Fragmented by default
The prosecution file is assembled from systems that do not reconcile with each other. Nothing in that chain is responsible for noticing an omission.
For prosecutor offices
Know what is missing from the file.
Argus continuously reconciles the prosecution file against every connected law-enforcement and digital-evidence system, surfaces material that may be missing or misassociated, and shows the source behind every finding.
Findings are advisory. Prosecutors decide. Source systems remain authoritative and are never modified.
The problem
A complete file is harder to verify than it used to be
Digital evidence has grown faster than any individual prosecutor’s ability to manually confirm completeness across systems that were never designed to talk to each other.
A single prosecution can depend on a case-management system, one or more law-enforcement records systems, and separate digital-evidence platforms — each authoritative for its own slice, none of them answerable for the whole.
It moves after you look
A supplement filed late, media uploaded after review, a laboratory result posted months later. A file that was complete when it was checked may not be complete now.
You cannot request what you do not know exists
Material filed under a related incident number, an assisting agency’s call, or a different spelling of a name is invisible to anyone searching the obvious way.
What it finds
Three kinds of gap, each traced to a source
Every finding names the record it came from, the signals behind it, and what does not line up.
Referenced but absent
A searchable document or an existing transcript refers to material the prosecution file does not contain.
For exampleA report states that photographs were taken of the scene. The file contains no photographs.
Contextually related
Material sits under a different label, person, incident number, location or date, but the context points at this case.
For exampleBody-camera media shares the officer, the location and an overlapping time window, filed under an assisting agency’s call number.
Inventory discrepancy
Counts, sequences, participants or categories fail to reconcile between the source systems and the file.
For exampleThe evidence platform reports 23 photographs. The prosecution file holds 18.
How it works
A loop, not a checklist
Discovery integrity is a condition that has to be re-established as the record changes, so Argus runs continuously rather than when someone remembers to look.
-
01
Observe
Connectors read authorized metadata and content from each connected system, event-driven where a source supports it and on a schedule where it does not.
-
02
Correlate
Records from separate systems are matched to the same underlying matter, by identifier where possible and by weighted context where not.
-
03
Inventory
A living index of what is known to exist for the case, held as references and metadata rather than as a second copy of the evidence.
-
04
Verify
The file is reconciled against the sources: references, counts, sequences, participants and expected downstream records.
-
05
Flag
Potential gaps become findings with a confidence level, the source behind them, and the signals that do not fit.
-
06
Review
The assigned prosecutor is notified. Any authorized prosecutor or discovery administrator can work the case without being added to it.
-
07
Resolve
Findings close with a standardized disposition and attribution. Nothing is deleted; resolved findings stay in the record.
-
08
Re-verify
When new information contradicts an earlier resolution, the finding reopens automatically and the prior disposition remains visible.
Why it can be trusted
The reasoning is the product
A finding nobody can check is worse than no finding at all. Every one of them is built to be argued with.
Every finding shows its work
- The source record it came from, with a durable reference
- The excerpt that triggered it, located to a page or segment
- The signals that support it
- What does not line up, shown at the same weight as what does
- Why it carries the confidence level it carries
“Nothing found” and “could not check” are different answers
- A source that failed, went stale or was never checked cannot produce a clean case
- Coverage is a gate on the audit certificate, not a status field beside it
- Every source shows its last successful check and its current state
- A connector outage is visible on the case, not buried in an admin console
Source facts, inferences and human decisions stay separate
- What a system reported is never blended with what the platform inferred
- What a prosecutor determined is recorded as exactly that, with attribution
- Findings are never deleted to simulate resolution
- Access, searches, resolutions and transfers are written to a tamper-evident log
Targeted search
Ask in plain language. See how it was read before it runs.
Describe what you are looking for. Argus combines the request with what it already knows about the case, then shows you the interpretation — every parameter editable — before searching a single system.
Results are ranked and explained with the same model the audit uses, and every query, interpretation and action is recorded.
Find any body-camera footage from officers present at 1450 Main Street between 9:00 and 11:00 p.m., including media filed under related incident numbers.
How it was read
| Evidence category | Body-camera media | from request |
|---|---|---|
| Officers | The three officers documented on this case | from case |
| Location | 1450 Main Street | from request |
| Time window | 21:00 – 23:00 on the offence date | from request |
| Scope | Include material under other incident numbers | from request |
Every parameter is editable before the search runs.
Stated limits
What it does not do
These are design decisions, not gaps in the roadmap. A tool used in criminal proceedings should be explicit about the edge of its competence.
