Searcharxiv⌕ Search

arXiv subjects

Bojie Li

Publications and source records attributed to Bojie Li.

At least 19 recordsLinked to original sources
arXiv2026

The Model Knows Your Project, Not You: Measuring Recognition in LLMs with NameRank

What a frontier model recalls about a person or tool from its own weights -- before any retrieval step -- often shapes the first description a human sees, making that parametric corpus presence a measurement problem. Citations explain about a third of whether a model recognizes a researcher; we target the residual and build NameRank, a [0,1] recognition score: each of 4,685 entities in 54 cohorts is probed with one open-ended question across 36 models, and an independent judge returns a binary verdict against a curated gold -- did the model state a specific, non-guessable fact about this exact entity? -- so hallucination, context echo, and guesses earn nothing. Synthetic-null entities hold the floor near zero, and verdicts track the entity, not the model. One thesis organizes the findings: recognition is paid to named, indexable artifacts, not to credentials or titles. Every Olympic-style credential sits below a working-researcher baseline, because no named artifact ships with the medal, yet the ranking inverts at the marquee tier, where Nobel, Turing, and Fields laureates saturate the panel. For independent creators the tool out-ranks its maker, and the credential that does propagate is a named method or awarded paper. Being one of many named contributors to a celebrated artifact, by contrast, earns almost nothing -- the authors listed on a flagship model report or system card sit near the recognition floor -- because recognition attaches to the artifact's own distinctive name, not to the roster behind it. No bibliometric predicts recognition well; top-density institutions out-recognize peers at matched citations; and on 258 news events recognition loads on peak salience, not persistence. A self-report probe shows introspection reads a corpus prior, not its own knowledge.

cs.AI↗
arXiv2026

Interaction Scaling: Grounding the Third Axis of Test-Time Compute

There are two standard ways to spend more compute at test time: let a model reason longer, or sample more attempts and keep one. Both share a hidden limit: they are internal. Every extra token comes from the same frozen weights and the same prompt, so neither can tell the model anything it does not already know. We study a third way, interaction: the model proposes an artifact, an external instrument observes how it actually behaves, and the model revises. Each cycle imports a real observation, so interaction breaks through the ceiling the other two hit. We argue that a single variable governs this third axis, grounding, and that it must hold on both sides of the loop. The feedback that drives revision must come from an instrument that actually observes the flaw, and so must the metric that scores the result. On hard coding tasks at a fixed token budget, reasoning-only and best-of-N sampling both plateau (the latter even when an oracle picks the best sample), while every interaction strategy keeps improving; our proposer-reviewer harness reaches a perfect 100% pass rate with no run-to-run variance, and the gain holds across three model families. On rendered visual artifacts, the usual judge (a vision-language model, or VLM, reading a screenshot) rates 14 of 15 visibly broken figures "perfect," because the screenshot hides the flaws before the judge can see them. A tool that measures the real layout instead shows the loop removing 40-74% of defects across four modalities; and that same VLM, used as the reviewer, makes slide layouts worse where the measuring tool repairs them. Interaction scaling is real and distinct from reasoning and sampling, but only visible when both the feedback and the metric are grounded.

cs.AI↗
arXiv2026

The Trade-off Was in the Labels: Causal Supervision for Turn-Aware Streaming ASR

A voice agent must decide, moment to moment, whether the user has finished; silence rarely settles it: a caller reading a phone number pauses mid-digits, a one-word "Stop!" ends a turn, a long question carries pauses longer than real turn-gaps. A voice-activity detector plus a silence timeout (the deployed default) cannot separate these, because within-turn pauses routinely exceed between-turn gaps; what distinguishes them is whether the words so far form a complete thought: what a recognizer computes to produce a transcript. We present the first open training recipe and benchmark for turn-aware streaming ASR: a small LoRA adapter on Qwen3-ASR-0.6B, trained in hours on one GPU, that transcribes, detects end-of-turn from meaning and silence, handles dictation, and grounds transcription in context. On a deployment-matched benchmark it reaches 0.97 boundary recall at 0.39 s median latency with 0.3 false fires per speech-minute, replicated on a fresh test set; no silence timeout reaches this point. The recipe rests on one principle: every streaming-decision label must be computable from input up to the decision point. Offline corpora violate it, encoding the future; such clairvoyant labels manufactured oscillation and a phantom recall-versus-precision trade-off, exposed when one appended second of silence raised a "broken" model's end-of-turn recall from 0.10 to 1.00. The same leak recurred with context: an always-matching biasing prefix became a copied shortcut (40% intrusion), and counterfactuals disagreeing with the audio cut this to 0.8% while keeping most of a +28.9 pp entity-recall benefit.

