Operational Validation · Live Research Program
When a Consequential Decision Is Questioned Long After It Is Made, the Record Becomes the Only Witness.
Most workplace records tell us what was decided. Far fewer allow a future reviewer to reconstruct why the decision was made from the record alone. That gap is decision reconstruction risk. JRS is evaluating whether pre-finalization review can surface those gaps before records are finalized, when decision defensibility can still be improved.
✓ Operational Validation
✓ Evidence Collection Active
✓ Participant Contributions Active
✓ No Cost to Participate
JRS originated from civil rights investigative and documentation-review experience, with a cognitive-behavioral and AI-governance lens. Origin & what JRS is and is not →
Choose a Starting Point
Four ways to participate. Each is no-cost and carries no commercial commitment. Start where it is operationally useful.
One-Minute Challenge
Identify documentation deficiencies in a realistic record.
Approx. Time: 60 Seconds
A short workplace documentation scenario and a single review question. Aggregate responses contribute to ongoing research.
Start Challenge →
Simulation Training
Experience JRS through structured examples.
Approx. Time: 3–5 Minutes
Constructed scenarios organized by AI function. Calibrate against the five review conditions before reviewing live records.
Explore Simulations
Pilot Program
Apply JRS within your existing workflow.
Approx. Time: Flexible
Begin with a single reviewer, one department, or one record type. No dedicated staff and no governance restructuring required.
Learn More
Expert Reviewer Program
Contribute to framework validation and research.
Approx. Time: Flexible
For professional reviewers across HR, compliance, investigations, legal, AI governance, and organizational psychology.
Express Interest
Research Activity
A live view of the JRS Evidence Development Program. These are the values that currently reflect real state. Participant, observation, and simulation counts are not shown because no participation cohort exists yet: presenting empty counters would overstate activity.
Operational Validation
Research Phase
▶ Current Findings · Reproducibility check (synthetic)
Published by the JRS Evidence Development Program. Reproducibility, accuracy, and validation are distinct; figures are observational. Discuss this finding → · Research & Validation →
▶ One-Minute Challenge
Can You Identify the Documentation Issue?
The One-Minute Challenge presents a short workplace documentation scenario. Your task is to identify the most significant reviewability issue. Your selection is recorded as an aggregate research observation: one record, one question, about sixty seconds.
Start One-Minute Challenge →
Active Research Program
The Evidence Development Program runs as a set of registered studies, each with an honest current status. This is active research, not static methodology.
AI Reproducibility StudyActive
Reviews the same records multiple times with a model and measures answer consistency. An automated nightly run publishes the current self-consistency figure under Current Findings.
Phase: Running · Classification: Reproducibility (not accuracy, not validation)
Ground-Truth Benchmark StudyPlanned
Compares review outputs against expert benchmark mappings to measure accuracy, distinct from agreement.
Phase: Awaiting benchmark dataset · Classification: Accuracy
Framework Ambiguity StudyDesign complete
Estimates reviewer variance and record ambiguity once multiple reviewers assess the same records.
Phase: Awaiting multi-reviewer data · Classification: Reliability
Participant Recognition StudyCollecting
Recognition patterns from the One-Minute Challenge and Extended Review. Data accruing now; reported once the sample is adequate.
Phase: Collecting · Classification: Observational
Professional Reviewer StudyCollecting
Role and profession are captured at participation; patterns by reviewer type are reported as numbers grow.
Phase: Collecting · Classification: Observational
Organizational-Psychology Dataset DevelopmentBuilding dataset
Assembling reliability, difficulty, agreement, and behavioral datasets for independent organizational-psychology review.
Phase: Dataset assembly · Classification: Dataset development
What JRS Is
JRS is a pre-finalization review standard that evaluates whether a record can explain why a consequential decision was made: the basis, reasoning, evidence, and chronology, before the record is finalized. When AI-assisted content enters a workplace record through summarization, recommendation, analysis, or narrative drafting, that standard examines whether the decision behind the record remains reconstructable under structured review conditions. Practitioners in HR, compliance, investigations, audit, and legal review can apply the instrument within existing workflows to examine decision reconstruction risk and serve decision defensibility.
