ONLINE CAUSAL CONFIGURATION

CAUSALUCB

Online Causal Configuration for Networked Systems
via Doubly Robust Steady-State Learning

SCROLL TO EXPLORE ↓
01 / NETWORK MODEL · INTERACTIVE NETWORK EXPLAINER

A living network.
A stable topology.

Think of a network as a community of interacting units: the nodes and connections stay in place, but signals and local states change whenever a configuration is applied.

Adjust the conceptual interference strength to see how strongly a change in one cluster can echo through its neighbors.

What you’re watching
Breathing clusters, moving signals, and highlighted interference paths—not network growth or learned graph construction.

Conceptual network visualization. Experimental measurements and estimator results are presented in the Evidence section.

NETWORK STATE / 5 CLUSTERS
FIXED EDGES
Five stable clusters connected by fixed links.
Local state Signal propagation Applied configuration
Selected cluster: C · Neighborhood response: moderate
0.45
LocalizedCross-cluster influence

Drag to explore how the illustrated network responds to stronger local interference.

Configuration C
NeurIPS 2026 · Accepted

Online causal configuration for reliable steady-state decisions.

Exploratory telemetry can misrank configurations when measurement perturbs local actions and shifts the post-burn-in context law. CAUSAL-UCB evaluates no-exploration causal value using joint-exposure doubly robust scores and calibrated predecision certificates.

Preview of the NeurIPS 2026 research poster NEURIPS 2026 · RESEARCH POSTERThe paper, at a glance.Identification · certificates · online decisions
CAUSAL DECISION LOOP

Reliable decisions need both identification and deployment.

Global decisions change the actions of many interdependent units. Exploratory measurements can improve the identification of a counterfactual value, yet also alter the operating conditions that the final system should optimize.

The target is no-exploration steady-state utility. Action–exposure overlap alone cannot transport post-burn-in contexts; CAUSAL-UCB separates these sources of error and selects configurations using predecision certified upper values.

01

Interference-aware

Represent how a unit's neighborhood influences its observed outcome.

02

Counterfactual evaluation

Estimate deployment values from exploratory observations with overlap-aware correction.

03

Online configuration

Select future configurations using estimates and uncertainty, not raw telemetry alone.

What must be identified before a configuration can be trusted?

Exploration changes what is observed—and potentially the operating regime whose utility we seek. The paper isolates interference-aware action correction from steady-state context transport, then certifies their remaining uncertainty before selection.

IDENTIFICATION · PROP. 3

Same logs, different target values

Even identical exploratory logs can conceal different no-exploration steady-state values. Exposure overlap is not a guarantee of context transport.

THEOREM 1 · CERTIFICATE

Separate the uncertainty budgets

A predecision value radius separates network-window fluctuation, nuisance products, target-law error, clipping, exposure/response transport, context shifts, configuration mismatch and residual temporal dependence.

THEOREMS 2–3 · DECISION

Information has deployment cost

Measurement improves identification yet perturbs utility; regret guarantees require valid certificates and selected-information growth.

PROPOSITION 1 · RESTRICTED MODEL

One executable radius

Under known local interference, the same anytime calibrated radius supports estimation, optimistic selection and reported uncertainty.

From exploratory telemetry to deployment decisions.

A four-stage causal-optimistic loop.

Certificate scope: The executable calibrated guarantee concerns a restricted model with known ring neighborhoods, matched rewards, bounded noise and the stated assumptions. General network deployment requires verifying the corresponding conditions.

Evidence, from the paper.

Explore seven-method recorded comparisons, manuscript figures and separate certificate audits. Run an independent Python experiment using the tabs below.

EXPERIMENTAL RESULTS · 7 METHODS · 24 SEEDS

Seven methods. One interactive view.

Replay recorded decisions and compare all seven methods, or switch to the Python experiment above.