eess.AS↗
arXiv2026

RLVP: Penalize the Path, Reward the Outcome

Agents acting on our behalf in the real world (e.g. placing phone calls) must learn online from costly, often irreversible interactions rather than cheap simulator steps. Two things follow. First, deployability depends on the path, not only the outcome. An agent must respect outcome-neutral constraints such as not repeatedly calling an unresponsive user, respecting business hours, or completing required authentication constraints that outcome-based rewards cannot express, since violating them frequently improves apparent success. Second, because each interaction is expensive, the agent must learn efficiently from very few examples. Reinforcement learning from verifiable rewards (RLVR) is blind to both challenges: it optimizes solely on the outcome and wastes expensive rollouts on all-fail groups where group-relative advantage collapses to zero. Attempts to densify supervision by rewarding progress target the hard-to-verify direction. In contrast, real agentic environments can cheaply detect bad moves. Since group-relative advantage is equivalent to within-group variance, a dense signal helps only when it supplies variance the outcome lacks. A verifiable penalty on the path meets this condition reliably, while a progress potential helps only where partial progress is reachable. The resulting recipe "penalize the path, reward the outcome" achieves high task success with near-zero violations, where outcome-only training violates constraints on nearly every episode. We provide four design rules for effective penalties, including avoidance of the inaction trap that arises when a penalty is used in isolation.

cs.LG↗
arXiv2026

Parametric Multimodal User Memory: Storing What Captions Cannot Carry

A personalized agent needs a user memory: a persistent model of who its user is. Today it is almost always text -- transcripts and captions retrieved by similarity. This serves the captionable half of a person ("my cat is named Bibi"), but discards the perceptual half no caption can hold: how a voice sounds, how a face reads across age and lighting, how tired someone sounds. We measure this loss across five modalities: a strong caption-based re-identifier recovers as little as 0.11 of a dedicated encoder's recall, collapsing toward chance on non-nameable signals. We instead ground perceptual memory in the model, decomposing recall into two subproblems: a vision-language model grounds the referent in context (what and where), and a dedicated encoder extracts an identity key (who), stored as one inline token read by attention at generation with no external round-trip. Neither suffices alone -- the VLM identifies cross-age faces at only 0.54 recall where a face encoder reaches 0.81, and an ungrounded encoder recognizes a two-person-scene referent at 0.05 -- yet together they reach correct-region oracle (0.96), generalizing to multi-speaker audio and video. The recognition core is training-free: it reproduces the encoder's recall on any frozen model at O(1) registration cost. On PerceptMem (12 domains, 1,080 tasks) perceptual identity is capacity-limited while exact facts are binding-limited: identity belongs in a parametric bank, facts in a text store. The two memories compose cleanly: an agent with both can remember not only what its user said, but also what they are like.

cs.CL↗
arXiv2026

Fine-Grained Computation Offload for Off-the-Shelf Servers in Tens of Lines

Hardware accelerators now sit on the critical path of online serving. GPUs, FPGAs, and increasingly remote services such as hardware security modules, post-quantum KEMs, and inference servers. For fine-grained offloads (microseconds to a few milliseconds) the classic responses to the resulting stall both fail: a context switch costs as much as the offload, and a busy-wait burns the core. Overlapping the offload with other requests is the fix, and prior systems obtain it by adding concurrency: an async-framework rewrite, a new runtime or dataplane OS, or a hand-tuned point integration. We observe that the concurrency already exists: serving concurrent requests is suspending and resuming them, so every server ships the machinery overlap needs. Overlap is then a routing problem, not a rewrite problem: submit the offload to an executor, suspend the request with the server's own deferred-response primitive, resume it on completion. Across ten off-the-shelf servers spanning every production concurrency model, this recipe takes 22-138 lines added, at most one modified, and recovers 1.2-5.4x on real hardware; the server's concurrency model and the offload's weight predict both numbers in advance, and the win is bounded by device throughput and the server's own overlap capacity. At the limit, an LD_PRELOAD fiber runtime injects the reroute into an unmodified thread-per-connection binary (17.3x) within a characterized envelope. Rerouting suspends run-to-completion atomicity; a measured taxonomy confines the hazard to unlocked shared aggregates, and a transparent page-protection detector guards exactly those, validated on stock Redis.

