Deterministic Time Classification
Every classification,
traceable back to a rule
ZoikoTime classifies workforce records using policy-bound rules — not AI. Approved facts, context, and rule precedence produce a result you can explain, trace, and review, every time.
- Rule-based, not AI
- Every result is explainable
- Human review, always
Rule Trace
Record #48213 · synthetic data
- 1
Source Facts
Clock-in 09:07 · Site: Meridian+
Worker type: Hourly · Source: Mobile app
- 2
Context
Scheduled start 09:00
Grace window: 5 min · Timezone: America/Chicago
- 3
Rules
Policy v4.2 · Grace-Window-Tolerance
Precedence: site override > org default
Classification Result
Late Clock-In · exceeds 5-min grace by 2 min
Review State
Pending Review
Worker context requested
Rule version 4.2 · effective Jul 1, 2026
What Deterministic Means
Rule-based classification, explained side by side
Deterministic (ZoikoTime)
- Same facts + same policy = same result, always
- Every result traces to a named, versioned rule
- Policy owners can read and approve the exact logic
- Changes are versioned, dated, and auditable
Black-box ML classification
- Same facts can produce different results over time
- Results are hard to trace back to a specific cause
- Model logic is difficult for a policy owner to inspect
- Drift can occur silently as models retrain
ZoikoTime uses AI only for verified anomaly flagging or signal quality — never for classification itself.
Classification Anatomy
What’s inside every classification result
| Component | Content |
|---|---|
| Source facts | Captured event, timestamp, source, and worker/site context |
| Context | Schedule, grace window, timezone, and applicable exceptions |
| Rules applied | Named rule, policy version, and precedence path |
| Result | The classification outcome, stated in plain language |
| Review state | Whether the result requires human review before use |
Rule Trace
Source Facts → Context → Rules → Result → Review
The same five-stage trace behind every classification — explorable in full for any record.
Source Facts
Approved event data, exactly as captured
Context
Schedule, policy scope, and jurisdiction
Rules
Named, versioned rule and precedence order
Classification Result
Plain-language outcome, always explainable
Review State
Flows to human review where policy requires it
Rule Trace Explorer · Recent records
Policy v4.2
| Record | Rule applied | Result | Review |
|---|---|---|---|
| #48213 | Grace-Window-Tolerance | Late Clock-In | Pending Review |
| #48214 | Break-Completeness-Check | Verified | No review needed |
| #48215 | Duplicate-Event-Dedupe | Duplicate Clock-Out | Reviewed · Corrected |
Policy & Jurisdiction
Rules apply only within their configured scope
Effective-dated policy
Every rule has a start and end date — a classification always reflects the policy version in force at the time of the event.
Scoped by organization
Rules apply at the org, site, or team level as configured — never assumed globally across your organization.
Jurisdiction-gated
Jurisdiction-specific rule content is applied only where that jurisdiction has been configured and approved — never inferred.
Neutral States
A result is a classification — not a verdict
| State | Meaning |
|---|---|
| Verified | Facts evaluated against policy with no unresolved conflict |
| Needs Context | A required fact is missing or in conflict with policy expectation |
| Pending Review | Context received; awaiting an authorized reviewer's decision |
| Corrected | An authorized change has been made and recorded with full history |
| Approved | Result confirmed and ready for downstream use |
Worker Explanation
“Why was I classified this way?” — answered in plain language
No jargon, no black box
Every worker-facing explanation uses the same plain language as the underlying rule — not a simplified summary that hides what actually happened.
- "Late Clock-In — exceeds 5-minute grace window by 2 minutes"
- Not "Suspicious" or "Policy Violation" — the exact rule outcome
- A direct path to add context or request correction
Human Review
Classification informs review — it never replaces it
Rules propose, people decide
A deterministic result is the starting point for review, never the final word — consequential decisions stay with authorized people.
Reviewer sees the full trace
Reviewers see the same facts, context, and rules the system used, so a decision rests on the same evidence.
Override is logged, not silent
Where a reviewer's decision differs from the classification, the override, actor, and rationale are all preserved.
No autonomous action
A classification result never triggers pay changes, discipline, or any other automated consequence on its own.
Policy Governance
Every rule change is versioned and previewed
- 1
Draft change
A policy owner proposes a rule change with rationale and scope.
- 2
Impact preview
ZoikoTime shows which records would classify differently under the proposed change, before it's published.
- 3
Approve & version
The change is approved, versioned, and given an effective date — never applied retroactively without record.
- 4
Publish
The new policy version takes effect; prior versions remain fully readable for historical records.
Product UI Showcase
Classification Operations Center
Production-faithful, shown here with synthetic data.
Classification Operations Center
Aug 5, 2026
- 8,412
- Classifications this week
- 31
- Review queue
- 100%
- Policy health
- 99.6%
- Source freshness
Policy v4.2 active
9 due today
No conflicting rules
All sources syncing
| Record | Result | Rule | Review state |
|---|---|---|---|
#48213 Meridian+ | Late Clock-In | Grace-Window-Tolerance | Pending Review |
#48216 Northwind Ops | Verified | Break-Completeness-Check | Approved |
#48219 Astera Retail | Duplicate Clock-Out | Duplicate-Event-Dedupe | Corrected |
Policy change & impact preview
- Grace window 5 → 7 min · Site B142 records affected
- New rule: Split-Shift-DetectionDraft · not published
Source health
- Mobile appSyncing
- Kiosk · Site CSyncing
- Scheduling integrationSyncing
Evidence & History
Nothing about a classification is a mystery later
Aug 5, 2026
Policy v4.2 published
Grace-Window-Tolerance updated from 3 minutes to 5 minutes, with rationale and approver recorded.
Aug 3, 2026
Rule precedence clarified
Site overrides now documented ahead of org defaults, with every affected record listed in the impact preview.
Jul 29, 2026
Duplicate-Event-Dedupe added
New rule introduced with an explicit effective date; earlier records remain classified under the prior version.
System Boundaries
What Deterministic Time Classification is not
- Not AI decision-making
- Not payroll approval
- Not a legal compliance guarantee
- Not worker surveillance
- Not productivity scoring
- Not misconduct detection
Classification results never trigger an autonomous decision — pay, discipline, or otherwise. A human always decides.
Enterprise Readiness
Built for the teams who have to defend the outcome
Auditors & compliance
Trace any result back to its exact rule, version, and effective date.
Explore Evidence Ledger →
HR & IT
One governed rule set, understandable by policy owners and engineers alike.
Visit Trust Center →Questions
Deterministic Time Classification — answered
No. Classification runs on versioned, policy-bound rules — not a machine-learning model. Where AI appears in ZoikoTime it is confined to verified anomaly flagging and signal quality, and it never performs classification itself.