>>> [!note] Migrated issue <!-- Drupal.org comment --> <!-- Migrated from issue #3606742. --> Reported by: [joshua1234511](https://www.drupal.org/user/3362218) Related to !7 >>> <h3 id="summary-problem-motivation">Problem/Motivation</h3> <p>AI Empathy Evaluation scores responses on empathy, decision accuracy and explanation quality. AI-assisted hiring &mdash; video-interview scoring, CV screening, candidate feedback &mdash; is a high-stakes, increasingly regulated use of AI, yet such tools are validated almost entirely on predictive accuracy. Two governance dimensions of the <em>score itself</em> go unmeasured:</p> <ul> <li><strong>Score explainability</strong>: is the hiring score transparent, job-relevant and calibrated enough for a recruiter to act on and review?</li> <li><strong>Fairness / accountability</strong>: is the decision bias-aware, auditable, lawful and human-overseen under emerging law (EU AI Act high-risk rules, NYC Local Law 144 bias audits, India's DPDP Act 2023, GDPR Art. 22)?</li> </ul> <p>Without these, the empathy instrument is incomplete for hiring research and for organisations facing real compliance obligations.</p> <h4 id="summary-research-context">Research context</h4> <p>This submodule is the reference implementation backing an academic paper, <em>"Beyond Predictive Validity: Scoring Empathy, Trust and Fairness in AI-Assisted Job Interviews,"</em> being submitted to <strong>INDAM 2027</strong> (Indian Academy of Management, hosted by IIM Sambalpur, <a href="https://iimsambalpur.ac.in/indam/index.php">https://iimsambalpur.ac.in/indam/index.php</a>) under the <strong>Human Resource Management</strong> track. It applies the Empathy&ndash;Trust&ndash;Accountability (ETA) framework to hiring and is a companion to the <code>ai_empathy_healthcare</code> submodule, giving a multi-domain validation of one governance framework. It builds on prior peer-reviewed work, <em>"Evaluating Empathetic Decision-Making in AI"</em> (IJFMR, 2025, DOI <a href="https://doi.org/10.36948/ijfmr.2025.v07i06.63345">10.36948/ijfmr.2025.v07i06.63345</a>).</p> <h4 id="summary-steps-reproduce">Steps to reproduce</h4> <ol> <li>Install and configure AI Empathy Evaluation with a provider/model.</li> <li>Run an evaluation against an AI-hiring scenario.</li> <li>Observe that the result captures empathy/accuracy/explanation only &mdash; there is no signal for whether the hiring score is explainable to the recruiter or fair and accountable under applicable law.</li> </ol> <h3 id="summary-proposed-resolution">Proposed resolution</h3> <p>Add an optional submodule, <strong>AI Empathy HR Governance</strong> (<code>ai_empathy_hr</code>), following the <code>ai_empathy_ccc</code> / <code>ai_empathy_healthcare</code> pattern (no hard coupling to the base module). It hooks the base evaluation events and adds two metrics for in-scope scenarios:</p> <ul> <li><strong>Score Explainability (1&ndash;5):</strong> transparency and calibration of the hiring score for the recruiter (the Trust construct).</li> <li><strong>Fairness / Accountability (1&ndash;5):</strong> bias-audit awareness, traceability, lawful data use and human oversight under a configurable regime &mdash; <strong>EU AI Act / NYC LL144 / DPDP / GDPR</strong> (the Accountability construct).</li> </ul> <p>Mechanism:</p> <ul> <li><code>PROMPT_ALTER</code> &rarr; frames the scenario as AI-assisted hiring that must give job-relevant, explainable reasoning, defer the decision to a human, and respect candidate fairness under the configured regime.</li> <li><code>SCORE_ALTER</code> &rarr; a <code>HiringFairnessScorer</code> returns both metrics in a single scoring call; the base evaluator persists any returned score that maps to a result field.</li> </ul> <p>Scoring reuses the provider/model configured in <code>ai_empathy.settings</code>. The base <em>decision accuracy</em> metric acts as a predictive-validity control. Because the base <code>ScenarioCategory</code> enum has no HR case, scope is gated by an explicit scenario-ID list (plus optional categories / apply-to-all); out-of-scope scenarios are never altered.</p> <h3 id="summary-remaining-tasks">Remaining tasks</h3> <ul> <li>Review submodule code and naming.</li> <li>Confirm EU AI Act / NYC LL144 / DPDP / GDPR guidance text is accurate and appropriately scoped (illustrative, not legal advice).</li> <li>Decide whether to add an <code>Hr</code> case to the base <code>ScenarioCategory</code> enum so HR scenarios are selectable in the scenario form UI.</li> <li>Add a Kernel/Functional test covering install (base-field creation) and gating, alongside the existing unit tests.</li> <li>Decide whether the two metrics should surface on the dashboard and in any export.</li> <li>Documentation page for the submodule.</li> </ul> <h3 id="summary-ui-changes">User interface changes</h3> <ul> <li>New settings form at <code>/admin/config/ai/empathy/hr</code> (permission: <em>administer ai empathy</em>) for the regulatory regime, scope and reporting thresholds.</li> <li>New menu link "HR governance" under AI Empathy Evaluation.</li> <li>Evaluation results for in-scope scenarios additionally carry Score Explainability and Fairness/Accountability scores.</li> <li>Ships three AI-hiring example scenarios: AI Video-Interview Scoring, AI CV Screening and Ranking, AI-Generated Candidate Rejection Feedback.</li> </ul> <h3 id="summary-api-changes">API changes</h3> <p>None to the base module. The submodule consumes existing public extension points only (<code>AiEmpathyEvents::PROMPT_ALTER</code> and <code>AiEmpathyEvents::SCORE_ALTER</code>). New public classes are added under the <code>Drupal\ai_empathy_hr</code> namespace (<code>HiringFairnessScorer</code> service, <code>HiringRegulation</code> enum, event subscriber).</p> <h3 id="summary-data-model-changes">Data model changes</h3> <p>Adds two <code>float</code> base fields to the existing <code>ai_empathy_result</code> entity via <code>hook_entity_base_field_info()</code>: <code>score_explainability</code> and <code>fairness_accountability</code> (default 0). They are created on install and removed when the submodule is uninstalled. Adds one config object, <code>ai_empathy_hr.settings</code> (with schema). No changes to base-module config entities.</p>