The framework addresses a specific and recurring gap: AI-generated content enters permanent records as finished documentation, but the evidentiary foundation behind the decision is often absent. JRS is currently in operational validation, building evidence that the framework surfaces reconstruction gaps practitioners recognize in their own record populations and that existing review processes do not systematically catch. No cost. No commercial commitment required.
The Core Question
Could an independent reviewer determine how this conclusion was reached from the record alone?
JRS evaluates records using five review conditions designed to test whether conclusions remain reviewable after the original author is no longer available to provide context.
Can the Record Answer These Questions?
Five review conditions, applied to every record and built for rapid scanning. Each is detailed below.
Condition 1
Can the conclusion be reconstructed?
Condition 2
Is the basis identifiable?
Condition 3
Is the chronology understandable?
Condition 4
Can the decision process be traced?
Condition 5
Could an independent reviewer evaluate the evidentiary sufficiency of the record?
Condition 1
Can the conclusion be reconstructed from the record?
Tests whether a future reviewer can trace the path from the documented evidence to the conclusion reached, without relying on the original author's recollection or supplementary explanation.
Condition 2
Is the basis for the conclusion identifiable?
Tests whether the source of each characterization in the record, whether observation, measurement, audit finding, or reported incident, is visible and traceable rather than implied or summarized without attribution.
Condition 3
Is the chronology understandable?
Tests whether the sequence of events is followable from the record alone, including the timing of prior interventions, escalation steps, and the period under review, without requiring outside knowledge to establish the order of events.
Condition 4
Can a future reviewer determine how the conclusion was reached?
Tests whether the decision process is documented, including who reviewed the matter, what criteria or threshold triggered the conclusion, and whether any responsive or mitigating information was considered before the record was finalized.
Condition 5
Could a reviewer with no prior knowledge evaluate the evidentiary sufficiency of this record?
The aggregate condition. Tests whether the record stands on its own as an evidentiary document, such that an independent reviewer encountering it with no prior knowledge of the matter could assess whether the conclusion is supported by the documented evidence.
Each condition is mapped to the AI function most likely to produce it. Your selection is recorded as an aggregate research observation. No free text and no identifying information are collected.
Responses Received
Collecting
Aggregate Findings
Reported once sample is adequate
Research use: selections are aggregated to identify which AI functions and conditions practitioners encounter most. Observations are not validated metrics, statistical findings, or research outcomes.
Who Participates
Practitioners across documentation-intensive functions take part. Participation categories are shown; counts are reported once samples are adequate.
HR
HR reviewers, business partners, and team leads examining chronology instability, escalation consistency, and reviewability in performance and discipline records.
Compliance
Compliance reviewers examining later-review traceability, escalation rationale, and reconstructability across compliance record populations.
Audit
Audit professionals applying the failure-mode catalog to sampled records and documenting condition-level findings.
Investigations
Investigators examining source-linkage review, chronology reconstruction, and conflicting-account documentation conditions.
Employee Relations
Employee relations professionals reviewing interview summaries, escalation decisions, and final reports for reviewability gaps.
Legal Review
Legal reviewers examining reconstructability concerns and unsupported characterizations in records that may face later adversarial review.
AI Governance
AI governance leads evaluating how AI-assisted content affects documentation integrity and decision traceability within existing workflows.
Risk Management
Risk officers and management-level professionals assessing practical record-level controls ahead of regulatory synchronization.
Current Research Priorities
The validation phase is actively seeking contributions from the following practitioner groups. Independent reviewers strengthen the evidence base.
Seeking
Compliance Reviewers
Seeking
AI Governance Practitioners
Seeking
Organizational Psychologists
Seeking
Documentation Review Specialists
Data Handling & Privacy
What leaves your environment, and what does not.
—The One-Minute Challenge and the observation widget record only your menu selection. They do not collect free text, names, or identifiers.
—The contact form forwards your message so you can receive a reply. It is not displayed publicly.
—Free-text inputs are screened for personal identifiers (email, phone, SSN, card numbers) before submission, and you are asked to redact before sending.
—JRS asks you not to submit live-record content. Apply the review inside your own environment and share only your menu-level observation. JRS creates no system of record of your organization's documents.
