Uncertainty Externalization Overview

• Uncertainty externalization is the process of transferring latent risks, assumptions, and value judgments from core analysis to an external, explicit interface.
• In energy policy modelling, this approach critiques false certainty by highlighting hidden assumptions, while in LLMs it improves calibration and error control.
• Across domains like MBSE, agent systems, and FX market making, externalization serves to either obscure or operationalize uncertainty, balancing risk and actionable insights.

Searching arXiv for the cited papers and topic-specific references to ground the article. arxiv_search(query="^^^^^^^^8\8 uncertainty in the modelling process for energy policy making^^^^^^^^8\8 OR ^^^^^^^^8\8 Should Express Uncertainty Explicitly^^^^^^^^8\8 OR ^^^^^^^^8\8 Reasoning in LLMs through Strategic Information Allocation under Uncertainty^^^^^^^^8\8 OR ^^^^^^^^8\8 Laundering in Agent Systems^^^^^^^^8\8 OR ^^^^^^^^8\8 All Uncertainty Is Equal^^^^^^^^8\8 OR ^^^^^^^^8\8 Modeling for SysML v^^^^8 OR \8^^^^^^^^^^^^8\8 OR ^^^^^^^^8\8 Uncertainty with Uncertainty^^^^^^^^8\8 OR ^^^^^^^^8\8 making by an FX dealer^^^^^^^^8\8 max_results=^^^^8 OR \8\8^^^^) Uncertainty externalization denotes a family of operations in which uncertainty is displaced from one locus of analysis or control to another rather than eliminated. In energy-policy modelling, it refers to pushing assumptions, value judgments, and gaps in knowledge outside the formal analysis, thereby creating an “excess of certainty” around particular policy options (&&&^^^^8\8^^^^&&&). In recent LLM work, the same expression is used in an almost opposite sense: latent uncertainty is made explicit as a trainable interface for abstention, retrieval, verification, and intervention (&&&^^^^8 OR \8^^^^&&&). Related literatures use the term for preserving uncertainty across agent handoffs through latent carriers (&&&^^^^8 OR \8^^^^&&&), encoding uncertainty as first-class structure in SysML v^^^^8 OR \8^^^^ (&&&^^^^8 OR \8^^^^&&&), transferring inventory risk from an FX dealer’s book to the inter-dealer market (&&&^^^^8 OR \8^^^^&&&), or introducing randomness into the decision rule itself through “Randoptimization” (&&&^^^^8 OR \8^^^^&&&). Across these uses, the central question is not whether uncertainty exists, but whether it is hidden, exposed, propagated, or operationalized at the decision interface.

^^^^8 OR \8^^^^. Domain-specific meanings

The term has no single universal definition across arXiv literatures. Instead, it names several distinct interface phenomena.

Domain	Meaning of uncertainty externalization	Representative paper
Energy policy modelling	Uncertainty is pushed outside formal analysis	(&&&^^^^8\8^^^^&&&)
LLM inference	Uncertainty is expressed explicitly for control	(&&&^^^^8 OR \8^^^^&&&)
LLM reasoning	Latent uncertainty is verbalized as epistemic information	(Kim et al., 16 Mar 2026)
Agent systems	Fragility is preserved across handoffs via latent carriers	(&&&^^^^8 OR \8^^^^&&&)
MBSE / SysML v^^^^8 OR \8^^^^	Uncertainty is modeled explicitly in system structure	(&&&^^^^8 OR \8^^^^&&&)
Decision theory / operations	Randomness is introduced into the policy itself	(&&&^^^^8 OR \8^^^^&&&)
FX market making	Inventory risk is transferred to the market	(&&&^^^^8 OR \8^^^^&&&)

In the energy-modelling literature, externalization is a pathology: “black-box” models, single cost-optimal pathways, and reliance on a single model or indicator suppress deep uncertainty and close off alternatives (&&&^^^^8\8^^^^&&&). In LLM interface design, by contrast, externalization is treated as a design goal: uncertainty should be a first-class, trainable interface that the model exposes explicitly to downstream controllers (&&&^^^^8 OR \8^^^^&&&). In agent systems, the emphasis shifts from local estimation to uncertainty-preserving interface design, because uncertainty propagates only when it survives the handoff between components (&&&^^^^8 OR \8^^^^&&&).

