Research method
Is this actually new, buildable, and better?
Space → Reality → Power → Time
RESEARCH & ENGINEERING SYSTEMS
The
Gauntlet
10 tools I built for research, engineering, verification, and review.
Give the system a professional task. It identifies what must actually be established, selects the method that owns each obligation, executes the check, and keeps the result scoped to the evidence.
What I built.
Five specialist research methods. Five control and review systems. They can run independently or as one routed workflow.
Research Gems
MindFormal reasoning, proof, probability, counterexamples. SpaceLiterature, prior art, repositories, standards. RealityMethod synthesis after a verified constraint gap. PowerImplementation, execution, integration, verification. TimeBaselines, ablations, uncertainty, stop/go.Control & review
SoulResearch orchestration and obligation routing. MirrorAdaptive reasoning complement for the current task. GauntletProcess assurance and false-green detection. MeditateDecision preflight before consequential work. CouncilStructured independent review.Professional workflows.
The route changes with the job. The background is the same system view: task, active methods, evidence returning, release.
Formal claim
Can this be proved or refuted?
Mind → Council
Software release
The tests are green. What could still be wrong?
Power → Gauntlet → Council
Stalled project
What is the next experiment worth running?
Meditate → Soul → Time
Adaptive work
What part of this task is least covered?
Mirror → Relevant Gem
The five Gems.
Each one is a different evidence-producing mechanism. As you move through them, the system field reconfigures into that Gem and runs its actual conceptual workflow.
MIND
Formal Reasoning
Turn a claim into explicit objects, assumptions, a proof obligation, and a claim-native verifier.
- Use
- Theorems, derivations, formal specifications, probability, exact counterexamples.
- Returns
- Proof, measured result, cited theorem, counterexample, or unresolved state.
SPACE
Research Discovery
Search the evidence landscape before novelty, current-state, or reuse claims are made.
- Use
- Literature, prior art, current technical facts, standards, reusable software.
- Returns
- Verified source set, nearest established class, concrete differentiator, search limits.
REALITY
Method Synthesis
Only invent after the nearest known methods fail a specific, named constraint.
- Use
- New algorithms, mechanisms, or research methods after a verified gap.
- Returns
- Candidate mechanism, assumptions, invariants, failure modes, verifier plan.
POWER
Engineering Verification
Move software claims through the real implementation and the real entrypoint.
- Use
- Implementation, debugging, integration, regression, executable verification.
- Returns
- Executed checks, defect-class coverage, output hashes, untested limits.
TIME
Evaluation & Benchmarking
Make candidate and baseline compete under the comparison that actually answers the decision.
- Use
- Experiment design, matched baselines, ablations, cost analysis, stop/go.
- Returns
- Matched comparison, uncertainty, failures, exclusions, decision consequence.
One control model.
Frame the task. Create load-bearing obligations. Route each one to the method that owns it. Keep the receipt. Release only what the evidence supports.
Task→Obligations→Selected methods→Receipts→Scoped result
Complete toolkit.
The `SKILL.md` is the reasoning contract. The linked Python runtime supplies execution, state, receipts, hooks, or gates where that tool has mechanical work to do.
Download repositorySoul · Research Orchestrator/soulRoute complex work across the toolkit.
Mind · Formal Reasoning/mindProof, logic, probability, formalization.
Professional use
Theorem checking, mathematical derivations, formal specifications, probabilistic reasoning, and counterexample search.
How it works
Defines objects and assumptions, states the exact claim and negation, chooses a claim-native verifier, searches edge cases, and reports scope.
Space · Research Discovery/spaceLiterature, prior art, repositories, standards.
Professional use
Literature reviews, prior-art checks, technical due diligence, current-source verification, and reusable implementation discovery.
How it works
Translates the mechanism into search terms, expands across neighboring fields, inspects primary sources and repositories, and records source limits.
Reality · Method Synthesis/realityNew mechanisms after a verified gap.
Power · Engineering Verification/powerImplementation, debugging, integration, execution.
Professional use
Building or reviewing software, debugging failures, integration work, release checks, and verifying executable claims.
How it works
Inspects the real repository, maps edits to requirements, implements the smallest complete change, runs targeted and regression checks, and exercises the real entrypoint.
Time · Evaluation & Benchmarking/timeBaselines, ablations, uncertainty, stop/go.
Professional use
Experimental design, capability comparisons, matched baselines, ablations, cost analysis, and stop/go decisions.
How it works
Defines the construct and endpoint, selects a strong baseline, matches relevant budgets, preserves failures, reports uncertainty, and applies stop/go criteria.
Gauntlet · Process Assurance/gauntletAudit the process behind a result.
Meditate · Decision Preflightorchestrator-invokedGround consequential work before acting.
Professional use
High-cost decisions, unstable task frames, repeated failure, or moments when execution is moving faster than the evidence state.
How it works
STILL → GROUND → ORIENT → WEIGH → RELEASE: identify authoritative state, separate facts from assumptions, choose one high-value next action.
Council · Evidence Review Panel/councilStructured independent review.
Professional use
Important artifacts that benefit from independent correctness, measurement, implementation, novelty, or adversarial review.
How it works
Creates distinct review seats, freezes first-pass conclusions before reveal, cross-critiques them, normalizes scope, and compares against a direct control.
Mirror · Adaptive Reasoning Complement/foilFind the smallest missing complement for the task.
Professional use
Work where the right next method depends on what the current task and available evidence already cover.
How it works
Classifies the task first, keeps competing gap hypotheses, uses a minimal discriminator when needed, supplies the smallest useful complement, then verifies transfer.
Inspect the work.
Source, runtime contracts, reproducibility, security, governance, and research notes remain directly accessible.