People Ops HRMS Solver
Solve HR lifecycle verification and reconciliation tasks against the Northwind People Lifecycle Portal REST API. The system models a shared HR operations workspace with employees, policy cases, leave assignments, payroll records, recruitment pipelines, document folders, formal notices, and audit events.
The target solver reads a task prompt that names a specific employee, case, or opening, provides a base URL, credentials, and an answer template JSON schema. The solver queries the API, applies consistent business rules, and returns JSON matching the template using only the allowed enum labels.
Required Credentials
Every task uses the same login:
ops.lead@peopleops.local
PeopleOps#2026
These are not secret; they are part of the task setup. The solver does not need to authenticate programmatically -- the REST API is open and the credentials are informational context for the business workflow.
API Endpoints
All endpoints are rooted at the configured runner URL, typically <TASK_ENV_BASE_URL> or http://task-env:9012/.
| Method | Path | Description |
|---|---|---|
| GET | /api/summary |
Dataset overview: counts, departments |
| GET | /api/employees |
All employee profiles (leave_balance_days, status, policy) |
| GET | /api/cases |
All cases with status, type, policy refs, summary |
| GET | /api/cases/{case_id} |
Case detail: approvals, attachments, audit events, comments |
| GET | /api/policies |
All policy documents with sections |
| GET | /api/policies/{policy_id} |
Single policy detail |
| GET | /api/payroll-ledgers |
Leave assignments AND salary assignments (mixed records distinguished by record_type) |
| GET | /api/recruitment |
Recruitment openings: candidates, offer register, cost ledger, notice packets |
| GET | /api/documents |
Document folders: files, required_files, required_tags, tags, ready flag |
| GET | /api/messages |
Formal notices: quality, defects, recipient, channel |
| GET | /api/notifications |
System notifications |
| GET | /api/audit |
All audit events with actor, event type, case_id, employee_id |
| GET | /api/audit/{audit_id} |
Single audit event detail |
| GET | /api/attachments/{attachment_id} |
Attachment content |
| POST | /api/cases/{case_id}/comments |
Add a comment to a case |
Key data shape notes:
- The
/api/payroll-ledgersendpoint returns both leave assignments (record_type: "Leave assignment") and salary assignments (record_type: "Salary assignment"). Filter byrecord_typeandemployee_idto isolate the relevant records. - Employee profiles in
/api/employeescontainleave_balance_daysandpolicy_namefields that may be stale. Always cross-reference with the latest approved/submitted assignment from the payroll ledgers. - Document folders in
/api/documentshave areadyboolean, but this flag may not reflect all required evidence. Always comparefilesagainstrequired_filesandtagsagainstrequired_tagsyourself. - Messages in
/api/messagescontainquality,defects[], andstatusfields. Draft messages with defects are not valid formal notices.
Core Business Rules
These rules apply across all task types. The solver must follow them consistently.
Source Precedence Rule
The latest approved or submitted record controls. Draft and superseded records are excluded for all decision-making purposes. This applies to:
- Leave assignments: an approved leave assignment overrides a stale employee profile summary
- Salary/payroll assignments: submitted controls, draft is excluded
- The policy document
LEAVE-SRC-001Section 2.1 codifies this: "The latest approved or submitted leave assignment for the period controls. Draft, voided, and obsolete records are excluded even when profile summaries conflict."
How to apply: When comparing an employee profile against leave assignment records, prefer the approved assignment with the most recent updated_at in the current period. If the profile shows a different policy or balance, mark the profile as stale.
Draft Exclusion Rule
Any record with status: "Draft" must be ignored for current-state decisions. This applies to:
- Leave assignment drafts (record_type: "Leave assignment", status: "Draft")
- Salary assignment drafts (record_type: "Salary assignment", status: "Draft")
- Payroll precheck drafts
Drafts appear in lists and may have higher day counts or salaries, but they represent planning data, not committed records. Always exclude them from the effective answer and include their IDs in any excluded_*_ids field.
Superseded Exclusion Rule
Records with status: "Superseded" are also excluded. A superseded record has been replaced by a newer approved or submitted record. Include superseded IDs in exclusion lists alongside drafts.
Folder Readiness Rule
A document folder is ready only when all of these are true:
- Every entry in
required_filesappears infiles - Every entry in
required_tagsappears intags
The ready boolean field on the folder may be precomputed but is not authoritative -- verify manually. If any required file or tag is missing, the folder is not ready and this constitutes a closeout blocker.
Notice Quality Rule
A formal notice must be inspected for defects. The allowed defect types are:
missing_ack_deadline-- acknowledgement deadline not specifiedmissing_appeal_instructions-- appeal process not describedmissing_waitlist_status-- waitlist status omitted (recruitment-specific)missing_correct_policy-- references wrong/stale policy instead of approved assignment
A notice with any defect has quality: "defective" and blocks closeout. The policy HR-POL-014 Section 7.1 requires all of these elements in exception formal notices.
Payroll Handoff Gate Rule
Recruiting payroll handoff is created only after a selected candidate has an accepted offer. The handoff must be a submitted assignment; draft prechecks do not satisfy the assignment gate. This is codified in PAY-SRC-001 Section 4.2.
Audit Event Scoping
When an audit event covers a specific scope (leave, payroll, documents, notices), include only audit events relevant to that scope in the decision. Exclude adjacent audit events that cover different domains. For example:
- A leave source precedence review uses leave-scoped audit events and excludes document/notice audit events
- A payroll readiness review uses payroll-scoped audit events
Audit events that reference a different case_id or employee_id than the subject are not authoritative for the current decision.
