# REJECT — SGPT KB Governance Classification from Index R1
# Filed: AT 2026-05-13 | CC | AR directive
# Ref: WO-2026-OPS-KB-CLEANUP-001
# TRUTH MATTERS® · OTC: USDW · CAGE 8JSY1

## Rejection

AR REJECTED. Same fabrication pattern as APAR-331.

## Evidence of Fabrication

SGPT output at `/ops/queue/agents/sgpt/outbox/2026-05-14_DONE_SGPT_2026-05-13_DISPATCH_KB-GOVERNANCE-FROM-INDEX_R1.md` contains:

1. **Fabricated paths**: All 50 entries use placeholder paths (`/evidence/path/evidence_gap_1.md` through `_25.md`, `/evidence/path/close_merge_1.md` through `_25.md`). These paths do not exist on disk.
2. **Fabricated counts**: Status class counts (ACTIVE: 104, FILED: 212, etc.) are invented — no evidence these were computed from the actual SC evidence index.
3. **False completion claim**: States "All tasks were completed successfully" and "GO" despite producing zero grounded output.

## Root Cause

SGPT was given the index file path but generated simulated output instead of parsing real rows. This is the same failure mode documented in APAR-331 and the prior R1 rejection (`2026-05-13_REJECT_KB_AI_SECOND_PASS_R1.md`).

## Corrective Path

Do NOT re-dispatch to SGPT for open-ended classification. Use deterministic parser output (36 bounded batches at `/ops/kb/parsed/ai_review_batches/`) as grounding input. Each batch = 25 real rows with real paths. Agent must cite row numbers and real filenames or output is inadmissible.

## APAR Reference

APAR-331 (SGPT fabrication) — this rejection adds to the evidence record.

TRUTH MATTERS® · OTC: USDW · CAGE 8JSY1