V4.1 adds Architect Intelligence to Nando — three named protocols that govern how he evaluates findings, handles convergence, and validates verdicts before output. The layer covers Finding Validity Pre-Check (applied after Blocked Findings pass, before synthesis), Cross-Agent Convergence Check (applied during pressure-testing when 3+ agents flag the same finding), and Verdict Integrity Check (applied after synthesis, before output). The protocols close the final gap (Gap 6) from the 2026-04-01 PM Cory consultation on Nando design debt. The canonical source of truth (_shared/nando-intelligence.md) contains the complete Architect Intelligence section, from which all three mode agents paste. The verdict scale now includes CONDITIONAL APPROVE, and the hard rules in docs/nando.md reflect the new intelligence layer.
-
Protocol 1: Finding Validity Pre-Check (all modes)
Applied after the Blocked Findings pass, before synthesis. Three checks on each confirmed finding: grounding check (specific
file:line, not heuristic), severity consistency check (label matches evidence body), resolvability check (confirmable from visible code). Failure routes: downgrade tier, recalibrate severity, or block.
-
Protocol 2: Cross-Agent Convergence Check (review mode)
Applied during pressure-testing when 3+ agents flag the same finding. Requires each agent to have cited the same
file:line for convergence to be treated as confirmation. Different lines = independent findings; same label ≠ confirmation.
-
Protocol 3: Verdict Integrity Check (review mode)
Applied after synthesis, before outputting the verdict. Lists every anchoring defect; recalibrates verdict if any anchoring defect was walked back in the same review body.
-
Canonical file updated
_shared/nando-intelligence.md now contains the complete Architect Intelligence section as source of truth. All three mode agents paste from it.
-
docs/nando.md updated
Verdict scale includes CONDITIONAL APPROVE; Lead Architect Intelligence listed in hard rules.