cs.DC↗
arXiv2026

Metronome: Bound the Cache, Keep the Beat for Real-Time Interaction Model Serving

Real-time interaction models -- Moshi, MiniCPM-o, Qwen-Omni -- turn serving into a periodic real-time task: on every frame a session ingests streaming audio and must respond by a recurring wall-clock deadline, while its KV cache grows monotonically and stays pinned for the whole conversation. This regime hides a dangerous failure mode. On a real full-duplex stack, sustained load does not degrade serving gracefully: it falls off a cliff, jumping in one step from milliseconds per frame to a stalled engine when accumulated session state exhausts the KV pool. The collapse is metastable -- identical five-minute runs collapse or survive on run-to-run variance -- and silent: latency and deadline-miss metrics read healthy throughout. We show one move restores both stability and observability: bound each session's resident state, and latency starts telling the truth. Metronome's in-engine KV window eliminates the collapse (0/20 vs. 14/20 runs across two batches) and turns per-frame latency into a monotone load signal, on which an online admission controller discovers the schedulable concurrency; without the window, the identical controller over-admits into the wall. A first-order model predicts the collapse time within a few percent on the headline model, and a quality probe validates the bound's design by ablation: the window alone is quality-free in turn-based decoding, and its few pinned attention-sink tokens are what keep free-running generation healthy. Everything is measured end-to-end on real audio, across four interaction models on one GPU.

cs.SD↗
arXiv2026

OneBarrier: What a Network Must Provide for Transparent Fault Tolerance to Be Free

Transparent fault tolerance -- making an unmodified server binary survive crashes -- has been pursued for four decades without reaching production. Every attempt paid three costs on the critical path: recording message arrival order for replay, coordinating a consistent snapshot, and holding each reply until the state that produced it was durable. This paper argues the costs are not intrinsic: they are the price of a network that guarantees neither order nor delivery. We state four conditions under which all three vanish. Three concern the network: Order (messages are delivered in one global sequence), Barrier (delivery is confirmed by a commit barrier), and Durability (each message is replicated to backups before its barrier completes). The fourth, Determinism, falls to the host: a user-space shim closes it for unmodified binaries at 2-10% overhead -- virtual time, virtualized randomness, and share-nothing sharding in place of thread scheduling. OneBarrier realizes all four conditions over an in-network total-order fabric (1Pipe) with microsecond round trips. Fifteen unmodified applications -- including Redis, Memcached, Nginx, Node.js, and a multi-process PostgreSQL -- recover byte-identically, and crash injection confirms linearizable, exactly-once histories; the core protocols are machine-checked in TLA+. A durable write placed inside the barrier adds 4.6 microseconds to a request; the same write placed after it adds three milliseconds. On a network that meets the conditions, fault tolerance is a property, not a tax.

cs.NI↗
arXiv2026

Whose Side Is Your Agent On? Multi-Party Principal Loyalty in LLM Agents

A rapidly growing class of LLM agents is multi-party: the agent acts for a principal (who briefs it, sends follow-ups, and receives results) while also conversing in a separate channel with a counterparty whose interests may diverge (negotiating with a vendor, screening inbound requests, or mediating between employees). Here "help whoever you are talking to" is the wrong objective. The agent must stay loyal to the principal it represents without over-refusing the principal's own cooperative asks. We study this multi-party loyalty problem and contribute a measurement instrument, two mechanisms, and a structural lesson. PrincipalBench is a 75-item multi-turn benchmark with leak probes, dual judges, and an integrity-audit gate. Across 13 frontier subjects it exposes a sharp split (<=20% vs. 53.6-75.3% harm) invisible to single-turn safety evaluations: a selective cluster that declines adversarial probes while still following the principal's legitimate requests, and an over-refusing cluster that refuses broadly. (M1) A prompt-time loyalty scaffold (a fixed system prompt of seven prioritized rules, open-coded from 50+ failure trajectories) holds Claude-Sonnet to 19.4% harm and all nine selective subjects to <=20%. (M2) A per-token-KL distillation recipe transfers a prompted Qwen3-32B teacher into 8B Qwen3 and Llama-3.1 students, the strongest open-weight recipe we measure. (Lesson) Both mechanisms only move along a common leak/over-refusal trade-off rather than crossing it: improving one axis costs the other, and the jointly favorable outcome stays out of reach.