Open Python experiment ↗
High mismatch · Regret1.732 vs. 6.326Certified DR / Certified IPS
Certificate width0.552 vs. 1.541Decision-active worst-pair diameter
Measurement trade-off30.607 vs. 23.217Deployment gap · DR costs more
Baselines 7/7
METRIC
VIEW
Archived target regret from recorded online decisions.Open Python experiment ↗
Data & display options ⌄
ARCHIVED · SEVEN-METHOD STUDY

Cumulative target regret

Round 200 / 200
Loading verified seven-method archive…
200
All seven methods show matched 200-round results; the five low-mismatch baseline traces are exact verified replays.
Loading original results…
All selected methods — source records24 paired seeds · round 200
7 methods · 24 seeds · 200 rounds
MethodTarget regret ↓Deploy. gap ↓Final ATE ↓Certified width ↓Configuration / opt. rateRecord source
Loading original records…

Target regret and deployment gap are cumulative through the selected round. Final ATE and valid certified width are reported at the end. The configuration column shows the modal choice (with its seed count), plus the final optimal-choice rate when applicable. Baseline methods do not have decision-active certified widths; their missing width is not zero. Low-mismatch baseline round traces are exact verified replays, not original archived intermediate logs.

Research insights View interpretations & comparisons
PAIRED-SEED INTERPRETATION

Decision advantage is conditional, not universal.

Loading paired seed results…

DECISION-ESTIMATION GAP

Loading reference-policy comparison…

RELIABILITY & COST

Loading deployment trade-offs…

About the experiments & source data ↗

Both mismatch regimes now contain 200-round trajectories for all seven methods, at 24 matched seeds. High mismatch and the two low-mismatch certified traces come from the original archive; the five other low-mismatch traces were deterministically reproduced from the frozen driver, with all 168 policy decision hashes and terminal metrics verified. They are newly replayed data, not independently archived round logs. The browser Python Lab is a separate conditional experiment.

Protocol & provenance

Original frozen SYN 2026-10-20 · known sparse ring and matched-noise assumptions · fixed exploration · 24 paired seeds × 200 rounds. Certified curves come from cert_dr_all_rounds.csv. High-mismatch curves for the other five methods come from the exact archived reference/main_per_round.csv.gz and were cross-checked against per_seed.csv. Low-mismatch reference round histories were newly replayed with the same generator and seeds and validated against the original per_seed.csv terminal values and decision hashes. See replayed_low_mismatch_audit.json. All displayed terminal values still come from per_seed.csv and summary.csv; paired intervals come from paired_regret_95CI.csv. No browser-side retraining occurs in this panel.

The paper, figure by figure.

Choose a figure to view its original artwork and the accompanying experimental interpretation.

Manuscript Figure 3: certified online selection, four panels
Fig. 3 · Certified online selection24 paired seeds and 200 rounds: network schematic, target regret, maximum pairwise ATE error and decision-active certified interval diameter.

Figure location in source manuscript · p. 8.

Read official paper ↗
APPENDIX · INDEPENDENT PROTOCOLSCertificate audits & stress testsFull-path coverage · sensitivity · paired regret · hidden violations
Figure 28: 300-seed full-path certificate coverage and OFU path inequality
FIG. 28 · PDF PAGE 61 · 300 SEEDS × 160 ROUNDS

Full-path coverage & OFU inequality

Both full-path target coverage and the OFU path inequality hold in all 300/300 independently seeded audit runs; intervals are 95% Wilson intervals.

INTERPRETATION BOUNDARY

The reported coverage applies to the stated finite-state calibrated protocol and its verified assumptions.

Source: supplied 66-page manuscript · Figure 28 · page 61

Open exact figure ↗Read official paper ↗

Reproduce the verified scope.

This repository contains the executable core comparison, archived benchmark data and figure assets. The supplied paper includes additional certificate audits and trace-driven semi-synthetic studies, which are not rerun by this website.

  • Controlled main experiment: 24 seeds × 200 rounds
  • Parameter-screening sweeps: 4 seeds × 100 rounds (exploratory)
  • Trace-driven semi-synthetic: 8 seeds × 180 rounds
▣ QUICK START ● ● ●
# From repository root
python -m pip install -r requirements.txt
python -m unittest discover -s src/tests -v
python src/fast_compare.py --out results_reproduced --seeds 24 --T 200 --scenarios high_mismatch,low_mismatch
Exact archive protocol for the certified known-interference special case. Other experiments have separate entry points.

Read. Reproduce. Discuss.

Choose the format that fits what you want to do: understand the paper, reproduce a claim, or exchange ideas with the community.

Paper opens the official OpenReview PDF. The archived chart records and source figure crops remain available here for exploration.

Visual sources & licenses

Visual montage: illustrative footage of transport and infrastructure from Pexels; interactive network artwork is procedural. Pexels license ↗

Video sources and replacement guide ↗