This suggests that the common denominator is representational location. What changes across fields is whether moving uncertainty outward is regarded as loss, preservation, or deliberate control.

^^^^8 OR \8^^^^. Energy-policy modelling and the critique of false certainty

In “Unpacking uncertainty in the modelling process for energy policy making” (&&&^^^^8\8^^^^&&&), uncertainty externalization is defined as the tendency of energy-system modelling to generate crisp scenarios or single “best-estimate” indicators while leaving unexamined assumptions, institutional choices, contested framings, and non-quantifiable uncertainties outside the formal analysis. The paper locates this problem in Post-Normal Science, where facts are uncertain, stakes are high, values are disputed and decisions are urgent. Externalization manifests as “black-box” models, fixation on one or two cost-optimal pathways, neglect of qualitative or “deep” uncertainties, and reliance on a single model or indicator for major policy commitments.

The proposed remedies are three reflexive-modelling methodologies. NUSAP decomposes a quantity into Numeral, Unit, Spread, Assessment, and Pedigree. Its pedigree component scores each assumption against proxy representation, empirical basis, methodological rigour, validation, and theoretical understanding. In matrix form, if PRESERVED_PLACEHOLDER_^^^^8\8^^^^ is the pedigree matrix, an aggregate score may be written as

PRESERVED_PLACEHOLDER_^^^^8 OR \8^^^^

Diagnostic diagrams then place each assumption on a two-axis scatter plot with pedigree on the horizontal axis and sensitivity on the vertical axis. The “Danger zone” is PRESERVED_PLACEHOLDER_^^^^8 OR \8^^^^: high sensitivity and poor pedigree. Sensitivity auditing extends beyond technical UA & SA to interrogate framing, scope-definition, rhetoric of numbers, hidden value judgments, and governance assumptions. Its mathematical hook is the first-order Sobol’ index

PRESERVED_PLACEHOLDER_^^^^8 OR \8^^^^

The four detailed case studies show how these methods recapture uncertainty that conventional modelling had externalized. In the ExternE case, a transdisciplinary workshop identified ^^^^8 OR \8\8^^^^ assumptions, reduced them to six critical ones, and found that all six received weak aggregate pedigree. In ESME, BioRES and CCSmbr lay in PRESERVED_PLACEHOLDER_^^^^8 OR \8^^^^, implying high influence on total system cost but very weak pedigree. In the NETs-in-IAMs case, Vaughan & Gough’s ^^^^8 OR \88^^^^-expert workshop found that almost all NETs parameters sat in the PRESERVED_PLACEHOLDER_^^^^8 OR \8^^^^ danger zone. In the Ecological Footprint case, the seven-point auditing checklist exposed omitted sink functions, hidden technical coefficients, absent error bars, and no published uncertainty or sensitivity analysis.

The significance of this literature is methodological rather than merely computational. It treats uncertainty as partly epistemic, value-laden, and institutional, so quality assurance requires workshops, pedigree matrices, diagnostic plots, and sensitivity auditing rather than only tighter confidence intervals.

