Attack assumptions
Interrogate hidden premises, boundary conditions, causal leaps, and untested dependencies.
Independent Scientific Red Team
Adversarial review for frontier physics claims, models, and decisions
The James Scott Institute for Adversarial Frontier Physics conducts external, structured examination of frontier scientific work. We stress-test findings, attack assumptions, probe vulnerabilities, and separate real effects from artifacts, theory from physical realizability, simulation from observation, and independent replication from dependent agreement, then state what the evidence permits.
Adversarial describes the treatment of claims and assumptions, not of people.
Adversarial evidence field
Illustrative schematicThe claim rests on a measurement record, and that record rests on its calibration history.
Positioning
An external assurance layer for high-consequence research
The institute operates as an independent scientific red team for frontier and exotic physics research. Its role is not to promote novelty, defend internal narratives, or reward mathematical ornament. Its role is to examine claims under pressure: to test whether an effect survives contact with metrology, whether a model outruns its validation domain, whether a result depends on hidden assumptions, and whether a decision is being made on evidence or momentum.
This is not certification. It is discriminative review. The output is a justified decision state: what appears supported, what remains uncertain, what is vulnerable to artifact, and what should be tested next before cost, reputation, or strategy compound the error.
Candidate signal
Possible artifact contribution
Unresolved residual
Separating possible contributions does not establish the origin of a residual. No measured data are shown.
What Red Team Means Here
Disciplined external challenge, not theatrical opposition
Red teaming is not cyber intrusion. It is disciplined external challenge applied to scientific claims, evidence packages, models, protocols, and decision logic. The goal is to find the failure mode before reality does.
Challenge measurement
Probe calibration chains, instrumentation drift, environmental confounders, timing errors, contamination pathways, and artifact sources.
Separate evidence classes
Keep observation, inference, simulation, reconstruction, and speculation distinct so conclusions do not outrun the record.
Test independence
Examine whether replication is truly independent or repeated agreement within a shared lineage of tools, data, assumptions, or personnel.
A sophisticated model can still fail physical reality. A repeated claim can still lack independent support.
Why Laboratories Engage
External scrutiny before error compounds
Before publication
Identify weak links in evidence, reasoning, and presentation before a claim hardens in public.
After an anomalous result
Determine whether an apparent effect reflects signal, artifact, confounding structure, or interpretive overreach.
When a program stalls
Locate the true bottleneck: physical, mathematical, computational, metrological, engineering, or governance-related.
Before major funding or partnership
Separate empirical support from simulation-heavy optimism and improve diligence quality.
Before scale-up or transition
Expose failure modes that appear only when a result leaves the bench and enters engineering reality.
When teams are too close to the narrative
Introduce external adversarial scrutiny without collapsing the research effort into politics or personality.
Service Entry Points
Five review paths, engaged singly or in sequence
The institute’s public service architecture is organized around five core lines. Laboratories and sponsors can engage a single review path or combine them in sequence, depending on the decision at hand.
Frontier Result Red Team
Independent, adversarial review of new or extraordinary scientific results. Designed to stress-test claims, surface hidden assumptions, map contradiction pathways, and determine whether the reported effect survives disciplined scrutiny.
A result appears novel, surprising, or strategically important.
Stalled Program Bottleneck Audit
Diagnostic review for research efforts that have slowed, fragmented, or failed to translate. Used to identify whether the true bottleneck is physical, mathematical, computational, metrological, engineering, data-driven, or governance-related.
Pre-Publication Review
External review of manuscripts, preprints, and technical reports before release. Focused on evidence traceability, claim discipline, interpretive restraint, and vulnerability to obvious expert challenge.
Pre-Investment Scientific Review
Evidence-first diligence for funders, sponsors, and strategic partners evaluating high-risk scientific programs. Separates direct evidence from simulation, inference, and speculative framing before capital or reputation is committed.
Lab Attach
Embedded, time-bounded assurance support inside qualified laboratories. Provides hands-on adversarial review, artifact elimination, and replication support without converting the institute into the operating laboratory.
