The scientific method as a universal problem-solving algorithm — goal-first, hypothesis-plural, falsifiable experiments, honest measurement.
Ask a generic AI to 'figure out why this is broken' and it guesses. One hypothesis, no experiment design, no measurement criteria — just pattern-matching toward a plausible answer. When the first guess is wrong, it guesses again with slight variation. There's no feedback loop, no falsification, no honest accounting of what the data actually showed. You get confident-sounding reasoning that collapses the moment you run it against reality.
The skill embeds the scientific method as a literal execution loop: DefineGoal first (success criteria before any action), then GenerateHypotheses with a hard floor of three candidates (single-hypothesis work is confirmation bias by construction), DesignExperiment to make tests falsifiable, MeasureResults to collect only goal-relevant data, AnalyzeResults against the original success criteria, then Iterate back to the hypothesis phase. QuickDiagnosis applies this cycle in 15-minute bursts for fast debugging. StructuredInvestigation runs the full multi-factor version for complex problems. The skill integrates Council for hypothesis generation, Evals for structured measurement, and RedTeam for stress-testing conclusions before you commit to them.
What you say to your DA, and what the Science skill actually does.
The thinking, frameworks, and architecture that distinguish this skill from a generic version of the same task.
Applies the scientific method as a general problem-solving algorithm: define the goal first, generate multiple hypotheses, design experiments that can fail, measure honestly, analyze against the goal, iterate. Seven core workflows plus two diagnostic shortcuts (quick 15-minute debugging and structured multi-factor investigation). It scales from micro (TDD) to meso (feature validation) to macro (MVP launch).
Most problem-solving is guessing dressed up as work. You pick the first idea that comes to mind, change something, and call it done when it "seems better" — which is confirmation bias, not progress. Without a clear definition of success you can't tell whether a change helped, so you keep tweaking forever or stop too early. Single-hypothesis thinking means you only ever test the idea you already believed. This skill forces the discipline that fixes all of that: a stated goal, at least three competing hypotheses, falsifiable tests, and measurement that compares to the goal rather than to your hopes.
The whole thing is one repeating cycle, and the goal anchors it — without clear success criteria you cannot judge results:
GOAL -----> What does success look like?
|
OBSERVE --> What is the current state?
|
HYPOTHESIZE -> What might work? (Generate MULTIPLE)
|
EXPERIMENT -> Design and run the test
|
MEASURE --> What happened? (Data collection)
|
ANALYZE --> How does it compare to the goal?
|
ITERATE --> Adjust hypothesis and repeat
|
+------> Back to HYPOTHESIZE
The answer emerges from the cycle, not from guessing.
| Resource | Description |
|---|---|
METHODOLOGY.md |
Deep dive into each phase |
Protocol.md |
How skills implement Science |
Templates.md |
Goal, Hypothesis, Experiment, Results templates |
Examples.md |
Worked examples across scales |
| Domain | Manifestation | Related Skill |
|---|---|---|
| Coding | TDD (Red-Green-Refactor) | Development |
| Products | MVP -> Measure -> Iterate | Development |
| Research | Question -> Study -> Analyze | Research |
| Prompts | Prompt -> Eval -> Iterate | Evals |
| Decisions | Options -> Council -> Choose | Council |
| Level | Cycle Time | Example |
|---|---|---|
| Micro | Minutes | TDD: test, code, refactor |
| Meso | Hours-Days | Feature: spec, implement, validate |
| Macro | Weeks-Months | Product: MVP, launch, measure PMF |
| Phase | Skills to Invoke |
|---|---|
| Goal | Council for validation |
| Observe | Research for context |
| Hypothesize | Council for ideas, RedTeam for stress-test |
| Experiment | Development (Worktrees) for parallel tests |
| Measure | Evals for structured measurement |
| Analyze | Council for multi-perspective analysis |
| Bad | Good |
|---|---|
| "Make it better" | "Reduce load time from 3s to 1s" |
| "I think X will work" | "Here are 3 approaches: X, Y, Z" |
| "Prove I'm right" | "Design test that could disprove" |
| "Pretend failure didn't happen" | "What did we learn?" |
| "Keep experimenting forever" | "Ship and learn from production" |
The answer emerges from the cycle, not from guessing.
Example 1: Quick diagnosis
User: "figure out why Surface time filters show stale items"
→ Quick diagnosis mode
→ Hypothesis: timestamp format mismatch in D1
→ Test: query D1 for actual stored format
→ Analyze: compare stored vs expected format
→ Result: ISO string vs Unix timestamp mismatch
Example 2: Full systematic investigation
User: "experiment with different prompt structures for better output"
→ Full cycle mode
→ 3+ hypotheses generated
→ Controlled experiments with measurements
→ Analysis identifies winning approach
→ Iterates until convergence
Workflows/`Workflows/DefineGoal.md`.md define the goal, what are we trying to achieve
Workflows/`Workflows/GenerateHypotheses.md`.md what might work, ideas, hypotheses
Workflows/`Workflows/DesignExperiment.md`.md how do we test, experiment design
Workflows/`Workflows/MeasureResults.md`.md what happened, measure, results
Workflows/`Workflows/AnalyzeResults.md`.md analyze, compare to goal
Workflows/`Workflows/Iterate.md`.md iterate, try again, next cycle
Workflows/`Workflows/FullCycle.md`.md Full structured cycle
Workflows/`Workflows/QuickDiagnosis.md`.md Quick debugging (15-min rule)
Workflows/`Workflows/StructuredInvestigation.md`.md Complex investigation
Say any of these to your DA and PAI activates the Science skill automatically:
Or invoke explicitly:
Skill("Science")Auxiliary files the skill loads at runtime — frameworks, guides, configs.
The thinkers, books, frameworks, and research this skill is built on. The ideas belong to them — the integration belongs to PAI.
Install PAI on your machine — your DA gets the Science skill plus 44 others, all hooked into one Life OS.