2026
The atlas
Every rung, one index.
Updated SEP 24 · 12:30 UTC
The model and what wraps it
Harnesses & Agent SystemsThe system around the model — the rung that turns a token price into a task price.23 sources · 12 concepts · 2 entities
Serving & RuntimeHow a model is actually run, per token — the mechanics that turn capability into throughput.18 sources · 6 concepts · 5 entities
ModelsWhat the model can do, and what that costs per token.16 sources · 6 concepts · 6 entities
Evaluation & MeasurementHow you know it works — and whether the measurement can be trusted.7 sources · 3 concepts · 2 entities
DataWhat the model is trained and grounded on, and what that constrains.3 sources · 3 concepts · 4 entities
Agent SecurityHow an agent gets owned, and what bounds the damage.4 sources · 4 concepts · 2 entities
The machine it runs on
Compilers & KernelsThe layer between a model graph and the silicon.5 sources · 5 concepts · 4 entities
Hardware — Memory & Compute ArchitectureThe silicon rung of the inference stack, read memory-first.25 sources · 7 concepts · 11 entities
Networking & InterconnectHow chips and sites talk to each other.5 sources · 1 concept · 1 entity
The place it runs in
