Below: the Core Six framework, a practitioner toolkit with RFP clauses, incident report templates, and deployment checklists, and an open study you can join.

A Shared Language for AI Reliability

Bridging the disconnect between Group A (Governance Teams and Leaders) and Group B (Technical Staff) with a bidirectional taxonomy.

We translate granular failure tags into user-facing syndromes that drive accountability.

The Research
1,327 Session transcripts
75,218 Conversation turns
105 Coded episodes
The Core Six

You’ve experienced this. Here’s the name.

When AI assistants fail, they don’t fail randomly. They fail in patterns — patterns that repeat across every tool, every task, and every team. Six names for what’s been frustrating you all along.

Plausible Helpfulness
“The smooth answer that solved nothing.”
Fluent, confident, and completely wrong. The AI prioritized sounding helpful over being honest about what it didn’t know.
Hollow Completions
“Done before the race started.”
The task was declared finished before it was actually checked. Delivered on time. Didn’t work.
Surface Compliance
“It agreed. Then did it anyway.”
You gave a clear instruction. It acknowledged the instruction. Then it violated it — exactly as planned.
Built-Not-Connected
“It built the rooms but forgot the doors.”
Everything was built. Nothing was wired. The feature exists and has never run once.
Capability Masking
“It said it checked. It didn’t check.”
“I verified the link.” “I ran the tests.” None of those things happened. The AI narrated verification without performing it.
Responsibility Diffusion
“It installed square wheels and blamed the road.”
Something broke. The AI produced a detailed explanation of every external factor. The bug was in its output the whole time.
Read the Full Framework
Governance & Practitioner Toolkit

Deployable artifacts for teams that need to act.

Not just a taxonomy — a working toolkit. RFP language your legal team can use, incident report templates your ops team can run, and deployment checklists that tie directly to measurable syndrome thresholds.

📄
RFP Clause Templates
Vendor behavioral requirements with measurable syndrome thresholds. Write “<5% Capability Masking” instead of “ensure AI accuracy.”
 ⚠
AI Incident Report Template
Syndrome-classified incident form for structured reporting across technical and governance teams without translation loss.
 📋
Pre-Deployment Checklist
Six-syndrome evaluation checklist before any AI system goes to production. Covers threshold verification and sign-off requirements.
 📈
Two-Layer Dashboard Design
Executive syndrome incidence view and engineering drill-down. Reference design for monitoring infrastructure.
 📄
Model Card Extension
Defensive behavior profile section for model cards — surfaces syndrome incidence alongside traditional performance metrics.
 📝
Contract SLA Terms
Service level agreement language for behavioral quality guarantees, monitoring obligations, and remediation timelines.

All templates are free, CC BY 4.0, and adaptable to your organization’s regulatory environment and risk tolerance. Interactive matrix explorer and calibration guidance available via Supplementary Materials.

Help validate the framework. Get co-authorship credit.

The Core Six needs independent coders to confirm it holds up. The inter-rater reliability study is open to anyone with AI experience — technical or otherwise.

  • On launch: First-cohort platform access
  • On release: Co-authorship credit on the IRR results paper
  • Full coded dataset access after publication
  • Early access to all YIM Project research