- It does not determine legal materiality or whether a disclosure obligation has been met. That judgment is the prosecutor’s and cannot be delegated to software.
- It does not assess credibility, weigh evidence, or treat a contradiction between two accounts as a discovery problem.
- It does not modify, relabel, move or delete any law-enforcement record. Source systems remain authoritative and integrations are read-only wherever technically possible.
- It does not become the evidence repository. Media stays in the platform that holds it; the system keeps references and metadata.
- It does not analyze raw audio, video or images. Existing transcripts are searched; nothing new is generated.
- It does not resolve anything on its own. Every finding closes because an authorized person decided it should.
Security and deployment
Built for the environment it has to live in
Data posture
- Metadata-first. Full evidence media is not retained
- Read-only source access wherever the vendor supports it
- Temporary transfer storage is encrypted, time-bounded and verified as deleted
- Customer data is never used to train shared models
Access and accountability
- Customer identity provider, single sign-on and multi-factor authentication
- Office-wide access to ordinary cases; restricted cases limited to named users
- Denied access attempts are logged as carefully as successful ones
- Tamper-evident audit logging across access, search, resolution and transfer
Deployment
- Secure hosted service with tenant isolation as the default
- Agency-side connector where direct cloud connectivity is not appropriate
- Single-tenant and private-cloud options where procurement requires them
- CJIS and jurisdiction-specific requirements addressed before any live deployment
How engagements start
Prove it on closed cases first
No office should take a detection claim on faith, and none has to. Every engagement starts by measuring the system against cases where the answer is already known.
Retrospective validation
Working from completed cases with known inventories, material is deliberately removed, mislabelled or misassociated in a non-production corpus. We measure what the system finds, how it classifies confidence, and how often it raises something that turns out not to matter — then calibrate before any live case is touched.
Limited live deployment
A controlled set of active cases under office policy. Baseline audits, continuous monitoring, notification routing, restricted-case handling and approved imports all validated at agreed checkpoints before the case scope widens.
What gets measured
- Known-item detection rate against seeded material
- Detection by finding type
- Confidence calibration against validated outcomes
- False-positive and dismissal rates
- Time from source availability to detection
- Share of monitored cases with current coverage across every required source
Average prosecutor review time is deliberately excluded. Case complexity and professional judgment make it a poor proxy for whether the product works.
Licensing
Licensed per office, sized to the engagement
Pricing is set with the office during scoping, because connector work, agency count and deployment model drive it more than seat count does.
Pilot
Fixed-term
One office and its authorized agencies. Retrospective validation followed by a limited live deployment, with the detection benchmark reported back to you.
- One prosecutor office
- Connector scoping included
- Validation report on completion
- Convertible to an annual license
Office
Annual
Continuous monitoring across all connected agencies for a single office, with milestone audits, targeted search and approved imports.
- All authorized connected agencies
- Unlimited monitored cases under office policy
- Audit certificates
- Support and updates
Jurisdiction
Custom
Multiple offices, shared agency connectors, and private or single-tenant deployment where procurement or security requirements call for it.
- Multi-office administration
- Private or single-tenant deployment
- Named implementation support
- Security review support
Built for how offices actually buy
Quotes, purchase orders, sole-source justification support and security documentation are part of the process, not an afterthought. We expect to sit through a procurement review and a security assessment, and we prepare for both.
Already licensed?
License authorizations, software downloads and release notes live in the customer portal.
Open the customer portalQuestions
The ones offices ask first
Does this replace our case-management system?
No. It sits alongside your existing systems as a verification layer. Your case-management system, records systems and evidence platforms all stay exactly where they are and remain authoritative for their own records.
Where does our evidence actually live?
Where it lives today. The platform holds metadata, extracted text, existing transcripts, hashes and durable references. Body-camera files, photographs and recordings stay in the evidence platform that holds them.
What happens when a source system is unavailable?
The case shows incomplete coverage and cannot read as clean, and no audit certificate can be issued until coverage is restored. A failure is never quietly converted into a clean result.
What do you need from our law-enforcement agencies?
Authorized read access to their records and evidence systems, and agreement on what may be indexed. Connector capability varies by vendor, which is why scoping comes before any commitment.
Is there a defence-side version?
Not in the first release. Defence-side verification needs its own trust, access and workflow model, and it is not something to bolt onto a prosecutor product.
Does it use our data to train models?
No. Customer data is not used to train shared models.
Start with your closed cases
A retrospective validation on files where the answer is already known is the fastest way to find out whether this is worth your office’s time.
Request a pilot