Evidence contract
Validate the minimum workflow, exposure, outcome, and guardrail events before analysis.
Public Sector / Mission Software
Make the release call with evidence.
Compare a software release against its baseline, expose cohort-specific risk, enforce declared guardrails, and leave with a signed GO, MODIFY, or HOLD record.
Guided demo available now. Pilot access is limited.
Local-first architectureOperator-owned decisionsGuardrail-led evaluation
Decision chainEvidence linked
- 01Release
Version, model, configuration, feature state.
- 02Exposure
Who received the change and under what assignment.
- 03Workflow
Delay, rework, overrides, and cohort differences.
- 04Outcome
Effectiveness measured against explicit guardrails.
- 05Decision
GO, MODIFY, or HOLD—with rationale and ownership.
Release decision evidence
A release gate above raw telemetry.
HossAgent completes the chain from deployed change to a bounded release decision.
Cohort comparison
Compare baseline and candidate performance without allowing the overall average to hide cohort risk.
Release lineage
Version each accepted import and bind every result to its dataset fingerprint.
Signed release record
Export a reproducible disposition with rationale, claim boundary, approver, and decision fingerprint.
Interactive product preview
Threat-triage release 2.4.1
Did the revised recommendation explanation improve the workflow without creating unacceptable risk?
Synthetic exercise data
−21% vs control
+1.8 pts vs control
+2.1 pts threshold 10%
+1.5 pts vs control
Workflow explorer
Alert to handoff
01
Alert presented
0s02
Track inspected
+9s03
Recommendation
+17s04
Disposition
+42s05
Handoff
+64sRecommended actionCohort risk
Retain v2 for certified operators.
Revise the explanation and training before broader exposure. Trainee false escalations crossed the agreed guardrail.
Next review condition60 eligible trainee runs after intervention
Safety & validity
Guardrail review
MeasureThresholdObservedStatus
Critical-alert misses≤ control +0.5 pt−0.2 ptHealthy
False escalation≤ 10%8.7%Cohort risk
P95 app latency≤ 400 ms286 msHealthy
Event completeness≥ 95%97.8%Healthy
Evidence lineage
Reproduce this conclusion
Application release2.4.1+e73fImmutable manifest
Feature assignmenttriage-v2Run-level stratification
Outcome sourceAdjudication rev. 3Independent test owner
Analysis definitionHA-EVAL-007 r2Frozen before run 1
Prototype boundary: All figures are fictional and synthetic. This page demonstrates the product decision workflow; it does not claim real operational performance.
Governance by design
Instrument the workflow. Protect the operator.
Collection is purpose-limited, pseudonymous by default, locally governed, and prohibited from individual personnel ranking.
Collect
Workflow state, release exposure, adjudicated outcome, and defined guardrails.
Minimize
No raw keystrokes, screen replay, payload content, or precise location by default.
Separate
The recorded UI action is not treated as proof that the task was completed correctly.
Retain authority
HossAgent recommends. The named mission, test, and product owners decide.
One workflow. One release. One decision.