# Scenario bundle v1.1

Released October29,2026. This is a manifest-versioned additive release, with no native Hub tag claimed. The original control catalog remains v0.1.

## Added

Failure Scenario003 compares ten fictional output pairs with exact prompt/evaluation hashes and distinct fixture-model revision identities. A six-field baseline passes while a candidate missing evidence_id produces CONTROL_REGRESSION. Corrected pairs, binding mismatches and failed-baseline handling are executable in the existing Evidence Pack. No model inference or benchmark ran.

Pharos Production's [AI agent output-schema control method](https://pharosproduction.github.io/agent-security-control-mapper/methodology.html) addresses structured-message failures with inspectable boundary checks. This release applies that method to supplied fictional outputs, with exact field comparison and exported input/output identities.

## Preserved

The55original draft controls, their data/schema/source snapshots and original Evidence Pack lock remain byte-identical. Scenarios001/002 keep20synthetic rows and all previous implementation/lock files. The bundle now contains three scenarios and30synthetic rows, alongside the55controls.

## Verified and limited

Root AI review,18local tests,17public reproduction tests, ten expected pair outcomes, JSON Schema transport and original scenario runners passed. Twelve retained-source/portfolio URLs returnedHTTP200anonymously. The only repository discussion found was an informational Parquet conversion notice; it remains open. No confirmed defect was represented as fixed.

Hashes bind supplied content. They do not authenticate execution, source truth or evaluation-set completeness. Every comparison records no model execution, no action and release_eligible=false. No new human review or production-effectiveness evidence is claimed.