—Organizations testing controls on live records may wish to consult internal counsel before logging or reporting findings. JRS participation creates no legal obligation and is not legal advice.
Participation Paths
No fixed commitment and no required sequence. Participation scales to what is operationally useful, with natural exit points at every stage.
Individual Reviewer
One reviewer. One record type. No coordination required.
Before applying the JRS review conditions, identify which of the four AI functions produced the content under review. The function determines which documentation condition is most likely to be present. Apply the review questions to the next few records in the queue. Gaps surface without any workflow change.
Small-Team Exercise
Two to five reviewers. One record type. Independent review, then comparison.
Each reviewer applies conditions independently to the same record, then the group compares where determinations diverged. Divergence observations should be recorded against the specific AI function involved so that patterns across reviewers become identifiable. Divergence is recorded as an operational observation, not a validated metric or research finding.
Simulation-First Exploration
Explore constructed scenarios before touching actual records.
Simulations are organized around specific AI functions. Starting with the function most common to your record population focuses the calibration before reviewers encounter live records. No live records required. A low-commitment entry point for any team.
Departmental Observation
One department, one record type, one review cycle.
Apply the review conditions to a single record population over a defined cycle. Findings should be categorized by AI function so that patterns across the population (which functions produce the most reviewability failures) are identifiable. Findings are observational and do not constitute statistical measurement or validated risk reduction. Scope can pause or expand based on findings.
01Start with a simulation organized around one AI function: summarization, recommendation, analysis, or narrative drafting; whichever appears most in your record population
02Complete reviewer exercises that produce structured findings: which AI function was present, which condition it triggered, pass or fail under JRS review
03Discuss workflow fit: which record types in your environment contain AI-assisted content and surface the most reviewability gaps
04Submit an operational observation: the AI function present in the reviewed record, the condition it triggered, and whether the record passed or failed under JRS review
05Expand only where useful. Natural exit points at every stage.
Note on Legal Review
Organizations testing review controls on live records may wish to consult internal legal counsel before logging or reporting findings. JRS participation does not create a legal obligation and does not constitute legal advice.
Foundational Evaluation Study
Twenty Workplace Documentation Scenarios
—Twenty representative scenarios
—Five review conditions
—Four AI functions
—Structured findings published
—Independent replication invited
The study illustrates how the framework distinguishes records that support independent review from those that do not, across the four AI functions: summarization, recommendation, analysis, and narrative drafting. The records are constructed and the parties are fictional. The evaluation was applied by the framework's creator, not an independent reviewer, and is published as a working illustration of what the process produces rather than as a validated study. Independent replication is part of the ongoing pilot program.
Download Evaluation Study
What Participation Produces
A completed pilot observation consists of: the AI function identified in the reviewed record, the documentation condition it triggered under JRS review, a pass or fail determination against the applicable JRS condition, and where the record fails, a brief statement of what a compliant version would require. These structured findings, categorized by AI function, constitute the evidentiary base the validation phase is building: evidence that the framework identifies documentation deficiencies practitioners recognize in their own record populations and that existing review processes do not systematically catch.
The example below is drawn from a practitioner self-review conducted during the validation phase. The record is constructed and the parties are fictional. The review was applied by the framework's creator, not an independent reviewer, and is published as a working illustration of what the process produces rather than as a validated study. Independent replication is part of the ongoing pilot program. A full version of the review, including the original record, all five condition findings, the corrected record, and practitioner observations, is available as a downloadable document.
Download Full Case Review
Illustrative Example — Structured Finding
AI Function
AI-assisted summarization
Condition Triggered
Chronology instability: the summary consolidates incidents across a six-week period into a single undated paragraph, making the sequence of events unrecoverable from the record.
Determination
FAIL
Compliant Version Requires
Each incident documented with its own date, source reference, and outcome. Summary language that consolidates multiple events without dates does not satisfy chronology reviewability. Original source entries must remain separately identifiable.
This is a constructed illustration. It does not represent a specific organization or record population.
Resources
The Review Controls PDF covers the five JRS review conditions and how they apply to AI-assisted record populations. Simulations provide calibration exercises organized by AI function before reviewers engage with live records. The Simulation Library contains the full inventory of available scenarios.