Research Payload Family
Research Payload Family shared core v0.1.0
Portfolio map for promoting the next research instruments from observed decisions and evidence needs, with shared companions, transfer limits, and explicit promotion triggers.
v0.1.1
Promote a new payload when a recurring decision needs a reusable path from parameters to evidence, interpretation, completion, and handback. This portfolio keeps the next instruments visible without turning every useful question into a new system.
Portfolio purpose
The Research Atlas already contains deep reports, executable research payloads, worked examples, and source syntheses. This page describes the next useful instrument horizon and the evidence that would justify promotion.
A candidate earns a full payload when several conditions converge:
- the receiving decision recurs across teams or domains;
- the source hierarchy can be made explicit;
- a shared method can travel while local facts remain local;
- an output contract can be evaluated;
- a stopping condition can be observed;
- one worked specimen exposes useful boundary behavior;
- another collaborator can run the instrument without reconstructing its intent.
A modular annex or parameter pack is the better form when a method strengthens several payloads while retaining little domain-specific research of its own.
Operating family
Instrument | Current role | Primary use |
Shared core | Common identity, evidence, output, completion, and run-receipt contract | |
Parameter pack | Choose faithful primitives, relationships, decision forms, state, and durable representation | |
Modular annex | Connect material claims to behavior, faults, oracles, recovery, and coverage | |
Executable payload | Move from dated US insurance evidence to a bounded local workflow test | |
Executable payload | Separate investment, operating value, commercial cohorts, cash, evidence, and capital gates |
These instruments establish the current reusable base. New candidates should reuse the core and the fitting companions rather than repeat them.
Promotion horizon
Priority 0 — immediate, cross-cutting decisions
Candidate | Decision served | Evidence classes | Intended form | Transfer limit | Promotion trigger |
What should be measured before release, in use, across change, and at a stop or rollback threshold? | System traces, task evaluations, human review, incident and recourse evidence, population and environment measures, current standards | Executable research payload | A generic monitoring taxonomy cannot establish the metrics, thresholds, or harms of a named deployment | One completed cross-layer specimen plus a run receipt that distinguishes pre-release evaluation from deployed monitoring | |
What evidence supports selection, contract terms, acceptance, renewal, transition, or exit? | Requirements, supplier claims, independent tests, contract language, data and model documentation, service evidence, portability rehearsal | Executable research payload | Public procurement guidance and vendor documents require adaptation to the named organization, jurisdiction, contract, and risk owner | One complete procurement-to-exit specimen with evidence requests, acceptance tests, change clauses, and an exit rehearsal |
Current P0 dry-run standing
- AI monitoring specimen and run receipt: the synthetic run established cross-horizon control flow, exposed source-version trace gaps, and produced focused v0.1.1 refinements. Candidate standing continues; an authorized real-world specimen and fitting review form the next promotion observation.
- Vendor-evidence specimen and run receipt: the synthetic run carried three routes through acceptance, change, renewal, portability, and deletion, then produced focused v0.1.1 refinements. Candidate standing continues; an instrumented run with accountable multidisciplinary review forms the next promotion observation.
These first receipts demonstrate that a completed dry run and payload promotion are distinct decisions. A useful run may sharpen a candidate while keeping the next evidence horizon explicit.
Priority 1 — organization and information boundaries
Candidate | Decision served | Evidence classes | Intended form | Transfer limit | Promotion trigger |
Work Redesign, Hiring & Capability Development | Which tasks, roles, team interfaces, and learning investments should change as production reach expands unevenly? | Task analysis, workflow observation, workforce surveys, labor research, output and quality measures, worker experience | Executable research payload | Occupational exposure and survey evidence can orient a local study; local role design still needs direct observation and participation | One role-family specimen that connects task change, hiring criteria, team capability, quality, workload, and recourse |
AI Data, Retrieval, Privacy & Provenance Boundaries | Which sources may enter retrieval, training, generation, evaluation, publication, and durable records under which authority? | Data inventory, rights and consent, provenance, retention, access, retrieval tests, public/private boundaries, incident routes | Executable research payload | A source can be technically reachable while its allowed use, disclosure, or evidentiary standing remains distinct | One end-to-end source journey with public, restricted, expired, corrected, and conflicting cases |
Organizational Adoption, Learning & Change Evidence | Which conditions let a capability become safe, useful, revisable practice rather than an isolated efficiency? | Adoption surveys, workflow observation, training evidence, support demand, quality and productivity measures, trust and recourse | Executable research payload | Self-reported readiness and productivity are contextual signals; receiving organizations need observed local behavior and outcomes | One adoption tranche that follows purpose, capability, practice, measures, learning, and a return decision |
Priority 2 — regulated, accessible, and resilient operation
Candidate | Decision served | Evidence classes | Intended form | Transfer limit | Promotion trigger |
Regulated Scientific Operations & Model-Change Credibility | What evidence supports a model, data, workflow, or software change across intended use, lifecycle, review, and recovery? | Context of use, risk analysis, validation, data governance, model and software lifecycle, change control, audit evidence | Executable research payload | Regulator principles orient evidence design; product, jurisdiction, quality system, and accountable review define the actual obligation | One scientific workflow specimen with versioned evidence, change impact, independent review, and rollback |
Accessible Collaborative Systems & Acquisition Evidence | What demonstrates that a system is accessible in design, procurement, implementation, and continued use? | Requirements, ACR or VPAT claims, manual and automated testing, assistive-technology use, user evidence, contract acceptance | Executable research payload | Accessibility claims, conformance reports, and automated scans each carry bounded evidentiary value | One acquisition specimen that traces requirements through vendor evidence, hands-on validation, acceptance, and remediation |
Operational Resilience, Replay & Recovery | What evidence supports restart, retry, replay, rollback, reconciliation, and return after partial failure? | Failure injection, durable receipts, backup and restore, dependency behavior, observability, runbooks, recovery exercises | Modular annex | Recovery objectives and failure modes remain specific to the service, data, and operating environment | Repeated reuse across two payloads or systems plus one recovery exercise whose evidence changes a design decision |
Candidate detail
AI Evaluation, Field Testing & Post-Deployment Monitoring
Why now. NIST AI 800-4, published in March 2026, distinguishes controlled pre-deployment evaluation from the continuing need to observe real-world operation. It describes deployed monitoring methods and terminology as an active, still-developing field. That combination supports a payload whose value is a structured local monitoring decision rather than a universal checklist.
