Hypotheses, observations, comparison, replication, criticism, and provisional rather than absolute closure.
Products
OBJECTProduct
PUBLIC STATUSRESEARCH PRODUCT · OPERATIONAL SUITE V0.1
FAMILYRESEARCH METHOD / AGENTIC INFRASTRUCTURE
Agentic Scientific Method
Make the inquiry itself inspectable.
An operational research protocol that makes goals, boundaries, state, evidence, criticism, repair, authority, and closure explicit enough for humans and agents to work together.
CURRENT PRODUCT BOUNDARYThe protocol, authority model, templates, validation ladder, and transition contract exist. External user pilots, comparative benchmarks, cross-domain replication, and formal validation remain open.
INQUIRY MACHINE · CONTROLLED TRANSITION VIEWScientific work as inspectable state change.
09 · COMPAREAnalyst
OPERATIONAL QUESTION
How do represented and observed states compare under the declared correspondence rule?
DURABLE ARTIFACTcorrespondence_report.md
FROMTrace
→CURRENTCompare
→TODiagnose
CONTROL RULECapability is not authority, and authority is not evidence.
WHAT THE METHOD ADDS
Make the research process explicit enough to inspect, criticize, and continue.
A research instruction like “form a hypothesis and test it” hides many practical decisions: what question is being asked, what is in scope, what alternatives matter, what would distinguish them, who may change the test, and what counts as a mismatch.Bind. Map. Select. Act. Trace. Compare. Diagnose. Repair. Close. Preserve.
Explicit objectives, state spaces, action options, planning under uncertainty, memory, stopping rules, and model updating.
Declare what matters before optimizing, expose omitted distinctions, preserve invariants, localize mismatch, and keep a repair path.
Stabilize sources, track claim status, preserve external memory, require critic passes, and separate work from promotion.
BOUND→LOAD→TRACE→TEST→REPAIR→CLOSE
MODEL VS. OBSERVATION
Write down what the model says, then compare it with what actually happened.
An ASM inquiry keeps the model and the observation record distinct. One describes what the current explanation expects or allows; the other records what actually happened under the declared test conditions.MODEL / EXPECTATIONMODEL
S0
→S1
→S2
→S3
States, relations, transitions, constraints, invariants, uncertainty, and predicted consequences declared before the outcome is known.
CORRESPONDENCECOMPAREMISMATCH?
OBSERVATION RECORDTRACE
O0
→O1
→O?
→O3
Raw and transformed traces, measurements, annotations, and registered consequences—kept distinct from the interpretation later placed on them.
FIND THE SOURCE OF THE MISMATCH
A failed prediction does not automatically tell you what was wrong.
It may be the hypothesis. It may also be the boundary, instrument, action, correspondence rule, or closure criterion. ASM treats localization as a separate scientific task before repair begins.
01Purpose02Boundary03State distinctions04Transition grammar05Invariant06Candidate model07Action selection08Intervention fidelity09Instrument / observer10Trace transformation11Comparison rule12Closure criterion13Authority
HUMANS, AI, AND RESPONSIBILITY
Separate who performs a task from who is responsible for approving the result.
Being able to perform a task does not grant decision authority, and approval does not make a claim true.
Owns purpose, consequence boundary, resources, stopping authority, and accountability.
May propose mappings, models, searches, experiments, and repairs; may not promote its own consequential claims.
Runs frozen symbolic, numerical, search, benchmark, theorem-prover, test, or pipeline operations.
Records raw or minimally transformed traces without silently changing the action after seeing results.
Attempts to break the object with counterexamples, adversarial cases, ablations, perturbations, and dependency attacks.
Audits the whole bounded inquiry, including provenance, compression, repair history, and requested closure.
Causes consequential transitions such as canonical replacement, public claims, publication, deployment, or authoritative merge.
GENERATORPropose
→EXECUTORRun
→OBSERVERRecord
→FALSIFIERAttack
→CRITICAudit
→HUMAN AUTHORITYPromote / withhold
A critic can find the work sound enough for the next review state without becoming the authority that turns it into a public claim, canonical replacement, deployment, or other consequence-bearing transition.
DURABLE RESEARCH MEMORY
Important research decisions should survive beyond one chat, person, or tool.
Each phase emits an inspectable artifact. The point is not paperwork for its own sake. It is to let another person reconstruct what was believed, what was tested, what changed, who authorized it, and where the next repair must begin.01
Inquiry charterPurpose, trigger, affected parties, accountable owner, exclusions, and stop conditions.
02
Boundary declarationIncluded and excluded entities, scales, sources, interfaces, exchanges, and boundary-failure signals.
03
State-space mapEntities, states, relations, temporal/modal states, uncertainties, and candidate transitions.
04
Candidate registerAlternatives, null baselines, predictions, assumptions, and likely failure points.
05
Action + run recordWhy this test, which frozen specification, which tools, versions, parameters, operators, and deviations.
06
Observation traceRaw observations separated from transformed data, annotations, interpretations, and agent inference.
07
Correspondence + defect reportExpected versus observed behavior, mismatch localization, and competing explanations.
08
Repair lineageWhat changed, why, which claims moved, what new degrees of freedom appeared, and what must be retested.
09
Closure + provenancePermitted claim, remaining uncertainty, reopening conditions, authority, critic findings, and preserved history.
FIRST OPERATIONAL RUN · SELF-HOSTING
Use the method to criticize and repair the method.
The first bounded ASM run applies the method to itself: stabilize sources, extract claims, compare prior art, run an independent critic, preserve defects, repair v0.1, and promote only what survives.
NON-NEGOTIABLENo self-approval.
HOW THE METHOD SHOULD BE TESTED
The immediate question is whether it improves real research work.
Does the protocol improve bounded inquiry compared with the baseline practice?
Stable definition, separated vocabulary, explicit claim ceiling, no unresolved contradiction in the core protocol.
Use ASM to criticize and improve its own specification without allowing self-approval.
Apply the method across existing Lab work such as software, knowledge systems, games, audits, or formal experiments.
Compare ad hoc practice, classical hypothesis-test summaries, and ASM on defect discovery, provenance, rework, and reviewer agreement.
Give the protocol to an independent user and measure learnability, ambiguity, completion, misuse, burden, and artifact quality.
Demonstrate useful transfer across discrete, continuous/noisy, and consequence-bearing human systems.
Test state reachability, permissions, closure gates, repair transitions, provenance integrity, and unsafe self-promotion.
The operational suite, governance contracts, and machine-readable transition model are implementation evidence. They are not yet an executed, externally validated scientific research machine.
Operational protocol existsAuthority state machine existsSelf-hosting run definedInternal benchmark portfolio definedExternal pilot still neededFormal / cross-domain validation still needed
NEXT REAL-WORLD TEST