^^^^8 OR \8^^^^. Explicit uncertainty in LLMs: global confidence and local signals

“LLMs Should Express Uncertainty Explicitly” reframes uncertainty as an interface for control rather than a latent quantity estimated post hoc (&&&^^^^8 OR \8^^^^&&&). It distinguishes two decision problems. Outcome uncertainty—“Should I trust this final answer?”—is addressed by a global scalar. Process uncertainty—“Am I hitting a fragile step in my chain-of-thought?”—is addressed by local signals during generation. The global interface appends a decimal confidence token PRESERVED_PLACEHOLDER_^^^^8 OR \8^^^^ after generating a reasoning trajectory PRESERVED_PLACEHOLDER_^^^^8 OR \8^^^^ and an answer $\hat y$, with

$$c \approx \Pr(Y=1\mid h_{1:T}),\qquad Y=\mathbf1\{\hat y=\text{gold}\}.$$

The local interface allows the model to emit the literal token “^^^^^^^^8\8 during reasoning, with

PRESERVED_PLACEHOLDER_^^^^8 OR \8\8^^^^

Both interfaces are learned through a light post-training stage using a variant of GRPO. The idealized update is

PRESERVED_PLACEHOLDER_^^^^8 OR \8 OR \8^^^^

For verbalized confidence,

PRESERVED_PLACEHOLDER_^^^^8 OR \8 OR \8^^^^

which rewards justified confidence and penalizes over-confident errors. For “^^^^^^^^8\8 the reward order is

PRESERVED_PLACEHOLDER_^^^^8 OR \8 OR \8^^^^

implemented as PRESERVED_PLACEHOLDER_^^^^8 OR \8 OR \8^^^^.

The reported effects are large. On ^^^^8 OR \8^^^^WikiMultihopQA, verbalized confidence changes accuracy from PRESERVED_PLACEHOLDER_^^^^8 OR \8 OR \8^^^^, ECE from PRESERVED_PLACEHOLDER_^^^^8 OR \8 OR \8^^^^, Brier from PRESERVED_PLACEHOLDER_^^^^8 OR \8 OR \8^^^^, NLL from PRESERVED_PLACEHOLDER_^^^^8 OR \88^^^^, and the over-confidence gap from PRESERVED_PLACEHOLDER_^^^^8 OR \8^^^^9; wrong answers with PRESERVED_PLACEHOLDER_^^^^8 OR \8\8^^^^ drop from PRESERVED_PLACEHOLDER_^^^^8 OR \8 OR \8^^^^ to PRESERVED_PLACEHOLDER_^^^^8 OR \8 OR \8^^^^ (&&&^^^^8 OR \8^^^^&&&). In Adaptive RAG aggregated over five datasets, Verbal-Calibrate reaches EM/F^^^^8 OR \8^^^^ PRESERVED_PLACEHOLDER_^^^^8 OR \8 OR \8^^^^ at PRESERVED_PLACEHOLDER_^^^^8 OR \8 OR \8^^^^ retrieval, while Uncertain-Calibrate reaches PRESERVED_PLACEHOLDER_^^^^8 OR \8 OR \8^^^^ at PRESERVED_PLACEHOLDER_^^^^8 OR \8 OR \8^^^^ retrieval. Mechanistically, the verbal interface concentrates token-level KL divergence almost exclusively at the final confidence digit and preserves layer geometry, whereas the special-token interface induces progressive late-layer CKA drop to PRESERVED_PLACEHOLDER_^^^^8 OR \8 OR \8^^^^, indicating a broader reorganization.

A complementary theoretical account is given in “Understanding Reasoning in LLMs through Strategic Information Allocation under Uncertainty” (Kim et al., 16 Mar 2026). That paper decomposes reasoning into procedural information and epistemic verbalization. If PRESERVED_PLACEHOLDER_^^^^8 OR \88^^^^ is a latent internal variable, then making it verbalized and therefore conditionable yields

PRESERVED_PLACEHOLDER_^^^^8 OR \8^^^^9

The paper further argues that purely procedural reasoning can become informationally stagnant, whereas sporadic epistemic verbalization can ensure continued information acquisition. Empirically, masking epistemic tokens in top-performers drops accuracy by PRESERVED_PLACEHOLDER_^^^^8 OR \8\8^^^^ and PRESERVED_PLACEHOLDER_^^^^8 OR \8 OR \8^^^^; removing all epistemic verbalization from high-quality procedural traces degrades Qwen^^^^8 OR \8^^^^-^^^^8 OR \8 OR \8^^^^B base from PRESERVED_PLACEHOLDER_^^^^8 OR \8 OR \8^^^^ pass@^^^^8 OR \8^^^^ on AIME^^^^8 OR \8 OR \8^^^^.