Source seeds
- NIST AI 800-4 — Challenges to the monitoring of deployed AI systems — final NIST report, published 2026-03-06.
- NIST AI Risk Management Framework and relevant evaluation guidance — current version to be bound at execution.
- Named domain regulation, organizational policy, incident history, and affected-population evidence — local and date-bound.
- The Registry’s evidence-bearing verification and administrative-AI instruments — method sources rather than external-fact authorities.
Required dimensions
- context of use;
- pre-release capability and failure evaluation;
- deployment conditions;
- population and subgroup effects;
- data, model, prompt, tool, and workflow change;
- monitoring signals and detection limits;
- human review and recourse;
- incident and near-miss handling;
- thresholds for continue, revise, restrict, pause, rollback, or retire;
- receipt and reassessment date.
Vendor Evidence, Procurement, Renewal & Exit Portability
Why now. AI capability increasingly arrives inside ordinary products, services, and platforms. The useful research object is the whole acquisition and operating relationship: problem definition, evidence requests, evaluation, contract terms, acceptance, change notification, monitoring, renewal, portability, and exit.
Source seeds
- UK Guidelines for AI Procurement — official guidance that frames procurement as multidisciplinary, data-aware, benefit- and risk-bearing work; current revision should be checked at execution.
- Section508.gov — Buy Accessible Products and Services — official US federal acquisition route connecting requirements, market research, vendor evidence, acceptance criteria, hands-on validation, and lifecycle testing.
- Named procurement rules, contracts, data-processing terms, sector requirements, and organizational authorities — local and jurisdiction-bound.
- Independent evaluation, migration rehearsal, and receiving-system evidence — execution-specific.
Required dimensions
- challenge and affected population;
- alternative routes, including process or non-AI options;
- data and rights boundary;
- supplier evidence request;
- model, system, and subcontractor change;
- accessibility and interoperability;
- service levels and incident routes;
- audit, retention, deletion, and export;
- price, usage, switching, and concentration risk;
- acceptance tests;
- renewal evidence;
- transition and exit rehearsal.
Work Redesign, Hiring & Capability Development
Why now. The ILO 2025 research brief uses task-level, expert, and model-assisted analysis and reports that exposure more often points toward job transformation than full redundancy. The UK AI Adoption Research, updated in February 2026 and based on 3,500 business interviews plus qualitative follow-up, reports uneven adoption and readiness, with skills, identified need, ethics, cost, and regulatory uncertainty among material factors. These sources justify a local task-and-team research instrument while preserving their population and method boundaries.
Required dimensions
- purpose and work outcome;
- task bundle rather than job title alone;
- existing expertise and tacit coordination;
- automation, assistance, review, and escalation;
- quality, safety, access, and workload;
- amplified production reach;
- hiring criteria and onboarding;
- team interfaces and role clarity;
- learning, supervision, and recovery;
- employee voice and affected-population input;
- local comparison and review date.
AI Data, Retrieval, Privacy & Provenance Boundaries
Why now. Retrieval, generation, evaluation, and publication can use the same source under different authorities. A reusable payload can keep technical reachability, allowed use, evidence standing, citation, retention, and disclosure distinct.
Required dimensions
- source identity and owner;
- acquisition route;
- consent, license, policy, and contractual authority;
- personal, confidential, regulated, or public status;
- allowed purpose;
- retrieval and transformation lineage;
- model and tool exposure;
- retention and deletion;
- correction and supersession;
- benchmark and evaluation separation;
- citation and public projection;
- incident, audit, and withdrawal route.