cs.AI↗
arXiv2026

Agent-Computer Observation Interfaces Enable Dynamic Computer Use

SWE-agent established the action interface as an underexplored design axis for software-engineering agents; we make the analogous case for the observation interface in computer-use (CU) agents. Current CU agents, closed and open-source alike, tie observation to action--one screenshot every 3-5 s, no audio--leaving them blind and deaf between screenshots to video, animations, transient UI events, meetings, and spoken instructions. We introduce the Agent-Computer Observation Interface (AOI), a model-agnostic perception layer that decouples continuous, adaptive observation from discrete actions through three gated components: inter-step keyframe capture, volume-gated audio transcription, and CU-model-generated visual narration that persists as text. Each produces almost nothing on static, silent content, reducing to the standard loop without degrading it. On DynaCU-Bench (100 dynamic browser tasks plus a 50-task static control), CU models from 7B to frontier scale gain +17 to +48 pp over their screenshot baselines with zero retraining, turning tasks that are near-impossible from periodic screenshots into largely solved ones. The gap is starkest on audio: on a spoken-content subset AOI agents solve every task, whereas streaming voice models hear accurately but cannot act on what they hear without the scaffold. The decomposition is as informative as the headline gain: keyframe selection turns out not to matter--the value comes from narrating captured frames into persistent text--and the interface is not a fixed bundle, since on a newer model (Gemini 3 Flash) the keyframe stream actively regresses through image-token dilution, so its components must be selected per model rather than shipped as one configuration.

cs.AI↗
arXiv2026

The Latent Bridge: A Continuous Slow-Fast Channel for Real-Time Game Agents

A real-time agent for general computer use - with games as the most demanding case - must act within tens of milliseconds while still planning over seconds. These two regimes sit at opposite ends of the latency-quality tradeoff. A reasoning VLM (Qwen3-VL-8B-Thinking) deliberates effectively but requires ~1.5 s per response - far too slow for a 15 Hz control loop. In contrast, a reactive VLM (MiniCPM-o 4.5) acts in milliseconds but underperforms on planning-heavy tasks. We couple two frozen models of matched scale (9B reactive, 8B reasoning), leaving the communication channel as the sole trainable component. The standard coupling is a Text Bridge (T): the slow model writes a suffix the fast model reads. We introduce a learned continuous Latent Bridge (L) that projects the slow model's residuals into the fast model's input-embedding space in a LLaVA-style manner, avoiding any text round-trip; both are compared against Fast-Only (F). On 7 Atari games and a driving domain (MetaDrive), tuning the action decoder per channel on held-out seeds, the Latent Bridge matches or beats the Text Bridge in every domain: it significantly improves two games (MsPacman +57%, RoadRunner +28%) and is a safe drop-in elsewhere. Combining both channels interferes destructively (RoadRunner -96%), so only one should be used. The benefit is highly predictable: the bridge helps if and only if slow reasoning already beats fast reaction (T > F) - the Latent and Text gains over Fast-Only move together at r=0.93. MetaDrive is the controlled negative, where the Latent Bridge is demonstrably inert because the Text Bridge adds no value. We release replay recordings and reproducible pipelines.

cs.AI↗
arXiv2026

User as Engram: Internalizing Per-User Memory as Local Parametric Edits

Personal memory in a language model is two problems: content and reasoning skill. The brain keeps the two apart (a sparse, local engram in the hippocampus for each episode, a slow neocortex for the shared skills that interpret it), so a new fact need not overwrite everything else. Most personalization today keeps a user's facts outside the weights, in a natural-language memory file or a retrieval index. When facts are written into the model instead, the standard recipe is the per-user LoRA adapter, which does the opposite of the brain, folding content and skill into one global weight delta. Writing a user's facts as a LoRA contaminates text unrelated to them; writing the same facts as local Engram rows leaves it mathematically untouched, resulting in a roughly 33,000x smaller memory footprint. We therefore propose User as Engram: store a user's content as surgical edits to the hash-keyed memory table of an Engram model, and carry the reasoning skill in one shared adapter. This layered design matches per-user LoRA's direct recall while delivering 5.6x higher indirect-reasoning accuracy on average, and never makes a single user worse at reasoning than the untouched base. The edit is a glass box: writing a fact switches on its lookup at exactly the trigger, adds the value the answer needs, leaves every other position unchanged to the last bit, and fails if written into the wrong layer. Because different users' facts land in disjoint hash slots, their edits compose: many users live in one shared table at once, stacking additively and losslessly, where a per-user LoRA, a single global weight delta, admits only one. Upon retrieval, a per-user Engram table does not grow with the population the retriever must search, so past ~100 facts it overtakes a retrieval pipeline on a 2.5x larger model.

