Files · code · data · local research objects
APPARATUS
Tools that keep research inspectable, testable, and transferable.
A paper can show the result. Serious research also needs a durable record of how that result was reached.
Boundary First Labs builds tools for tracking questions, sources, experiments, claims, evidence, criticism, failures, revisions, and responsibility. The goal is simple: another person should be able to see what happened, challenge it, and continue the work without depending on hidden context.
DESIGN POSTUREHuman-readable. Machine-checkable. Open to correction. Transferable without hidden dependence.
AI can propose. Tools can check. Evidence decides.
Separate sources of truth, connected deliberately.
The Lab does not force every kind of research state into one master database. Papers, experiments, products, source records, and operations keep their own responsible homes, while shared directories and links make them easier to find.
Tools can search, compare, validate, transform, and summarize that material, but using a tool does not give the tool authority to declare a scientific claim true or make a consequential decision on its own.
The Lab machine is federated rather than monolithic.
Where does authority live when registries, transforms, instruments, evidence, and public pages all touch the same work?
RL · PROD · EXP · PUB · ATLAS · XFORM · EVENT · MACH
Addresses · classes · owners · lifecycle · authority
Typed edges · adapters · preservation / loss contracts
Queues · validators · Corpus Forge · Observatory · Workbench
Claims · sources · events · receipts · revision state
Snapshot · Research · Apparatus · Publications · Open Lab
- Source state→Durable identitiesassigns addressable continuity
- Identities→Registries / contractsdeclares ownership and authority
- Contracts→Relations / transformsconstrains connection
- Transforms→Operators / instrumentssupports bounded operation
- Operations→Evidence / historyleaves witnesses and receipts
- Evidence / history→Public projectionssupports bounded rendering
Make the work easy to inspect without making the machinery the authority.
TECHNICAL MACHINERY DIRECTORY
What tools exist, what they do, and where they stop.
human_reviewed_reconciliation_v0_2
- LAB REVISION
- 5dafabd14f89
- REGISTRY DATE
- 2026-09-23
Discovery, routing, interoperability planning, and integration status only. This registry does not supersede component-local contracts or confer scientific, publication, institutional, or execution authority.
Inspect canonical machinery registryAgent Control
organized_library_curated/999_Library/04_Operations/01_daily_operations__operations/Agent_Control/- document
organized_library_curated/999_Library/04_Operations/01_daily_operations__operations/Agent_Control/00_AGENT_CONTROL.md - script
organized_library_curated/999_Library/04_Operations/01_daily_operations__operations/Agent_Control/runtime/bfl_agent_control/__main__.py - document
organized_library_curated/999_Library/04_Operations/01_daily_operations__operations/Agent_Control/agent_registry.yaml
None declared in the registry.
Bounded Agent Control admission/routing/scheduling/dispatch/execution-evidence/verification authority under explicit packet, transaction, capability, scope, budget, cadence, and human-gate contracts. A4 promotion and A5 external consequence remain human-authorized; Agent Control may not self-admit arbitrary work or silently enlarge scope.
Adapt Agent Control run/verification/cadence evidence into the common Machinery Event envelope and include its package-local manifest in ordinary generated-registry validation/refresh.
Agentic Artifact Protocol
organized_library_curated/999_Library/04_Operations/02_pipeline_tooling__operations/agentic_artifact_protocol/- document
organized_library_curated/999_Library/04_Operations/02_pipeline_tooling__operations/agentic_artifact_protocol/README.md - script
organized_library_curated/999_Library/04_Operations/02_pipeline_tooling__operations/agentic_artifact_protocol/validate_agentic_artifacts.py - script
organized_library_curated/999_Library/04_Operations/02_pipeline_tooling__operations/agentic_artifact_protocol/build_agentic_artifact_registry.py
None declared in the registry.
Representation and execution contract for explicitly opted-in research objects only; passing validation or successful execution cannot promote claims, publication status, canonical source, or institutional stage.
