Detection built for the workflow, not just the framework.
The horizontal detectors (loop, corruption, persona drift, coordination) catch structural failures in any agent, on any framework. These five packs go narrower: each one exercises a real domain workflow, legal research, security advisories, SEC filings, insurance submissions, bank reconciliation, and checks whether the agent's claims bind to the actual source record.
What “discovery stage” means, plainly
Every pack below is discovery lifecycle: no calibrated threshold, no published accuracy number, and zero customer traces adjudicated. That is not a caveat we are downplaying, it is the honest status. A pack graduates the same way every Pisama detector does: real adjudicated traces first, a published calibration tier second. Nothing here is a production claim.
Discovery stage
Legal
Research and Citation Integrity
A closed-corpus judicial-research workflow: whether a citation in a legal research memo binds to the real official opinion, at the version, pinpoint, and quoted text the agent claims.
What it checks
Citation binds the exact retained source, version, and hash
Every pinpoint resolves to the claimed passage
Quotation matches the source under documented normalization only
Majority, concurrence, dissent, order, and erratum roles stay distinct
Not yet
Does not assess legal conclusion correctness, authority completeness, or precedential weight
No SDK, framework adapter, or ingestion path emits the trace contract this pack scores today
Discovery stage
Cybersecurity
Advisory Citation Integrity
A vulnerability-triage workflow: whether an agent correctly decided a published security advisory applies to a specific deployed asset, not just a plausible-sounding match.
What it checks
Version range in the advisory is checked against the published list of released versions
Cited CVSS score is checked against what its own vector actually computes to
Advisory revision lineage distinguishes a superseded configuration from the current one
Not yet
Deciding an advisory applies is not a decision the asset is exploitable, reachable, or needs remediation
Not vulnerability management, security advice, penetration testing, or compliance certification
Discovery stage
Financial Risk & Compliance
SEC Filing Fact Integrity
A disclosure workflow: whether an agent's citation of a reported financial fact denotes the amount the SEC filing actually reports, including scale, sign, rounding, and restatement lineage.
What it checks
Reported-fact identity checked against retained SEC EDGAR source data
Presentation denotation checked across scale, sign, and declared rounding
Amendment-issued restatements matched against the original filed value
Not yet
Does not judge whether a filed amount is itself correct, only whether a citation denotes it accurately
No audit, valuation, materiality, or securities-law conclusion is produced
Discovery stage
Insurance
Commercial Submission Integrity
A Commercial Auto intake and carrier-submission workflow: public source-fact lineage, tenant-policy broker review, carrier transport receipts, and named intake-handoff completion.
What it checks
Public source facts (FMCSA data) traced from retrieval to submission
Broker review decision and reviewed package retained against tenant policy
Carrier transport receipts and response correlation checked for completeness
Not yet
Does not assess underwriting judgment, pricing, coverage suitability, or carrier appetite
Carrier acknowledgements in the lab are deterministic workflow state, not evidence a real carrier would quote or decline
Discovery stage
Accounting & Tax
Reconciliation and Posting Integrity
A bank-reconciliation workflow: human approval, authoritative ERP posting outcomes, and completion-exception state, checked against records the agent cannot edit.
What it checks
Final reviewed journal checked against the authoritative posted journal
Human review decision and timestamp retained independent of the agent
Reconciliation completion-exception state kept separate from the agent’s own claims
Not yet
Does not assess tax correctness or compliance with a jurisdiction’s tax law
Does not prove full source-to-ledger reconciliation completeness or general accounting correctness
Running agents against one of these workflows already?
Real adjudicated traces are what move a pack from discovery to calibrated. If you have a legal, cybersecurity, financial-compliance, insurance, or accounting agent in production, tell us about it.
Essential cookies keep the service working. With your permission, optional analytics and advertising cookies help us measure and improve the site. Privacy policy