cs.AI↗
arXiv2026

PreAct: Computer-Using Agents that Get Faster on Repeated Tasks

Computer-using agents drive real software through the screen -- clicking and typing -- but they solve every task from scratch: asked to repeat a task, an agent re-reads the screen, re-reasons every tap, and pays the full cost again. We present PreAct, which lets such an agent get faster on tasks it has done before. The first time it succeeds, PreAct compiles the run into a small state-machine program-states that check the screen, transitions that act-and on later runs replays it directly instead of invoking the agent 8.5-13x faster, with no per-step language-model calls. Replay is not blind: at each step PreAct checks that the screen matches what the program expects before acting, and hands control back to the agent the moment something is off. PreAct applies the same discipline when deciding what to keep: a freshly compiled program enters the store only if, re-run from a clean state, an independent evaluator confirms it solved the task-catching programs that replay to their last step yet leave the task undone. Across a mobile, a desktop, and a web benchmark, this store-time check separates repeated runs that improve from ones that degrade as faulty programs accumulate, worth 1.75-2.6 tasks per benchmark, the same direction on all three; a fallback that explores afresh when no program fits brings PreAct level with a strong record-and-replay baseline. We also report what did not matter: prompt wording, runtime guardrails, and whether a language model or a plain embedding retriever selects which program to reuse.

cs.AI↗
arXiv2026

User as Code: Executable Memory for Personalized Agents

A personalized AI agent needs a user memory: a persistent model of who the user is, built across many conversations and consulted on each new one. Today this memory is almost always stored as unstructured text, a knowledge graph, or a flat store of facts, and consulted by retrieval -- fetching the entries most similar to the current request. Such "bag-of-facts" memory recalls individual facts well, but because storing a fact and acting on it are separate steps, it struggles to resolve contradictions, aggregate over many records, or enforce rules. We argue that user memory should instead be executable. We introduce User as Code (UaC), a paradigm in which an agent's model of a user is a living software project: typed Python objects hold the user's state and ordinary Python functions encode the rules that govern it, so representing and reasoning about the user happen in one medium an interpreter can run. The enabling mechanism is a two-phase pipeline: an append-only log that never discards a fact, periodically checkpointed into typed code. This changes what memory can do. On standard long-term conversation benchmarks, UaC matches both a full-context upper bound and the strongest prior memory systems on recall (78.8% on LOCOMO). Its advantage emerges where representation matters most. On aggregate questions over a user's history -- "how many international trips did I take last year?" -- retrieval-based memory collapses (6-43%) while UaC stays near-perfect (99%), because the answer is a one-line computation over typed state rather than a search over text. And because its rules execute deterministically whenever the state changes, UaC can surface unsolicited, safety-critical alerts -- such as a newly prescribed drug that conflicts with an allergy recorded months earlier -- a capability query-driven memory cannot provide.

cs.AI↗
arXiv2026

Models Take Notes at Prefill: KV Cache Can Be Editable and Composable

Prefix caching reuses prefill only across an exactly shared prefix, so one changed field invalidates the entire downstream cache. Yet overwriting the field's own key/value vectors and reusing the rest leaves the model acting on the old value. The reason, established causally across four model families: at prefill the model has already written the field-conditioned conclusion onto downstream notes; the field's own key/value drives under 1% of the decision. Read as a notebook of memoized conclusions, two capabilities follow. (1) It is editable. A salient erratum amends the notes; and with chain-of-thought, editing the field alone recovers the decision (1.00 at 8B, ~1% compute), while without CoT it is ignored. (2) It is composable. The notes are position-portable, so a precompiled skill can be RoPE-repositioned and spliced into any context, indistinguishable from full recompute (logit cosine 0.90-0.999, twelve models) at O(L) rather than O(L^2) time-to-first-token. A unified edit+compose agent stays decision-identical to recompute at up to 14.9x lower latency. The approach applies to any per-token attention KV cache, validated across scale, quantization, Mixture-of-Experts, and multimodal caches, and extends to several attention variants through small adapters. Because the erratum is append-only, it composes with production prefix caching: in an online vLLM benchmark it keeps the prefix cache-aligned (98.5% hit-rate), cutting p90 time-to-first-token by 53-398x.

