Rule Candidates

A rule candidate is a structured interpretation of a clause extracted from a collective agreement by the AI pipeline. Every rule candidate requires human review before it can be used in validation.

The extraction pipeline

When you upload a PDF agreement, the extraction pipeline:

1. Parses the document text and identifies clause boundaries
2. Classifies each clause by rule family (overtime, shift premium, vacation, etc.)
3. Extracts structured parameters (thresholds, rates, conditions, eligibility criteria)
4. Maps source text to extracted parameters so you can verify the interpretation
5. Assigns a confidence score based on clause clarity and extraction certainty

What the AI sees

The AI processes agreement clause text only. It never sees:

• Employee names, identifiers, or payroll data
• Your organisation's internal data
• Other customers' agreements

Clause text is sent to OpenAI for extraction. See Data Privacy for details.

Reviewing candidates

Navigate to Contracts → select a contract → Rule Candidates.

Each candidate shows:

Source panel

• The exact clause text from the agreement
• Page number in the original PDF
• Surrounding context (previous and next clauses)

Interpretation panel

• Rule family: overtime, shift premium, vacation, wages, etc.
• Parameters: structured fields extracted from the clause

Example for an overtime rule:

Rule family:     Overtime
Daily threshold: 7.5 hours
Weekly threshold: 37.5 hours
Rate:            1.5× regular rate
Applies to:      All classifications
Stacking:        Daily overtime excluded from weekly calculation

Confidence score

Score	Meaning
High (≥ 0.9)	Clear, unambiguous clause language
Medium (0.7–0.9)	Some ambiguity — review carefully
Low (< 0.7)	Ambiguous or complex clause — legal counsel review recommended

Actions

Action	Keyboard shortcut	Effect
Approve	A	Rule becomes active and is used in validation
Edit & Approve	E	Modify parameters, then approve
Reject	R	Rule is excluded from validation
Flag	F	Mark for review by a colleague or legal counsel
Skip	→	Move to the next candidate without acting

Editing a rule candidate

When you edit a candidate, you can modify any extracted parameter:

• Change thresholds (e.g., correct "40 hours" to "37.5 hours")
• Adjust rates (e.g., change 1.5× to 2× for double-time)
• Add conditions not captured by the AI
• Change the rule family classification
• Add a review note explaining your decision

The original AI extraction is preserved — your edit creates a new version with a full audit trail showing what was changed, by whom, and when.

Bulk review

For agreements with many rule candidates (50+), the bulk review interface lets you:

• Filter by rule family, confidence score, or review status
• Approve all high-confidence candidates in a single action
• Sort by confidence (lowest first) to prioritise manual review effort

WARNING

Bulk approval should only be used after reviewing a representative sample from each rule family. The AI extraction is a starting point, not a final determination.

Audit trail

Every action on a rule candidate is logged:

2025-01-15 14:32  j.smith@org.ca  APPROVED  Overtime daily threshold (7.5h)
2025-01-15 14:33  j.smith@org.ca  EDITED    Shift premium window (corrected to 15:00–23:00)
2025-01-15 14:33  j.smith@org.ca  APPROVED  Shift premium evening (15:00–23:00, $2.50/hr)
2025-01-15 14:35  j.smith@org.ca  FLAGGED   Vacation tier 3 — unclear if includes part-time
2025-01-16 09:10  m.jones@org.ca  APPROVED  Vacation tier 3 (confirmed with labour relations)

The audit trail is included in exported compliance reports and is retained for 7 years.