A frequent misconception is that uncertainty expression is merely a stylistic overlay. These two papers argue the opposite: explicit uncertainty can alter calibration, retrieval control, and the internal routing of computation.

^^^^8 OR \8^^^^. Propagation across interfaces: agent handoffs and model-based systems engineering

In multi-step agent systems, the central problem is not only whether uncertainty exists locally, but whether it is preserved at component boundaries. “Confidence Laundering in Agent Systems: Why Uncertainty Needs a Latent Carrier” defines uncertain decision handoff as the transfer of an intermediate decision made under uncertainty (&&&^^^^8 OR \8^^^^&&&). If PRESERVED_PLACEHOLDER_^^^^8 OR \8 OR \8^^^^ is an upstream distribution over possible artifacts and the interface applies a deterministic commitment map

PRESERVED_PLACEHOLDER_^^^^8 OR \8 OR \8^^^^

then distinct epistemic states can collapse to the same PRESERVED_PLACEHOLDER_^^^^8 OR \8 OR \8^^^^. Downstream components receive a procedurally valid artifact but lose access to its fragility. That failure mode is called confidence laundering. The proposed remedy is a latent uncertainty carrier PRESERVED_PLACEHOLDER_^^^^8 OR \8 OR \8^^^^, attached to the committed artifact through an augmented handoff

PRESERVED_PLACEHOLDER_^^^^8 OR \8 OR \8^^^^

In HotpotQA web-search experiments, latent carriers achieve AUROC PRESERVED_PLACEHOLDER_^^^^8 OR \88^^^^ in binary uncertainty detection, versus PRESERVED_PLACEHOLDER_^^^^8 OR \8^^^^9 with scalar scores and PRESERVED_PLACEHOLDER_^^^^8 OR \8\8^^^^ with answer-only.

A structurally analogous agenda appears in “Uncertainty Modeling for SysML v^^^^8 OR \8^^^^” (&&&^^^^8 OR \8^^^^&&&). There, uncertainty is externalized into explicit model constructs by incorporating the PSUM metamodel into SysML v^^^^8 OR \8^^^^. The core stereotypes include PRESERVED_PLACEHOLDER_^^^^8 OR \8 OR \8^^^^, PRESERVED_PLACEHOLDER_^^^^8 OR \8 OR \8^^^^, PRESERVED_PLACEHOLDER_^^^^8 OR \8 OR \8^^^^, PRESERVED_PLACEHOLDER_^^^^8 OR \8 OR \8^^^^, PRESERVED_PLACEHOLDER_^^^^8 OR \8 OR \8^^^^, and PRESERVED_PLACEHOLDER_^^^^8 OR \8 OR \8^^^^. Uncertainty carries attributes such as PRESERVED_PLACEHOLDER_^^^^8 OR \8 OR \8^^^^, PRESERVED_PLACEHOLDER_^^^^8 OR \88^^^^, PRESERVED_PLACEHOLDER_^^^^8 OR \8^^^^9, and PRESERVED_PLACEHOLDER_^^^^8 OR \8\8^^^^. Propagation is represented by explicit references from uncertainties to effects, so downstream consequences remain traceable rather than implicit. The extension was validated on seven publicly available SysML v^^^^8 OR \8^^^^ models totaling ^^^^8 OR \8^^^^,^^^^8 OR \88 OR \8^^^^ lines of model and over ^^^^8 OR \8^^^^,^^^^8\8\8\8^^^^ elements.

The common significance is interface preservation. In agent systems, preservation requires a latent carrier attached to decision handoffs. In MBSE, preservation requires stereotypes, constraints, measurements, and effect links. In both cases, uncertainty becomes actionable only when it is retained in a form that downstream processes can consume.