Expose Agentic Artifact validation/registry operations through the common machinery registry while retaining package-local agentic manifests as execution authority.
Architecture Observatory
organized_library_curated/999_Library/04_Operations/02_pipeline_tooling__operations/01_control_and_architecture/diagrams/lab_architecture_atlas_mermaid_v0_1/observatory_v0_1/- document
organized_library_curated/999_Library/04_Operations/02_pipeline_tooling__operations/01_control_and_architecture/diagrams/lab_architecture_atlas_mermaid_v0_1/observatory_v0_1/viewer/README.md - script
organized_library_curated/999_Library/04_Operations/02_pipeline_tooling__operations/01_control_and_architecture/diagrams/lab_architecture_atlas_mermaid_v0_1/observatory_v0_1/viewer/serve.mjs - script
organized_library_curated/999_Library/04_Operations/02_pipeline_tooling__operations/01_control_and_architecture/diagrams/lab_architecture_atlas_mermaid_v0_1/observatory_v0_1/viewer/scripts/build_catalog.mjs
None declared in the registry.
Read/derive-only architecture inspection and representation-analysis authority. The Observatory may build derived navigation catalogs, route questions, compare declared views, expose gaps, and prepare projection requests; it may not create source truth, durable identities, KTG relations, machine admission, promotion authority, deployment state, or scientific validation.
Enroll the Observatory in common machinery health/refresh and complete browser-level verification before considering a higher integration level.
BLF Translation Program
organized_library_curated/999_Library/04_Operations/02_pipeline_tooling__operations/blf_translation_program__operations/- document
organized_library_curated/999_Library/04_Operations/02_pipeline_tooling__operations/blf_translation_program__operations/
- Publication Mine
Controlled representation change across audiences/artifact types while preserving source relationships and claim ceilings; translation does not change source authority.
Review operation-specific contracts and add explicit representation-transport adapters.
Structural and Semantic Control Plane
organized_library_curated/CONTROL_PLANE.md- document
organized_library_curated/CONTROL_PLANE.md - command
python helper_scripts/bfi_run.py --run refresh_control_plane - health_surface
organized_library_curated/00_CONTROL_PLANE_HEALTH.md
None declared in the registry.
Routing, generated measurement, derived projection, freshness, and bounded validation authority only; no source, claim, publication, or institutional promotion authority.
Add package/local machinery manifest representation and enroll machinery registry generation/validation in refresh_control_plane after schema calibration.
Corpus Forge
organized_library_curated/999_Library/04_Operations/02_pipeline_tooling__operations/corpus_forge/- document
organized_library_curated/999_Library/04_Operations/02_pipeline_tooling__operations/corpus_forge/corpus_forge_protocol.md - document
organized_library_curated/999_Library/04_Operations/02_pipeline_tooling__operations/corpus_forge/corpus_forge_job_packet_template.md
- Timeline
- Publication Mine
- Research Lane Register
- Experiment Register
- Atlas of Atlases
Source-bounded research refinement, criticism, evidence production, and candidate repair. Operator and critic output cannot cross the Human Promotion Gate automatically.
Create Machinery Event adapter and first Corpus Forge -> durable projection candidate vertical slice.
May emit candidate projection updates when a governed operation materially changes a durable cross-cutting object; acceptance remains with the relevant register/source authority.
Intake Audit and Commit Closure
organized_library_curated/998_Intake/- workflow
audit -> adjudicate -> commit -> refresh -> re-audit - document
organized_library_curated/998_Intake/active_handoff.md - script
organized_library_curated/999_Library/04_Operations/02_pipeline_tooling__operations/helper_scripts/audit_intake.py - script
organized_library_curated/999_Library/04_Operations/02_pipeline_tooling__operations/helper_scripts/commit_intake.py
None declared in the registry.
Bounded ingestion transaction after explicit adjudication; must preserve source hashes, path containment, collision checks, append-only commit evidence, and post-mutation refresh. Does not promote research claims.
