Stress-test research plans by comparing possible scenarios, such as sample losses, unavailable materials, or methods failing to transfer. Record assumptions and evidence, compare response options, and define contingency plans and observable triggers for changing course.
Supports structured what-if scenario analysis for research planning, experimental contingencies, and scientific project decisions. Explores favorable, reference, adverse, wild-card, contrarian, and second-order scenarios with explicit assumptions, evidence, and decision triggers. Use to stress-test a research plan under uncertainty; scenario narratives do not estimate causal effects or calibrated forecast probabilities.
Run score Not tested
See how to installwhat-if-oracle helps examine how a research plan might fare under several possible conditions. It organizes scenario comparisons, records supporting evidence and assumptions separately, and outlines contingency plans. Scenarios can include favorable, reference, adverse, unexpected interruption, alternative hypotheses, and downstream effects. It also helps identify observations and thresholds that could prompt a change in response.
Use it to prepare for sample loss, difficulty obtaining reagents or data, or a method that may not transfer to another instrument or study group. It can help compare options by cost, feasibility, and reversibility, and identify what additional measurement could change a decision.
The qualitative workflow needs no package, API key, or network connection. It is designed for research planning. Scenario narratives do not estimate causal effects or event probabilities. Numerical claims need supporting evidence and a calculation method. The skill does not carry out experiments, purchases, or plan changes. It is licensed under CC BY-NC-SA 4.0; commercial use requires a license from the author.
This overview was written by AI (OpenAI Codex) from the public SKILL.md. Check the original below for the exact behavior.
AI checked the SKILL.md for operations that need care and for setup requirements. The score comes only from actually running the skill (hands-on test).
Not counted in the score. Rated by AI (OpenAI Codex) as of Oct 24, 2026. Treat it as a guide and read the original before installing.
Review the instructions the agent will receive before you install.
Explore a bounded set of possible futures before committing research resources. The output is a scenario comparison, an evidence ledger, and a contingency plan. It does not cover the full possibility space or establish that a narrated mechanism will occur.
The six branch prompts and 0·IF·1 mnemonic come from the upstream What-If framework. Treat the mnemonic as a framing aid: a potential change (0), a stated condition (IF), and a resulting scenario (1). It is not a scientific law or a computational model.
The author's What-If Paradigm and Unified Theory of Digital Consciousness are catalogued by Zenodo as preprints. Their deposit and DOI establish provenance, not forecasting accuracy or empirical validation of this workflow.
Use for questions such as:
Keep the analysis attached to a scientific decision. This is not a general business, personal advice, or software architecture skill. If the question requires an intervention effect, a power calculation, or a numerical simulation, identify the required design/model and evidence; do not manufacture those results with scenario prose.
No package, API endpoint, credential, or network connection is required for the qualitative workflow. Verify current external facts when they matter, or label the analysis as based only on supplied material. For research-specific prompts and a worked example, see references/scenario-templates.md.
Write one operational question containing:
For example: "If the usable-sample fraction is 60% rather than the planned 80%, can a batch of 100 samples deliver 80 usable samples this month, and which contingency should we prepare?" Both fractions may be planning assumptions until supported by data.
Use the user's supplied framing when sufficient. State reasonable working assumptions and proceed; ask for a missing fact only when it materially changes the analysis.
Maintain a short evidence ledger: claim, source or data version/date, relevant population/protocol, uncertainty, and status (observed, model-derived, or assumed). Record conflicting evidence. An LLM-generated rationale is not an additional observation or an independent expert assessment.
Select distinct scenarios suited to the question, usually four to six. A short screening can use favorable, reference, and adverse scenarios. The labels below are exploration lenses, not mutually exclusive outcomes or a probability distribution.
| Lens | What to examine | Scientific boundary |
|---|---|---|
| Ω Favorable (best case) | Favorable conditions within justified bounds | Not a proven upper limit; specify whose outcome improves |
| α Reference case | A stated baseline or central set of assumptions | Call it "likely" only if comparative evidence supports that ranking |
| Δ Adverse (worst case) | Plausible failures, including dependent failures | Not a proven lower limit or the worst imaginable outcome |
| Ψ Wild card | A concrete disruption outside the main assumptions | Do not claim to enumerate unknowable events or assign a rarity from the label |
| Φ Contrarian | An alternative to a dominant hypothesis | Evaluate supporting and disconfirming evidence; disagreement alone is not support |
| ∞ Second order | Downstream effects, delays, feedback, or changed behavior | Can occur within any other scenario; mark untested mechanisms as hypotheses |
Retain important shared drivers and correlations across branches. Avoid implausible combinations created by independently toggling quantities that share a cause. Use descriptive scenario names in the final output so favorable/adverse labels do not substitute for evaluation.