Organizational Adoption, Learning & Change Evidence
Why now. Implementation gains are uneven because capability depends on purpose, workflow, skills, authority, support, measurement, and the surrounding ecosystem. A payload can help distinguish tool availability, task use, operational adoption, and durable organizational capability.
Required dimensions
- change purpose;
- current workflow and constraints;
- affected groups;
- readiness and support;
- pilot boundary;
- measures spanning quality, productivity, burden, access, trust, and recovery;
- local learning;
- decision cadence;
- scale, revise, pause, or redirect criteria;
- ownership after the pilot.
Regulated Scientific Operations & Model-Change Credibility
Why now. FDA and EMA’s Guiding Principles of Good AI Practice in Drug Development emphasize human-centered design, risk, standards, clear context of use, multidisciplinary expertise, data governance, model development, performance assessment, lifecycle management, and clear essential information. A research payload can translate those source dimensions into an evidence plan while leaving regulatory interpretation and product decisions with accountable specialists.
Required dimensions
- context of use and intended purpose;
- data provenance and representativeness;
- model, software, and workflow versions;
- validation and performance boundaries;
- change classification;
- traceability;
- human review;
- audit and quality-system fit;
- deployment and lifecycle monitoring;
- correction, rollback, and retirement;
- regulatory and quality authority.
Accessible Collaborative Systems & Acquisition Evidence
Why now. Accessibility is both a design property and a procurement, acceptance, and operating responsibility. Official Section 508 acquisition guidance connects requirements, market research, vendor conformance reports, hands-on testing, contract acceptance, and continued testing. That makes accessibility an especially strong cross-functional payload candidate.
Required dimensions
- affected users and contexts;
- applicable requirements;
- semantic and interaction design;
- assistive-technology behavior;
- content, document, media, and data-visualization alternatives;
- vendor claims and their evidence;
- manual and automated testing;
- acceptance criteria;
- defects and remediation;
- change and regression;
- continuing user feedback.
Operational Resilience, Replay & Recovery
Why annex first. Retry, replay, rollback, and recovery appear across research systems, data pipelines, web surfaces, deployment, and regulated operations. The method is broadly reusable, while recovery objectives and mechanisms remain local. Repeated use can reveal whether this should remain an annex or become a full payload.
Required dimensions
- service and data invariants;
- failure geometry;
- interruption point;
- idempotency and duplicate behavior;
- durable state;
- backup and restore;
- dependency and network uncertainty;
- replay and reconciliation;
- recovery objectives;
- evidence from exercises;
- accountable return decision.
Promotion rubric
Score only after a candidate has a named receiving decision and source spine.
Dimension | Candidate | Nearly runnable | Executable |
Receiving decision | Broad topic | Named decision and receiver | Bounded decision, authority, and completion state |
Parameters | Implied | Structured but incomplete | Required, optional, defaults, and pause behavior |
Evidence | Source list | Hierarchy and claim classes | Coverage ledger, recovery, independence, transfer limit |
Method | Useful questions | Ordered procedure | Ordered procedure with branching and stop conditions |
Outputs | General deliverables | Defined artifacts | Schemas, worked specimen, provenance, accessible forms |
Verification | Editorial review | Some observable checks | Completion gate plus run receipt |
Transfer | Plausible | Boundaries named | Sending evidence, receiving conditions, and local test explicit |
Re-entry | Informal | Review date | Reopen events and next receiver explicit |
Promotion requires the executable column on the decision-bearing dimensions. Editorial polish supports use and remains distinct from executability.
Portfolio sequencing
- Use the shared core, parameter pack, and verification annex across the two current domain payloads.
- Promote AI Evaluation, Field Testing & Post-Deployment Monitoring first because it supplies evidence methods needed by several later instruments.
- Promote Vendor Evidence, Procurement, Renewal & Exit Portability next because it connects current capability choices to requirements, contracts, acceptance, change, and exit.
- Develop Work Redesign and Organizational Adoption as coordinated but distinct instruments: one centers tasks and roles; the other centers organizational conditions and learning.
- Develop Data, Retrieval, Privacy & Provenance as a boundary-bearing instrument that can attach to every other payload.
- Promote the regulated and accessibility payloads from completed domain specimens.
- Keep Operational Resilience modular until repeated use demonstrates a stable research job beyond implementation and recovery testing.
This order is a working route. A live receiving decision can move a candidate forward when its evidence, authority, and accountable receiver are ready.
Run and portfolio receipts
Every promoted payload should produce at least:
- one dry-run receipt;
- one worked specimen;
- one partial or unavailable evidence case;
- one disconfirming or alternative route;
- one completed handback;
- one named reopen event.
The portfolio page should be reviewed when a payload is promoted, a candidate is retired or combined, a new recurring decision appears, or an external source materially changes the evidence landscape.
Public-use boundary
This portfolio names research instruments and source seeds. It does not convert current guidance, surveys, principles, or Registry interpretations into universal requirements. Each run binds jurisdiction, population, organization, date, system, authority, and evidence access.
The desired state is a smaller set of strong, reusable instruments whose outputs help another collaboration team make a bounded decision and continue the work.