piensalo

Check a deliverable against its actual consumer: who reads it, what they do next, what format and register they need, and what they will assert downstream on its basis. Activate when the deliverable has a named audience beyond the requester, or its text will be forwarded, announced, or acted on by someone absent from the conversation.

For any rule-like deliverable — code, regex, query, formula, validator, policy — enumerate the boundary inputs (empty, zero, one, max, duplicate, malformed, hostile) and actually exercise the most breakable three before shipping. Activate when the deliverable is executable or rule-like and will meet inputs the author didn't type.

State every claim at its actual evidence level — verified, inferred, assumed, guessed — and give credences only where an observable could score them; ban certainty language on unverified claims. Activate when a response makes claims about unobserved, future, or unverified state, or when the user asks how confident or likely something is.

Before intervening in a system, write the explicit causal chain from the proposed action to the intended outcome — every arrow marked verified or assumed — and check the weakest assumed arrow first. Activate when an intervention (fix, optimization, policy change, growth action) is proposed on a system whose mechanism has not been written down.

With multiple live hypotheses and a limited test budget, design the cheapest single test whose possible outcomes separate the hypotheses — with a results-to-verdict table committed before running anything. Activate when two or more competing hypotheses are on the table and the next step is choosing what test, experiment, or probe to run.

Enumerate every deliverable the request actually asked for and map each one to the work produced, so nothing asked-for is silently dropped, stubbed, or rounded up to done. Activate when a request contains multiple parts — an enumerated list, several 'and'/'also' clauses, or a spec with more than one required artifact — or before delivering a draft against such a request.

When inputs contain conflicting facts, stop and resolve the conflict explicitly — classify it, pick a resolution rule, and log what stays unresolved — before building anything on either side. Activate when two sources give different values for the same quantity, a document's claim disagrees with observed state, or the text itself flags inconsistency.

Before asserting a universal claim — always, never, all, none, guaranteed, impossible — spend one bounded search for a counterexample, starting at the edges of the quantifier's domain.

Before shipping, sweep for conditions that void the entire answer regardless of its quality — violated mandatory rules, missing required elements, forbidden resources, wrong output form. A hit is fixed, never caveated.

When enumerating causes, ideas, or hypotheses, generate candidates that are distinct causal mechanisms — not paraphrases of the symptom or of each other — including classes with no lexical footprint in the evidence.

Attack a proposal's feasibility before endorsing it: extract its binding requirements — resources, limits, permissions, dependencies, deadlines — and test each against known constraints. One hard infeasibility ends the endorsement.

Post-draft check that every explicit question and sub-request in the task is answered in the response, in the form the task asked for — nothing dropped, deflected, or answered in the wrong shape.

Extract every constraint buried in the task text — limits, exclusions, format rules, 'must/only/never' clauses — into an explicit list, and check the plan or draft against each one.

Identify which uncertain assumptions a conclusion actually rests on — the ones that flip the answer if wrong — state them explicitly, and attach the cheapest check to each. Activate when a recommendation, verdict, or go/no-go decision is being produced and at least one of its inputs is assumed or unverified rather than observed this session.

When a metric, KPI, or threshold drives a decision, check that the number measures the real objective before optimizing or killing anything against it. Diagnose the ruler first. Activate when a decision (fix, kill, ship, prioritize) is justified by a named metric, percentage, or threshold presented as ground truth.

Recover the objective behind a solution-shaped or vague request before executing it literally. Activate when a request names a solution or knob rather than an outcome ('increase the timeout'), or uses a vague referent ('fix it', 'make this better') with no stated goal. Single intervention: state the interpreted objective in one line, check the literal request against it, flag any divergence.

Independently recompute every derived quantity in the work — sums, percentages, ratios, deltas, projections — from stated inputs, by a different route than the one that produced it, before asserting it. Activate when the draft or its inputs contain computed numbers that the answer asserts or builds a decision on.

Before endorsing a change, project the effects of the effects: who or what adapts to the first-order outcome, and what that adaptation breaks — especially neighbors depending on current behavior. Activate when a change with downstream dependents is proposed (policy, pricing, migration, deprecation, incentive, default-change, quota, rollout) and the analysis so far lists only direct effects.

Trace each load-bearing claim to its primary source before believing it — especially self-labeled causes in error messages, logs, vendor explanations, and second-hand summaries. Activate when a diagnosis or decision is about to rest on a claim whose only support is a label, a log line, a doc/vendor statement, or someone's summary of a source you can open yourself.

Software build-and-debug workflow: symptom capture, minimal reproduction, ranked hypotheses, discriminating tests, minimal in-scope fix, regression proof with quoted output. Use when code is broken, failing, or behaving unexpectedly, or when implementing a change that must not regress anything. Trigger phrases: 'fix this bug', 'tests are failing', 'why is this crashing', 'implement X without breaking Y'.

Mathematical and quantitative reasoning workflow: formalize the claim, compute small cases honestly, conjecture, then prove or refute via independent derivation routes and boundary probes. Use for proofs, counting, probability, closed forms, numeric estimates, or any task where a pattern-extension 'proof' would be tempting. Trigger phrases: 'prove', 'compute', 'closed form', 'what are the odds', 'derive'.

Domain-general deep problem-solving loop distilled from curated expert reasoning traces: objective recovery, constraint and contradiction extraction, distinct mechanisms, cheapest discriminating test, adversarial verification, calibrated synthesis. Use when a task is non-trivial and a shallow first answer is risky - hard analysis, debugging, design, or any question where being confidently wrong is expensive. Trigger phrases: 'think this through', 'get this right', 'hard problem', or any multi-constraint ask.

Independent verification of a candidate answer against domain verifier criteria: deterministic checks first, adversarial probes, evidence-level grading, disqualifier sweep, per-criterion verdict table. Use when asked to check, grade, review, or attack a finished answer or artifact - yours or another agent's - before it is trusted or shipped. Trigger phrases: 'verify this', 'is this answer right', 'review before we ship', 'grade this output'.

Invention and mechanism-design program: extract the physics-level requirements, search prior art, generate mechanism candidates that differ in operating principle, attack feasibility with numbers, and converge on a buildable design. Use for 'design a device/mechanism/protocol that does X', novelty-sensitive proposals, and feasibility reviews of inventive claims. Trigger phrases: 'invent', 'design a mechanism', 'is this feasible', 'novel approach'.

Research and evidence-synthesis program: define the question, gather and grade sources, build a contradiction map, verify load-bearing claims at the claim level, and deliver a verdict with graded evidence. Use for 'what do we know about X', reconciling conflicting numbers or reports, literature or market questions, and any answer that must be assembled from multiple sources. Trigger phrases: 'research', 'reconcile these figures', 'what does the evidence say'.

Strategy and decision program: recover the real objective, enumerate constraints including contractual and irreversibility traps, generate genuinely distinct options, scenario-score them, attack the leader, and recommend with explicit kill conditions. Use for 'should we do X', pricing/roadmap/build-vs-buy calls, responses to competitor moves, and any decision under uncertainty with real downside. Trigger phrases: 'recommend', 'should we', 'what's the right move'.

Writing and content-creation program: build the claim ledger and constraint checklist first, draft divergently, run separate argument/accuracy/style critic passes, compress without losing ledger items. Use for high-stakes prose - announcements, notices, abstracts, memos, docs - where every claim must be defensible and omissions are costly. Trigger phrases: 'draft', 'write the announcement/memo/notice', 'tighten this'.