cs.LG↗
arXiv2026

Tiara: A Programmable Line-Rate ISA for Remote Memory Access

RDMA one-sided verbs are the natural primitive for memory disaggregation, but they require the client to supply the exact remote address. The 1-RTT performance breaks down when the target address depends on data that must first be read from remote memory, a pattern we call the Indirection Wall. Indirection is pervasive: graph traversals follow pointers hop by hop, address translation walks multi-level page tables, distributed coordination requires conditional multi-host logic, and disaggregated LLM inference must resolve paged KV caches through block-table lookups. Each level of indirection costs one sequentially dependent network round-trip, yet offloading to existing RDMA NICs either consumes remote CPU cycles or has limited throughput. We present Tiara, a compact, statically verifiable instruction set that executes on the memory-side NIC. Tiara operators are pre-registered programs, analogous to eBPF programs in the kernel, that resolve indirection locally, collapsing multi-RTT dependent chains into a single round-trip. On an FPGA-based prototype, Tiara reduces 10-hop graph-traversal latency by 2.85x over one-sided RDMA while sustaining 3.4x higher throughput, cuts page-table walk latency by 62%, reduces uncontended distributed-lock latency by 2.9x, achieves 2.8x throughput for disaggregated PagedAttention at 8 KB blocks, and 1.88x MoE expert-gather latency at 32 experts.

cs.AR↗
arXiv2026

OpenURMA: A Clean-Room Open Implementation of the Unified Bus Protocol

Modern datacenter RDMA is bottlenecked at the network interface, not the wire. A NIC running RoCE or InfiniBand holds per-connection state for every (application, remote-endpoint) pair - hundreds of megabytes at 1024-application fanout - and pays a four-traversal PCIe round trip on a 64-byte operation, inflating latency an order of magnitude beyond the wire. Both follow from the Queue Pair over PCIe abstraction RDMA inherits from InfiniBand. Huawei's Unified Bus (UB), a public 2025 specification, changes the abstraction: it decouples per-application endpoint state from per-host transport state so connection context grows additively, exposes ordering as opt-in, and reaches remote memory through native CPU load/store to an on-chip-bus controller. UB ships in Huawei's closed Ascend 950 silicon. OpenURMA is the first clean-room open implementation of UB's transport and transaction layers, realised at three tiers - synthesisable RTL on Alveo U50, a cycle-level two-node SystemC simulator, and a gem5 full-system scaffold - each with a matched OpenRoCE (RoCEv2 RC) baseline. The contribution is the implementation, harness, and controlled comparison closed silicon does not admit. On the canonical 64-byte remote fetch - LOAD on UB-spec Sec.8.3, READ on RoCEv2 RC - UB's load/store path delivers ~500 ns end-to-end, 4.37x below the matched baseline (2186 ns), sustains 2.80x higher throughput, and fits in ~14% of a U50's LUTs.

cs.AI↗
arXiv2026

Incompressible Knowledge Probes: Estimating Black-Box LLM Parameter Counts via Factual Capacity

Closed-source frontier labs do not disclose parameter counts. Storing F facts requires at least F/(bits per parameter) weights, so factual recall lower-bounds parameter count--an intrinsic, serving-independent signal, though (as we show) a coarse one. We introduce Incompressible Knowledge Probes (IKPs), a benchmark of 1,400 factual questions spanning 7 tiers of obscurity, designed to isolate knowledge that cannot be derived by reasoning or compressed by architectural improvements. We score with no hallucination penalty (lambda = 0: IKP accuracy is simply the fraction of probed facts answered correctly), which removes both the penalty hyperparameter and the per-tier flooring choice; a full lambda x flooring ablation shows the calibration is robust across scoring choices while individual estimates are not, motivating the no-penalty default. We calibrate a log-linear mapping from IKP accuracy to parameter count on 93 open-weight models (135M-1,600B) spanning 19 vendors, achieving R^2 = 0.910; leave-one-out cross-validation confirms generalization (median fold error 1.48x, 72% within 2x, 86% within 3x). The instrument is deliberately coarse--its 90% prediction interval spans ~3x in either direction, wider than inference economics--so IKP recovers order-of-magnitude effective capacity and relative rankings, not precise parameter counts. For Mixture-of-Experts models, total parameters predict knowledge (R^2 = 0.67) better than active parameters (R^2 = 0.41). We evaluate 201 models from 27 vendors on a curated probe set (1,311 of 1,400 probes surviving name-collision and label-ambiguity filters) and report effective knowledge capacity for all major proprietary frontier models as prediction bands rather than point estimates; for heavily safety-tuned models these are lower bounds, since refusal policy can suppress tens of percentage points of otherwise-answerable capacity.

cs.LG↗