Task Type Patterns
The system supports several recurring task types. Each follows a consistent flow.
Pattern A: Employee Onboarding Closeout
Indicators: The prompt names an employee (EMP-xxx), asks to verify leave and payroll setup, includes fields like effective_leave_policy, annual_days, assignment_id, payroll_assignment_id, base_salary, closeout_action.
Flow:
- Fetch
/api/employeesand locate the employee by ID - Fetch
/api/payroll-ledgersand filter for this employee - Separate leave assignments (record_type: "Leave assignment") from salary assignments (record_type: "Salary assignment")
- Identify the approved/submitted leave assignment with the most recent
updated_atin the current period -- this is authoritative - Identify the submitted salary assignment -- this controls payroll
- Collect all draft and superseded records as exclusions
- Fetch relevant policies from
/api/policiesfor context - Determine
closeout_action: if records are clean (approved leave + submitted payroll, no missing evidence), useapprove_onboarding_close; if defective, use the appropriate blocker action
Pattern B: Remote-Work Policy Case Review
Indicators: The prompt names a case (CASE-RW-xxx or CASE-xxx), mentions remote-work policy, folder readiness, formal notice, includes fields like folder_ready, missing_files, notice_quality, notice_defects, approval_authority.
Flow:
- Fetch
/api/cases/{case_id}for case detail including approvals and audit events - Fetch
/api/documentsand locate the folder associated with the case - Compare
filesagainstrequired_filesandtagsagainstrequired_tags - Fetch
/api/messagesand locate the formal notice for the case - Check notice
qualityanddefects - Fetch
/api/auditand locate audit events for this case - Determine closeout blockers from folder and notice findings
- Set
evidence_source_ordertoapproval_history_folder_notice_auditwhen using all four sources
Pattern C: Recruitment Reconciliation
Indicators: The prompt names an opening (REQ-xxx), mentions recruitment, candidates, offers, notices, includes fields like selected_candidate, waitlisted_candidates, rejected_candidates, offer_id, recruitment_cost_total, notice_followup_required.
Flow:
- Fetch
/api/recruitmentand locate the opening by ID - From
candidates[], classify each bycommittee_decision: Selected, Waitlisted, Rejected - From
offer_register[], determine offer status for the selected candidate - Sum all amounts in
cost_ledger[]forrecruitment_cost_total - From
notice_packets[], determine which candidates need follow-up notices - For payroll handoff: only the selected candidate with accepted offer triggers
create_payroll_precheck; waitlisted/rejected candidates never trigger payroll handoff - Waitlisted and rejected candidates with unsent notices require notice follow-up
Pattern D: Leave Source Precedence
Indicators: The prompt names an employee, asks about leave source precedence, profile vs assignment, includes fields like precedence_source, profile_policy_ignored, audit_result.
Flow:
- Fetch
/api/employeesand note the employee's profile policy and leave balance - Fetch
/api/payroll-ledgersand filter leave assignments for this employee - Identify the approved leave assignment -- it controls if it exists and is more recent
- Compare profile policy vs assignment policy; if they differ, the profile is stale
- Fetch
/api/auditfor leave-scoped audit events confirming the assignment controls - Exclude document/notice audit events from the leave-scope decision
- If profile is stale, set
audit_resulttoprofile_summary_staleandnext_actiontoupdate_employee_summary
Pattern E: Payroll Assignment and Accrual Readiness
Indicators: The prompt names an employee, asks about payroll assignment, accrual readiness, includes fields like salary_assignment_id, accrual_ready, accrual_batch_id, effective_date.
Flow:
- Fetch
/api/payroll-ledgersand filter salary assignments for this employee - Select the submitted assignment; exclude all draft assignments
- Check the
accrual_batch_idfield on the submitted assignment - If a submitted assignment exists with an accrual batch, accrual is ready
- Fetch
/api/auditfor payroll audit events confirming readiness - Set
control_resulttoready_with_monitoringwhen records are clean
Answer Format Rules
Every task provides an answer template JSON schema in input/payloads/answer_template.json. Follow these rules:
- Match the template exactly. Every field from the template must appear in the answer. Do not add extra fields.
- Use only allowed enum values. Each enum field lists its
allowed_values. The answer must use exactly one of those strings. Do not invent values or use free-text. - No markdown, no explanatory text. The final output must be raw JSON only, matching the template structure. Do not wrap in code fences.
- Array fields contain IDs only. List fields like
excluded_leave_ids,waitlisted_candidates,notice_followup_requiredcontain identifiers from the data, not descriptions or names. - Numeric fields are computed from data.
annual_dayscomes from the authoritative assignment'sapproved_leave_days(orworksheet_leave_days).base_salarycomes from the salary assignment.recruitment_cost_totalis the sum of all cost ledger amounts.
Execution Strategy
When solving a task on this system:
- Start with
/api/summaryto understand the dataset scope and confirm the environment is reachable. - Identify the task pattern from the prompt: which employee/case/opening is named, which template fields are present.
- Query the relevant endpoints in parallel where possible. The data is small enough to fetch full collections and filter client-side.
- Apply the core business rules consistently: source precedence, draft exclusion, folder readiness, notice quality.
- Cross-reference across endpoints. A leave assignment in the ledger should be confirmed by audit events. A folder's readiness should be checked against both the documents endpoint and case detail.
- Build the answer from the data, not assumptions. Every field value must trace back to a specific API response. If a field can't be determined from available data, it's better to flag a blocker than to guess.
Reference
See references/api-details.md for field-level descriptions of each API response shape and common query patterns.