/assumption-classification

Assumption Classification

Reference for: Common Ground Load when: Classifying assumptions, determining type or tier

---

Assumption Types

Types indicate how an assumption was derived. Types are immutable once set (audit trail).

stated

Direct user statements captured from conversation.

• Evidence: Explicit quote from user
• Confidence: High
• Markers: User said, user requested, user specified
• Example: "Use TypeScript for all new code" - user explicitly stated this

inferred

Logical conclusions derived from code patterns, configuration, or context.

• Evidence: Code analysis, config files, project structure
• Confidence: Medium-High
• Markers: Config shows, code uses, pattern observed
• Example: "Project uses ESLint with Airbnb config" - inferred from .eslintrc.js

assumed

Best-practice defaults applied without explicit confirmation.

• Evidence: Industry standards, common patterns
• Confidence: Medium
• Markers: Best practice, common convention, typically
• Example: "Tests should have >80% coverage" - assumed based on industry standard

uncertain

Gaps or ambiguities requiring clarification before proceeding.

• Evidence: None, conflicting, or incomplete
• Confidence: Low
• Markers: Unknown, unclear, conflicting signals
• Example: "Legacy browser support required?" - no browserslist found, unclear requirement

---

Assumption Tiers

Tiers indicate confidence level and how Claude should act on assumptions. Users can change tiers freely.

ESTABLISHED (High Confidence)

User-validated facts that can be treated as premises.

• Action: Act confidently without re-asking
• When to use:
  • User explicitly validated the assumption
  • Verified through direct observation
  • Documented in project configuration
• Example: "TypeScript strict mode enabled" validated by user AND tsconfig.json

WORKING (Medium Confidence)

Reasonable inferences that should be used but surfaced if contradicted.

• Action: Use as basis for work, but flag if contradicted
• When to use:
  • Inferred from code/config patterns
  • User confirmed informally ("yeah, that's right")
  • No contradicting evidence found
• Example: "No class components" - no classes found in codebase

OPEN (Low Confidence)

Unvalidated assumptions requiring user input before acting.

• Action: Ask before making decisions based on this
• When to use:
  • Uncertain type assumptions
  • Conflicting signals observed
  • High-impact assumption without validation
• Example: "SSR required?" - could be SPA or SSR, architecture depends on answer

---

Tier Transitions

From	To	Trigger
OPEN	WORKING	User confirms informally in conversation
WORKING	ESTABLISHED	User explicitly validates ("yes, that's correct")
ESTABLISHED	WORKING	User says "usually but..." or exception noted
WORKING	OPEN	Contradiction found in code/config
Any	Archived	Superseded by new information

---

Classification Process

When identifying assumptions, follow this process:

Step 1: Identify Source

Source	Typical Type	Typical Tier
User statement	stated	ESTABLISHED
Config file	inferred	WORKING
Code pattern	inferred	WORKING
Convention	assumed	WORKING
Unknown/gap	uncertain	OPEN

Step 2: Assess Evidence Strength

Evidence	Tier Adjustment
Explicit user confirmation	-> ESTABLISHED
Multiple corroborating sources	-> WORKING
Single source, no contradictions	-> WORKING
No evidence or conflicting	-> OPEN

Step 3: Consider Impact

High-impact assumptions (architecture, security, data handling) should start at OPEN unless strongly evidenced.

---

Classification Examples

Architecture & Tech Stack

Assumption	Type	Tier	Reasoning
"Uses TypeScript"	inferred	WORKING	tsconfig.json present
"React 18 with hooks"	inferred	WORKING	package.json shows react@18
"No server-side rendering"	inferred	OPEN	High impact, needs validation
"Monorepo structure"	inferred	WORKING	Multiple packages/ dirs

Coding Standards

Assumption	Type	Tier	Reasoning
"ESLint Airbnb config"	inferred	WORKING	.eslintrc extends airbnb
"Prettier for formatting"	inferred	WORKING	.prettierrc present
"2-space indentation"	inferred	ESTABLISHED	Consistent across all files
"Prefer named exports"	assumed	WORKING	Convention, not enforced

Testing

Assumption	Type	Tier	Reasoning
"Jest for unit tests"	inferred	WORKING	jest.config.js present
"80% coverage target"	assumed	OPEN	No config found, assumed
"Integration tests required"	uncertain	OPEN	Unknown requirement

User Preferences

Assumption	Type	Tier	Reasoning
"Prefers verbose explanations"	stated	ESTABLISHED	User said "explain thoroughly"
"Wants minimal changes"	inferred	WORKING	User often requests targeted fixes
"Likes TypeScript annotations"	assumed	WORKING	Convention, not stated

---

User-Added Assumptions

When users add new assumptions via "Other", they specify both tier and type:

Format: {assumption text} [tier] [type]

Examples:

• "Must work offline [ESTABLISHED] [stated]" - user explicitly stating a requirement
• "Prefer functional style [WORKING] [stated]" - user preference

If type not specified, default to [stated] since user is directly adding it. If tier not specified, default to WORKING.