^^^^8 OR \8^^^^. Externalization as controlled transfer: randomized policies and hedging

A different tradition treats uncertainty externalization as a deliberate control strategy. In “Fighting Uncertainty with Uncertainty: A Baby Step,” Kashyap calls this “Randoptimization” (&&&^^^^8 OR \8^^^^&&&). The idea is to choose a randomized policy rather than a single deterministic action, thereby pushing some randomness into the decision-maker’s own control. In the benchmark news-vendor problem, deterministic profit under order quantity PRESERVED_PLACEHOLDER_^^^^8 OR \8 OR \8^^^^ is

PRESERVED_PLACEHOLDER_^^^^8 OR \8 OR \8^^^^

with classical optimizer

PRESERVED_PLACEHOLDER_^^^^8 OR \8 OR \8^^^^

Randoptimization instead draws PRESERVED_PLACEHOLDER_^^^^8 OR \8 OR \8^^^^, yielding

PRESERVED_PLACEHOLDER_^^^^8 OR \8 OR \8^^^^

A sufficient condition for randomized performance to match or exceed the deterministic optimum is

PRESERVED_PLACEHOLDER_^^^^8 OR \8 OR \8^^^^

Under the simplifying constraint PRESERVED_PLACEHOLDER_^^^^8 OR \8 OR \8^^^^, Kashyap gives an additional bound on how spread-out PRESERVED_PLACEHOLDER_^^^^8 OR \88^^^^ may be. The paper sketches applications to school admissions, job-candidate selection, journal submissions, stock picking, and monetary policy.

In FX market making, externalization has a more literal market meaning. “Market making by an FX dealer: tiers, pricing ladders and hedging rates for optimal risk control” defines internalization as absorbing client flow imbalances on-book and using price-skewing to attract offsetting trades, whereas externalization means trading in the inter-dealer segment to immediately hedge inventory (&&&^^^^8 OR \8^^^^&&&). The dealer controls bid/ask skews and an external execution rate PRESERVED_PLACEHOLDER_^^^^8 OR \8^^^^9. The reference price follows

PRESERVED_PLACEHOLDER_^^^^8 OR \8\8^^^^

With execution cost PRESERVED_PLACEHOLDER_^^^^8 OR \8 OR \8^^^^, the optimal hedging rate is

PRESERVED_PLACEHOLDER_^^^^8 OR \8 OR \8^^^^

where PRESERVED_PLACEHOLDER_^^^^8 OR \8 OR \8^^^^. If PRESERVED_PLACEHOLDER_^^^^8 OR \8 OR \8^^^^, then PRESERVED_PLACEHOLDER_^^^^8 OR \8 OR \8^^^^: pure internalization. For a realistic EURUSD top-tier bank, the paper reports PRESERVED_PLACEHOLDER_^^^^8 OR \8 OR \8^^^^ bps/PRESERVED_PLACEHOLDER_^^^^8 OR \8 OR \8^^^^, PRESERVED_PLACEHOLDER_^^^^8 OR \88^^^^ dayPRESERVED_PLACEHOLDER_^^^^8 OR \8^^^^9, PRESERVED_PLACEHOLDER_^^^^8 OR \8\8^^^^ bps·day per M€, PRESERVED_PLACEHOLDER_^^^^8 OR \8 OR \8^^^^ bps, PRESERVED_PLACEHOLDER_^^^^8 OR \8 OR \8^^^^ bps per M€, and PRESERVED_PLACEHOLDER_^^^^8 OR \8 OR \8^^^^ bpsPRESERVED_PLACEHOLDER_^^^^8 OR \8 OR \8^^^^ per M€; under these parameters, the dealer fully internalizes for PRESERVED_PLACEHOLDER_^^^^8 OR \8 OR \8^^^^–PRESERVED_PLACEHOLDER_^^^^8 OR \8 OR \8^^^^ M€ and internalizes about PRESERVED_PLACEHOLDER_^^^^8 OR \8 OR \8^^^^ of flow.