Define Machinery Event adapter for audit/adjudication/commit lineage and candidate durable-projection updates.
Management Control
organized_library_curated/999_Library/04_Operations/02_pipeline_tooling__operations/management_control/- document
organized_library_curated/999_Library/04_Operations/02_pipeline_tooling__operations/management_control/
None declared in the registry.
Cross-Lab management routing and rollup only; generated management projections do not silently reprioritize, close, or supersede domain-local work authority.
Verify current executable entrypoints and enroll capability/authority semantics in the common manifest.
Publication Control and Paper Mine
organized_library_curated/999_Library/04_Operations/01_daily_operations__operations/publication_graph/- document
organized_library_curated/999_Library/04_Operations/01_daily_operations__operations/publication_graph/
- Publication Mine
Publication discovery, dependency, measurement, queue, projection, snapshot, and history state. Generated outputs do not silently promote publication or scientific status.
Bind publication events/projections to common Machinery Event identity without duplicating publication source authority.
Boundary First Executable Sandbox / Workbench Pilot
branch:agent/boundary-first-sandbox- workflow
PR #270 / branch agent/boundary-first-sandbox
None declared in the registry.
Prototype direct-manipulation/executable representation surface only; does not become institutional source authority by visualization or execution.
ADAPT: consume the common machinery registry and runtime authority/event contracts rather than defining a parallel institutional control plane.
Supporting Capture Protocol
organized_library_curated/999_Library/04_Operations/02_pipeline_tooling__operations/supporting_capture_protocol/- document
organized_library_curated/999_Library/04_Operations/02_pipeline_tooling__operations/supporting_capture_protocol/
- Timeline
Provenance and integrity only; no execution authority and no promotion of historical/supporting material into current theory.
Add manifest after verifying current validator/generator entrypoints and define provenance-event adapter only where useful.
UX Pattern Atlas Recipe Admission Gate Pilot
branch:agent/bfux-pattern-atlas-v0-1-2026-09-04- workflow
PR #268 / branch agent/bfux-pattern-atlas-v0-1-2026-09-04
None declared in the registry.
Scoped recipe/runtime admission prototype. Readiness is evidence, not authorization; explicit admission and revocation are distinct authority events.
COMPOSE: extract reusable admission/runtime-authorization semantics into the common runtime authority federation while preserving UX Pattern Atlas domain ownership.
WHAT THE TOOLS ARE FOR
Different tools, different jobs, clear limits.
Lab Registry Registrar
What durable records exist, where do they live, and which source is responsible for each one?
Acts as a directory for the Lab's registers, ledgers, catalogs, queues, maps, and indexes without taking ownership of the records inside them.
Which record systems exist, where their source of truth lives, who owns them, and how they are maintained.
A search index, generated page, or convenience view being mistaken for the source of truth.
May describe and route people or software to the Lab's record systems.
May not rewrite the underlying research, product, publication, or operational records simply because it can find them.
Research Lane Register
What durable questions is the Lab pursuing?
Keeps a long-running research question identifiable across experiments, conversations, papers, failures, revisions, and changes of direction.
Governing question, inquiry boundary, posture, sources, experiments, publications, dependencies, lineage, and next targets.
A long-running inquiry collapsing into disconnected papers, chats, or short-term memory.
May record the identity, history, and current state of a research program.
Being registered does not make the hypothesis correct.
Experiment Register
What has the Lab actually tried?
Keeps a durable record of experiments, simulations, benchmarks, replications, stress tests, attempts to disprove a claim, and negative or inconclusive results.
Bounded evidence-bearing operations and their outcomes.
Failed or inconvenient experiments disappearing from the institutional record.
May record what was tried and connect the result to the relevant research.
Recording an experiment does not make a scientific claim established.
Claim-control ledgers
What exactly are we asserting?
Keeps the status of a claim separate from polished prose: what is supported, still hypothetical, under test, disputed, revised, or superseded.
What a claim currently says, what supports it, what challenges it, and how confident the owning research program is allowed to be.
