Outputs are lossy
A final string collapses a long, structured computation into one token. Two runs can agree on the answer and think in completely different ways.
Reasoning telemetry & control · Fudan University + Shanghai Innovation Institute
The model’s white box is the next data layer.
Beyond final outputs, we read a model’s internal signals — activations, MoE routing, neurons — and its run-level process signals, and turn them into measurable, controllable data across inference, evaluation, and model construction.
CckFdu·white-box core + process telemetry
Read the inside. Build the moat.
Led by Kang Chen
Ph.D. researcher at Fudan University (advised by Prof. Yixin Cao) · first or co-first author on all six studies below · prior algorithm-research internships at SenseTime, Baidu ERNIE, and Alibaba Taotian.
The insight
Outputs are the black box. The real signal is inside — and along the way.
Almost all of AI reads a model by what it finally says. But the richest, least-tapped signals are how it computes and how a run unfolds: which neurons fire, which experts route — and where a trajectory commits, traps, or recovers. Both are measurable beyond the final answer. Six studies turn them into a substrate, united by one question — not by one identical pipeline.
White-box core · internal signals Activations · MoE routing · role-conditioned neurons NAD · RAD · SliceGraph · ARM
Process telemetry · run-level signals Partial traces · agent action–observation · trap & repair events TAAR · TraceGraph
One question: beyond the final answer, which signals make reasoning measurable and controllable?
Measurable — and mostly label-free
Activation keys and routing traces need no gold answers; the run-level layer adds only a light diagnostic probe. The signal scales with compute, not annotation.
One substrate, staged products
The same signals feed a control loop first, then a process-evaluation layer, then cheaper model construction — one core, entered where the ROI is most direct.
The engine
One substrate. A control loop, an evaluation layer, and model ops.
The control loop · the concept
How the signal becomes control.
The shape the research points to: read a model’s internal and run-level signals live, score each sampled trajectory, and act — continue, stop, restart, or select — so the same compute buys more correct answers.
Model runtime → Activation / routing / trace hooks → White-box telemetry → Agreement · trap · route-basin → Stop · restart · select → Cost per correct task ↓
16 sampled runs · live agreement example values
Actions this step
- stop 3 low-agreement runs
- restart 1 trapped run, before the trap
- select the densest route basin
From wedge to platform
Start where the ROI is most direct. Then compound.
- 1Inference controlstop / restart / select — cost per correct task
- 2Evaluation & diagnosticssee the process, not just the score
- 3Model construction & repairbuild & fix models without retraining
Each run of the control layer compounds into process data that feeds evaluation, and evaluation guides construction. Below: the research that validates each step — a direction, not a shipped product.
Line 1 · Inference control (the first wedge)
Turn compute into correctness — not just more tokens.
Internal agreement and control decide when to trust a run, when to stop, and when to restart — reading no answer strings. The same signal makes test-time compute pay off instead of piling up.
Up to 98%fewer generated tokens (NAD)NAD · AIME · early-stopping · research result
+4.6pton an 8B base model (TAAR)TAAR · base frozen; diagnostic policy trained separately
32tokens to a correctness signal (NAD)NAD · first early signal
Proven in:
Line 2 · Evaluation & diagnostics
See how models think — not just what they answer.
Raw signals become structured, process-level data: atlases of reasoning routes, decision landscapes, trap maps. A new evaluation and diagnostics layer that a leaderboard number can’t give — built from signals no one else is collecting.
60,622reasoning trajectories mapped (SliceGraph)SliceGraph · 954 cells · 6 models
85.5%same-answer runs take different routes (SliceGraph)SliceGraph · blinded-validated atlas
+3.1ppSWE-bench Verified, per provider (TraceGraph)TraceGraph · +3.8 pp on common-fired instances
Proven in:
Line 3 · Model construction & repair
Build and repair models — without retraining.
Reading the white box lets us compose and fix models cheaply: pick and graft the right internal circuits instead of running gradient descent. Better checkpoints, no training run.
Beatsthe oracle expert selector, training-free (ARM)ARM · Qwen3-8B & Qwen2.5-7B pools
2model families, one merge recipe (ARM)ARM · cross-family replication
0gradient steps — pure activation surgery (ARM)ARM · training-free merge
Proven in:
Proof
Not a pitch deck — a research portfolio.
Six papers on arXiv, each with real numbers, blinded validation, and reproducible methods. Every claim on this page links to the paper that proves it — read them yourself.
Peer-reviewed: NAD · ICML 2026 Spotlight TAAR · ACL 2026 Findings
NADwhite-box coreICML 2026 · Spotlight
Predicts correctness from internal-activation consensus, reading no answer strings.
up to 98% fewer generated tokens (AIME early-stop) · signal by 32 tokens Read the paper → RADwhite-box coreSelects the answer by MoE-routing agreement — a signal, not a verifier; works where answer-string voting is undefined.
on par where voting works · available where voting breaks Read the paper → TAARprocess telemetryACL 2026 · FindingsDiagnoses “thinking traps” and restarts before them — base weights frozen; a lightweight diagnostic policy is trained separately.
+4.6 pts on an 8B base model · base frozen Read the paper → SliceGraphwhite-box coreAn activation-key atlas of many CoT runs — same answer, different reasoning routes.
60,622 trajectories · 85.5% multi-route Read the paper → TraceGraphprocess telemetryPools agent runs into one decision landscape; its trap map lifts SWE-bench Verified resolved rate.
+3.1 pp per provider (+3.8 on common-fired) Read the paper → ARMwhite-box coreTraining-free agent merging via answer-free activation overlap — build models without retraining.
beats the oracle expert selector · training-free Read the paper →6papers · white-box core + process telemetry
60k+internal-signal trajectories (the core is label-free)
Answer-freeat inference — the selectors read no answer strings
Momentum: 6 papers · Sep 2025 → Jun 2026
Why now · the moat
The signal compounds — labels don’t scale, runs do.
A compounding data layer
Internal-signal and process datasets are generated by the act of thinking, not bought or annotated. Instrument once across architectures, and every run adds cross-model, cross-task, cross-failure data.
Engineering across architectures
The hard part is instrumentation and normalization — reading activations, routing and role neurons across model families and turning them into one comparable schema. The papers are public; the production data, calibration, and integration are not.
Grounded in six peer-reviewable papers from Fudan University and the Shanghai Innovation Institute. Every number here is real and reproducible — the vision is bold; the evidence is on arXiv.
Let’s talk
Let’s build the white-box layer of AI.
If you work at the frontier of model interpretability, reasoning, and inference, we’d like to talk — research collaborations and partnerships welcome.
Read the inside. It is the next data layer.