lubluniky/dale/dale-lenses

Dale Lenses

Change the representation of a hard problem before changing the answer. Apply a specific epistemic method with an observable artifact and a stop rule; never simulate rigor by naming several frameworks and repeating the same opinion.

Keep the method bounded

• Preserve the user's current phase. Analyze, decide, plan, or implement only to

the extent requested; invoking a lens does not authorize execution.

• Require a bounded question. When the user is still discovering the goal,

options, taste, or priorities, keep or move the conversation to ordinary brainstorming instead of imposing a lens prematurely.

• Stay in the current task. Do not create Codex tasks or subagents merely to

manufacture perspectives. Use tools only when actual evidence is needed and already authorized.

• Separate observations, inferences, assumptions, preferences, and unknowns.
• Redact secrets, credentials, raw environment values, personal data, private

transaction identifiers, and other sensitive operational evidence before it enters a tool call, artifact, or response.

• Select one primary lens. Add one challenging lens only when it makes a

materially different prediction, exposes a different failure mode, or could reverse the decision.

• Honor a lens named by the user. If it does not fit the problem, explain the

mismatch briefly and offer the closest fitting lens instead of silently substituting it.

• Decline the method when ordinary reasoning is sufficient. “No useful

reframe” is better than framework theatre.

Select by problem signature

Read references/lens-contracts.md completely before applying a lens.

Do not select by novelty. Select the lens whose required artifact could most change the next decision.

Run the lens

1. Bound the question

State the exact question or decision, system boundary, time horizon, consequence of error, and evidence currently available. Resolve an ambiguous scope from local context when safe; ask at most one question only when different answers would materially change the lens or decision.

2. Precommit the method

Name the primary lens and its concrete reason in one line. If using a challenging lens, state what different prediction or failure mode justifies it. Define the artifact and method-specific terminal condition before drawing a conclusion.

3. Construct, then attack

Build the artifact required by the selected lens. Preserve rival models and negative evidence. Record the strongest counterargument to the leading model before supporting it. Fix the falsifier, invalidation threshold, update trigger, or boundary of identification before interpreting any new result; do not move the gate after seeing evidence.

When evidence can be gathered safely, prefer the cheapest observation that distinguishes live models. Do not ask for measurements that cannot affect the decision. Never invent telemetry, probabilities, causality, constraints, or system boundaries.

4. Expose the crux

Compress the result into a decision-relevant crux map:

question: <bounded question>
evidence boundary: <observed sources and unavailable surfaces>
primary lens: <lens and why>
challenging lens: <lens and distinct contribution, or none>
model: <required lens artifact>
cruxes: <assumptions or observations that can change the conclusion>
terminal test: <prediction, threshold, counterexample, or update/kill trigger>
status: <method-specific terminal state justified by evidence and preferences>
decision impact: <what changes now, if anything>
unknowns: <material unresolved items>

Adapt the presentation to the problem rather than forcing a table when prose or a small diagram is clearer. Keep the fields and epistemic distinctions even when the rendering changes.

5. Stop honestly

Stop when the chosen lens no longer changes the model, evidence need, or decision. Return unidentified when accessible evidence cannot distinguish the live models. Return not applicable when the lens's required units, boundary, or falsifier cannot be defined. Do not synthesize a compromise merely because two lenses disagree; expose the disagreement and the evidence that would settle it.

Hand off without expanding scope

Answer the user's requested decision or analysis at the end, preserving the strongest counterargument and remaining uncertainty. Do not append an implementation plan, create an artifact file, invoke another Dale skill, or start execution unless the user explicitly asks for that next action.