Confident writing from sounding stronger than the underlying evidence.
May record claim status inside the research object that owns that claim.
There is no single Lab-wide claim scale that overrides the standards of every field.
Source & provenance registers
Where did this come from?
Tracks the literature, data, software, standards, historical material, conversations, and criticism that a piece of research depends on.
Where material came from, how it was used, and how it relates to the current work.
The Lab's interpretation from being confused with what the original source actually said.
May record source history and how material entered the work.
A source record does not prove that every interpretation of that source is correct.
Corpus Forge
How can software and AI help turn source material into research that a person can review?
Helps organize, compare, criticize, revise, and package clearly scoped research while preserving where it came from and what still requires human judgment.
The assigned task, source material, claim-and-evidence relationships, criticism, comparison results, revision history, and review state.
The shortcut 'an AI wrote it, therefore the Lab accepts it.'
May perform clearly scoped research operations such as organizing, comparing, criticizing, and proposing repairs.
Tool output, critic agreement, or a passing validator never makes a scientific claim official on its own.
Research Deployment Packet
How does research leave the environment in which it was developed?
Packages a clearly scoped body of research so another person, laboratory, organization, or machine can understand and continue it without reconstructing the original conversation.
The current editable work, claims, sources, context, open questions, next steps, and file inventory.
Research becoming inseparable from one chat, one computer, or one researcher.
May package and transfer the declared research record.
A complete packet does not make the claims inside it true.
Lab Machinery Registry
What executable or operational capabilities exist?
Keeps a technical directory of the Lab's software and operational tools: where each one lives, what it does, how mature it is, and what limits apply.
Which tools exist and how they fit into the Lab's operating system.
Knowing that a tool exists from being confused with permission to run it or trust its output automatically.
May support discovery, routing, integration planning, and technical maintenance.
Being listed does not give a tool permission to run or decide scientific questions.
Formal Metrology
What important features of a reasoning system can we make visible without pretending everything has one meaningful score?
A working program for measuring and inspecting things such as traceability, coverage, information loss, revision history, responsibility, repair paths, and whether a result can be challenged.
Whether claims can be traced, dependencies can be seen, information was lost, revisions are current, responsibilities are clear, and repair paths exist.
Important structural questions being ignored simply because they do not collapse into one number.
May trace, classify, compare, and measure where a meaningful measurement is available.
Not every important property is numeric, and this is not a universal score of research quality.
APPARATUS CONTEXT
How the Lab keeps tools useful without letting tools become the authority.
WHY APPARATUS MATTERS
Good ideas can fail because their machinery is opaque.
Apparatus makes research state inspectable instead of leaving it in memory, prose, or disconnected folders.
FEDERATED ARCHITECTURE
There is no single master database for all research.
Different kinds of work keep their own source records, and shared tools connect them without taking ownership.
HOW IT WORKS TOGETHER
A simplified research path.
Each step does its own job; no tool or record silently gains the decision rights of the next step.
HUMAN GATES
Human review does not mean avoiding automation.
Automation can do substantial work while consequential decisions remain with the people responsible for them.
PUBLICATION MODEL
A paper is the readable result, not the entire research record.
The article can remain the front door while supporting records preserve how the result was produced, challenged, and revised.
PUBLIC EXPOSURE
Transparency still needs boundaries.
Being inspectable does not mean publishing private, sensitive, security-relevant, or poorly contextualized material.
DESIGN POSTURE
Every tool should answer six questions quickly.
A reader should be able to tell what a tool tracks, why it exists, what decisions it may make, what decisions remain human, and who owns it after handoff.
TRUST + STEWARDSHIP
No tool gets to declare its own success.
Passing a check, becoming visible, being published, or looking complete never removes the need for responsible review, maintenance, and correction.
CLOSING TEST
Research becomes durable when another person can pick it up.
The Lab tries to preserve questions, sources, failed attempts, criticism, revisions, responsibility, and next steps—not only conclusions.