Capability Matrix
Review strategy families behind the five entry points
Explore categories, decision triggers, and example review patterns. Confidential workflows and proprietary internal playbooks remain outside the public catalogue.
5 of 5 strategy families shown.
Scientific validity and claim challenge
- Unsupported inference, hidden assumptions, or an incomplete causal account.
- A novel result depends on an interpretation that has not been sufficiently challenged.
- Claim map, contradiction analysis, alternative-hypothesis comparison, prioritized discriminating tests.
Experimental and metrology attack
- Artifact, calibration weakness, drift, confounding, or dependent replication.
- An apparent effect changes with instrument configuration, environment, processing, or laboratory conditions.
- Artifact pathway map, calibration-chain review, control recommendations, replication-dependency assessment.
Simulation, model, and AI-science assurance
- Numerical error, unbounded extrapolation, model-form weakness, or automation-driven overconfidence.
- A model or AI workflow supports a claim beyond its demonstrated evaluation domain.
- Assumption register, evaluation-domain map, numerical review findings, prioritized verification and validation needs.
Program, governance, and transition review
- Unresolved dependencies, weak decision gates, integration failures, or scale-up assumptions.
- A program stalls or approaches a consequential engineering transition.
- Bottleneck map, decision-gate review, dependency analysis, bounded next-step recommendations.
Diligence, communication, and anomalous-claim review
- Public or investment narratives that outrun the scientific record.
- An extraordinary claim is approaching funding, publication, partnership, or public attention.
- Evidence classification, claim-language review, decision-state memo, falsification priorities.
The five service lines, five analytical groupings, and underlying review strategies are related structures, not interchangeable taxonomies. The complete approved 30-strategy catalogue is not present in this project, so no numerical “view all 30” control is shown.
Attack Surfaces
Where frontier research fails under pressure
The institute does not attack people. It attacks the surfaces where frontier research most often fails under pressure. Select a surface to see where a weakness in it can propagate.
Figure R4
Attack-surface map
Illustrative schematicCore claim logic
Causal leaps, hidden premises, unsupported inference.
Core claim logic
Causal leaps, hidden premises, unsupported inference.
Experimental design
Controls, confounders, protocol fragility, sample handling.
Calibration chain
Drift, synchronization, shielding, contamination, sensor placement.
Data lineage
Provenance gaps, preprocessing opacity, selective inclusion, version ambiguity.
Simulation stack
Invalid assumptions, unbounded extrapolation, numerical instability, surrogate misuse.
AI-science workflow
Benchmark illusion, automation error propagation, interpretability gaps, overfit confidence.
Replication structure
Shared dependencies mistaken for independent confirmation.
Decision architecture
Milestone logic, readiness criteria, escalation thresholds, governance blind spots.
Transition pathway
Manufacturability, integration, reliability, operating-envelope collapse.
External narrative
Claims that outrun the underlying evidence base.
Illustrative review map. These are possible vulnerabilities, not findings about a particular laboratory.
Evidence Classes
Keeping unlike things from being treated as equivalent
A simulation result, an inferred mechanism, and a direct observation may all be useful, but they are not the same kind of evidence.
Observed / Measured
Direct instrument records, calibrated observations, and experimentally grounded measurements.
Computed / Simulated
Model outputs, digital reconstructions, scenario runs, and numerical predictions.
Inferred
Interpretations derived from patterns, correlations, or indirect indicators.
Hypothesized / Speculative
Proposed mechanisms or extraordinary possibilities not yet established by decisive evidence.
Decision-Relevant Unknowns
What remains unresolved but materially affects funding, publication, replication, safety, or transition decisions.
Replication dependency comparison
Engagement Workflow
A time-bounded, scope-limited review
- 1
Inquiry and qualification
A high-level, non-confidential inquiry describes the decision context.
- 2
Scope and triage
The institute determines fit, boundaries, inputs, and review path.
- 3
Evidence mapping and review
Claims, models, protocols, and records are mapped and challenged.