Use a compact record:
Scenario: [name and lens]
Condition and horizon: [operational definition]
Assumptions: [including dependencies and boundary conditions]
Evidence: [source/data version, supporting and conflicting observations]
Mechanism: [conditional narrative; identify untested causal links]
Consequences: [outcomes, units, range, and relevant time points]
Likelihood: [not estimated, or justified estimate with method and uncertainty]
Evidence confidence: [reasoned assessment, separate from event likelihood]
Discriminating observation: [what would weaken this scenario or favor another]
Trigger and response: [observable measure, threshold, review time, owner, action]
Residual risk: [what this response does not resolve]
For quantitative claims, show inputs, units, method, assumptions, and whether the calculation was executed. Distinguish variation in future outcomes from uncertainty in the inputs or model. Sensitivity checks should vary decision-critical assumptions, including plausible joint changes, and identify where the preferred action changes.
For causal claims, specify the intervention, comparator, population, outcome, and horizon. An observational association or a changed prediction is not by itself an intervention effect. State the identification assumptions and supporting design; for observational adjustment these include consistency, exchangeability, and positivity. A scenario can motivate a controlled pilot or measurement, but cannot replace that evidence. See the author's maintained Causal Inference: What If, especially Chapters 1–3, for this distinction.
Compare candidate actions against the scenarios using a table of outcome, cost, feasibility, reversibility, and unresolved assumptions. Include the current plan. State which objective or trade-off determines the recommendation.
Do not turn a recommendation into an executed experiment, purchase, or protocol change. The deliverable is the analysis and proposed response unless the user separately requested execution.
Use "not estimated" when evidence cannot support a numerical probability. If a probability is useful, define the event and horizon, identify a relevant base rate or fitted model, and record estimation uncertainty and transfer limitations. Qualitative plausibility is not a hidden numeric score. Confidence concerns the strength and consistency of evidence; it is not the probability that an event happens. This distinction is also used in the IPCC uncertainty guidance.
Normalize only a mutually exclusive, collectively exhaustive outcome partition under the same conditioning assumptions and horizon. For example, the final number of usable samples from one fixed batch of 100 can be partitioned as 0–59, 60–79, and 80–100. That defines valid outcome bins, but supplies no probabilities for them. The six scenario lenses do not form such a partition.
Keep resource allocation separate from event likelihood. A 61.8/38.2 golden-ratio allocation is not justified by this workflow. Choose allocations from evidence, consequences, costs, capacity, reversibility, and the stated objective.
P(A and B) = P(A) × P(B given A), not a product of marginal probabilities without justified independence. Stop when added detail cannot change the action or lacks evidence.The current Government Office for Science Futures Toolkit distinguishes scenarios from predictions and describes stress-testing options across possible futures. This skill adapts those general distinctions to research contingencies; the toolkit does not validate these six particular lenses, their probabilities, or a prescribed number of branches.
This documentation-only skill contains no forecast engine or causal estimator. Its worked example is synthetic and illustrates planning arithmetic, not demonstrated improvement in real research outcomes.
© 2026 Ashraf Hussein Kahoush / AHK Strategies. Licensed under CC BY-NC-SA 4.0. Commercial use requires a license from the author. This repository adaptation narrows the workflow to research planning and revises evidence, probability, and causal-claim guidance; it preserves attribution to the upstream framework and preprints.
No reviews yet. Tell others what you thought.
No login needed. Share how it worked for you, ask a question, or note where it did not run.
No comments yet. Be the first to write one.
Estimate intracellular reaction rates from steady-state carbon-13 tracing measurements. Validate carbon atom mappings and measurement inputs, simulate labeling, fit constrained fluxes, and diagnose which pathway rates the data can meaningfully constrain.
Plan protein characterization experiments with Adaptyv's cloud laboratory, estimate costs, submit sequences, track progress, and retrieve measured results. Supports binding, thermal stability, expression, fluorescence, and enzyme activity workflows through the API and Python SDK.
Analyze time series with aeon, a Python toolkit for data that changes over time. Supports classification, regression, clustering, forecasting, anomaly detection, and pattern search, with guidance on preprocessing, algorithm selection, and evaluation.
Investigate how DNA variants may affect gene activity using precomputed AlphaGenome Atlas results and model predictions. Rank candidate variants, examine tissue-specific effects, and inspect the features behind scores to develop research hypotheses about biological mechanisms.
Plan and document studies that assess whether an analytical procedure suits its intended use. Select the governing framework, evaluate calibration, accuracy, precision, and detection limits, and compare methods using acceptance criteria defined before data collection.
Organize single-cell measurements with cell and gene metadata, read and write h5ad or Zarr files, and prepare subsets or combined datasets. Includes guidance on handling large annotated matrices without loading all measurement values into memory.