These two literatures use the same word for distinct mechanisms. Randoptimization externalizes environmental uncertainty into a controlled sampling distribution. FX hedging externalizes inventory risk to the broader market. In both cases, uncertainty is not removed; it is transferred into a mechanism with explicit design parameters.

^^^^8 OR \8^^^^. Granularity, verification, and the limits of visibility

Making uncertainty visible does not guarantee that downstream actors will use it well. “Not All Uncertainty Is Equal: How Uncertainty Granularity Shapes Human Verification in LLM-Assisted Decision Making” studies output-level, relation-level, and token-level uncertainty displays in a between-subjects study with PRESERVED_PLACEHOLDER_^^^^8 OR \88^^^^ (&&&^^^^8 OR \8 OR \8^^^^&&&). Token-level uncertainty is computed from next-token distributions, for example through per-token entropy

PRESERVED_PLACEHOLDER_^^^^8 OR \8^^^^9

or through $\hat y$^^^^8\8^^^^, where $\hat y$^^^^8 OR \8^^^^. Relation-level uncertainty assigns a score $\hat y$^^^^8 OR \8^^^^ or $\hat y$^^^^8 OR \8^^^^–y^\hat y^^^^8 OR \8^^^^ to each reasoning step or claim. Output-level uncertainty provides a single scalar $\hat y$^^^^8 OR \8^^^^ for the entire response.

The behavioral effects differ sharply by granularity. Agreement with AI is Baseline $\hat y$^^^^8 OR \8^^^^, UQ-Token $\hat y$^^^^8 OR \8^^^^, UQ-Relation $\hat y$8, and UQ-Output $\hat y$9; UQ-Token exceeds Baseline with $$c \approx \Pr(Y=1\mid h_{1:T}),\qquad Y=\mathbf1\{\hat y=\text{gold}\}.$$^^^^8\8^^^^. Confidence in one’s own answers decreases under UQ-Output and UQ-Relation but not under UQ-Token. Relation-level uncertainty reduces external verification: link clicks fall from Baseline $$c \approx \Pr(Y=1\mid h_{1:T}),\qquad Y=\mathbf1\{\hat y=\text{gold}\}.$$^^^^8 OR \8^^^^ to $$c \approx \Pr(Y=1\mid h_{1:T}),\qquad Y=\mathbf1\{\hat y=\text{gold}\}.$$^^^^8 OR \8^^^^ with $$c \approx \Pr(Y=1\mid h_{1:T}),\qquad Y=\mathbf1\{\hat y=\text{gold}\}.$$^^^^8 OR \8^^^^, and web searches fall from $$c \approx \Pr(Y=1\mid h_{1:T}),\qquad Y=\mathbf1\{\hat y=\text{gold}\}.$$^^^^8 OR \8^^^^ to $$c \approx \Pr(Y=1\mid h_{1:T}),\qquad Y=\mathbf1\{\hat y=\text{gold}\}.$$^^^^8 OR \8^^^^ with $$c \approx \Pr(Y=1\mid h_{1:T}),\qquad Y=\mathbf1\{\hat y=\text{gold}\}.$$^^^^8 OR \8^^^^. Task accuracy and completion time show no reliable differences.

The result is a direct caution against an intuitive but overgeneral claim: more detailed uncertainty cues are not necessarily more responsible. Token-level heatmaps can increase agreement with the AI, whereas relation-level cues can steer users away from independent fact-checking. This suggests that uncertainty externalization must be evaluated not only by calibration metrics or representational fidelity, but also by the behavioral pathways it induces in verification, oversight, and intervention.

Across these literatures, the decisive issue is therefore not visibility alone. The consequential properties are whether uncertainty is retained at the right interface, encoded at the right granularity, and matched to an action that can actually use it.