- 4
Findings and decision state
The output states support, uncertainty, vulnerabilities, and next steps.
- 5
Secure follow-through
Sensitive exchange is separately arranged after qualification and agreement.
What a Qualified Review Typically Requires
A bounded evidence package and a clear decision context
High-level description of the claim, result, program, or manuscript.
Decision context and timeline.
Non-confidential summary of methods and technical domain.
Available evidence package description.
Status of experiments, models, or prototypes.
Known anomalies, contradictions, or replication concerns.
Relevant stakeholders and review audience.
Constraints involving security, export control, facility access, or timing.
Do not upload restricted, classified, export-controlled, personal, medical, or confidential research material through public website forms. Public intake is for qualification only.
Outputs and Deliverables
A decision-grade review product, not an endorsement
- Claim and evidence map.
- Contradiction and dependency analysis.
- Artifact and confounder review.
- Calibration-chain challenge summary.
- Simulation versus observation separation.
- Replication independence assessment.
- Decision-state memo.
- Prioritized next-test recommendations.
- Bounded executive briefing.
- 1Claim
- 2Evidence basis
- 3Assumptions
- 4Uncertainty
- 5Contradictions
- 6Next discriminating test
- 7Decision relevance
A clean review reduces a specific class of error. It does not remove uncertainty from frontier research.
Boundaries and Exclusions
A deliberately narrow role, stated in public
- It does not certify, license, accredit, or validate results.
- It does not rank researchers, laboratories, or institutions.
- It does not claim discoveries or co-author claims it reviews.
- It does not present simulation as observation or model agreement as physical truth.
- It does not guarantee breakthrough, publication, replication, investment, or program success.
- It does not provide covert collection, intrusive surveillance, or unauthorized data acquisition.
- It does not expose confidential methods or protected partner information.
- It does not replace legal, regulatory, IP, or safety authorities.
Adversarial rigor is applied to claims, assumptions, evidence, and decisions, never as a performance of hostility toward people.
Trust and Data Handling
The public site is not a channel for sensitive disclosure
Public content and inquiry forms are limited to non-confidential information such as professional identity, organization, role, review interest, decision timeline, and a brief non-sensitive summary.
Confidential workflows, protected evidence handling, and secure document exchange remain off the public site. Sensitive exchange proceeds only through authenticated, encrypted channels after scope alignment and mutual agreement.
01
Public qualification
Non-confidential information only
02
Scope agreement
03
Separately arranged sensitive exchange
FAQ
Questions about scientific red teaming
What does “red team” mean here?
Structured external challenge applied to claims, evidence, models, and decisions. It does not mean cyber intrusion or hostility toward researchers.
Who engages the institute?
Frontier laboratories, advanced R&D groups, technical sponsors, funders, strategic partners, and serious journalists evaluating high-consequence scientific claims.
Does the institute certify results?
No. It does not certify, license, accredit, or validate results. It produces a justified review state based on the evidence and defined scope.
Can you tell us whether our effect is real?
The institute can assess whether the current evidence better supports a real effect, an unresolved question, an artifact pathway, or a premature conclusion.
Do you review simulations and AI workflows?
Yes, where relevant. Simulated, inferred, and observed evidence remain separate, and model claims are bounded by their demonstrated evaluation domain.
Can confidential material be submitted here?
No. Public forms are for non-confidential qualification only. Sensitive exchange is separately arranged if a review proceeds.
Do you work inside laboratories?
In qualified cases, through time-bounded Lab Attach engagements. These are scoped assurance missions, not indefinite operating roles.
What is the output of a review?
A structured decision-grade assessment describing what appears supported, what remains uncertain, where vulnerabilities exist, and what should be tested next.
Bring the claim under pressure before the world does
If a result, model, manuscript, or program will influence funding, publication, partnership, or strategic direction, it should survive external adversarial review before the cost of error compounds.
Public inquiries must remain non-confidential. Sensitive exchange begins only after qualification.