Current demo state
Not found
Active contracts
Executed records only
ContractLedger AI
GVGuest viewerRead Only Auditor
Read-only public demo
You are browsing every register, approval, obligation and validation record with read-only access. Uploading documents, recording decisions and running AI analysis require a signed-in workspace role. All organisations, people and agreements shown here are fictional.
Portfolio-ready evidence
Contract Operations Case Study
A concise interview narrative, ten-minute demonstration path, and claim-safe evidence ledger for the complete ContractLedger AI lifecycle.
Portfolio thesis
Turn unstructured agreements into verified operational records without hiding the human decisions.
ContractLedger AI stages legacy data, separates drafts from official records, routes exceptions, reproduces effective terms, and converts dates into accountable work. Every material result remains tied to its source, reviewer, rule, and history.
7
Lifecycle releases
v0.2–v0.8
15
Evaluation cases
Fictional ground truth
7 / 13
Roles / permissions
Server enforced
101
Passing tests
2026-09-02 v1.0 baseline
Current demo state
Not found
Active suppliers
Not found pending onboarding
Current demo state
Not found
Open obligations
Not found calculated overdue
Current demo state
Not found
Open approvals
Not found blocked intakes
Case-study decisions
The product choices that make the workflow defensible rather than merely automated.
1
Draft ≠ official record
Reviewed proposals remain intakes. Only authorized executed agreements enter official totals and registers.
2
Source before confidence
Critical fields require a supporting page and quote or an explicit reviewer override reason.
3
Rules own operational decisions
Deterministic code controls approvals, arithmetic, overdue state, risk factors, and permissions.
4
History stays reproducible
Original terms, amendments, decisions, obligation events, and exported evidence remain linked.
Release evidence ledger
Each claim is paired with reproducible evidence and an explicit boundary.
| Release | Capability | Verified evidence | Claim boundary |
|---|---|---|---|
| v0.2 | Amendment lifecycle | USD 475,000 original + USD 75,000 verified amendment = USD 550,000 current value, with the original preserved. | Reproducible fictional scenario; no legal-outcome claim. |
| v0.3 | Approval controls | Five versioned rules, immutable decisions, accountable owners, deadlines, and an executed-registration gate. | The seeded 26-hour turnaround is fixture evidence. |
| v0.4 | Legacy migration | CSV/XLSX staging, mapping, normalization, duplicate review, correction export, transactional commit, and audited rollback. | No real-company time-savings or duplicate-precision claim. |
| v0.5 | Obligation execution | Five seeded obligations cover upcoming, overdue, evidence-required, and evidence-backed completion states. | Completion metrics expose their small sample. |
| v0.6 | AI governance | Fifteen controlled fictional cases, field-level ground truth, source coverage, correction evidence, and a regression gate. | Accuracy must name its run, version, and sample size. |
| v0.7 | Quality and permissions | Seven workspace roles, thirteen named permissions, denied-write coverage, and page-level document preflight. | Demonstrates policy controls, not enterprise IAM certification. |
| v0.8 | Operational handoff | Source-aware review PDF, eight-factor supplier risk, and durable integration-outbox events. | Integration-ready; external delivery is not active. |
10-minute demo route
A paced path from business problem to measured evidence.
Evidence boundaries
What this portfolio deliberately does not claim.
- No real-company time-savings percentage until repeated paired timing runs are saved.
- No extraction-accuracy claim without a named completed run, dataset version, and sample size.
- No legal-advice, negotiation-outcome, or contract-enforceability claim.
- No production duplicate-precision claim without labeled false-positive and false-negative evidence.
- No external-notification claim while outbox delivery remains intentionally unconfigured.
Timed workflow evidence
Repeat the same fictional scenario manually and with AI assistance. The median and sample size below are the only basis for a time-saving claim.
Recording a timed run requires the AI governance permission. The saved evidence below is readable by every role.
| Scenario | Manual median | Assisted median | Reduction | Evidence |
|---|
Recent runs
No timed runs recorded yet. Record at least 3 manual and 3 assisted runs of the same scenario before quoting a reduction.