Don’t Let a Silent Rejection
Kill Your Research Contract.

The Auditor is the only agent that operates in both Phase 1 and Phase 2. In Phase 1, it builds an independent baseline from patent filings and your SciENcv disclosures — before it knows what any other agent has found. In Phase 2, it receives anonymized findings from The Inquisitor and The Genealogist and cross-references all three sources to detect conflicts of interest, specifically personnel–entity conflicts.

No. Each agent has exclusive access to its own raw data, enforced architecturally. The Inquisitor cannot see cap-table data. The Genealogist cannot see publication records. All data flows route through The Architect, which itself is prohibited from accessing any agent’s raw data.

The protocol has a built-in reversal mechanism. If a downstream agent’s analysis contradicts an upstream finding, the contradiction is escalated to The Architect, which initiates a targeted re-query. The finding is never silently resolved — the discrepancy and its resolution appear in your Vetting Pack.

A single system creates a single point of failure and bias. Five independent agents with compartmentalized access means that when findings from multiple agents converge, it is the strongest possible signal. The architecture is adversarial by design: each agent challenges the others through independent analysis.

Every engagement includes three mandatory human review checkpoints — no exceptions, no automated overrides.

• Checkpoint 1 (Intake): Before any agent begins work, a human analyst reviews your engagement package — confirming entity identification, ITAR/EAR classification, personnel roster completeness, and SciENcv file integrity. If something is wrong at this stage, the engagement halts until it's corrected. This prevents a misidentified entity from cascading through all five agents.
• Checkpoint 2 (Flag Review): After Phase 1 discovery completes, all findings in the moderate-confidence range (0.60–0.74) are queued for human review. These are findings where the data is suggestive but not definitive — exactly the kind of call that requires human judgment, not an algorithm. The analyst reviews the evidence and issues a verdict: approved, rejected with a stated reason, or modified with specific changes. Findings above 0.75 propagate automatically because the evidence is strong enough to stand on its own. Findings below 0.60 require corroboration from a second agent or data source before they go anywhere.
• Checkpoint 3 (Vetting Pack Sign-Off): A senior analyst reviews and signs off on the final Vetting Pack before it reaches you. They check for false positive artifacts, verify that the narrative is coherent across all agent findings, and confirm that every mitigation memo recommendation is actionable — not generic boilerplate. No Vetting Pack is delivered without this sign-off.

Additionally, any time two agents produce conflicting results — for example, The Auditor and The Inquisitor parsing the same SciENcv field differently — the discrepancy is routed to a human analyst for resolution, not silently resolved by the system.