Richard Ketelsen PRO
CRAFTFramework
AI & ML interests
CRAFT applies structured programming principles to AI conversations. Think recipes for your workflows.
Recent Activity
repliedto their post 1 day ago
# A measurable QA layer for LLM working sessions
Hallucination is treated as an inherent LLM failure mode, but most production workflows respond to it with "be careful, double-check things." That doesn't scale past a handful of sessions, and it doesn't catch the failure mode that does the most damage: confident reconstruction in late-session context.
CRAFT for Cowork takes a structural approach. The QA framework runs verification at four levels — individual claims, recipe execution, file integrity, and cross-session consistency — and treats trust as a measurable property rather than a vibe.
**The four-gate verification sub-routine** (RCP-CWK-024) runs before any recipe reports a result:
1. *File-pointability* — claim traceable to a specific file
2. *Read-vs-reconstructed* — was data actually read this session
3. *Lessons-Learned conflict* — contradicts documented prior truth
4. *Untested assumption* — verified vs. assumed
**Confidence scoring** grades every factual claim 0-100 against a source hierarchy: evidence read from files (80-100), tool-output observation (50-79), design intent (30-49), pure reasoning (0-29). A 10-point penalty applies past 70% token usage to correct for late-session reliability decay.
**Cross-session consistency** is enforced by a longitudinal audit recipe (RCP-CWK-036) run every 5-10 sessions. It has caught ~40% drift in tracking-file state tables — drift that would otherwise propagate as silent ground truth.
**Concrete result:** A factual claim validation pass caught nine pre-publication content files referencing the framework with an incorrect license descriptor. Single pass, all nine corrected.
This is week 5 of an 8-week capability spotlight. CRAFT for Cowork is a free public beta.
Repository: https://github.com/CRAFTFramework/craft-framework
License: https://craftframework.ai/craft-license/ (Spec under BSL 1.1, converts to Apache 2.0 on Jan 1, 2029; content proprietary) posted an update 1 day ago
# A measurable QA layer for LLM working sessions
Hallucination is treated as an inherent LLM failure mode, but most production workflows respond to it with "be careful, double-check things." That doesn't scale past a handful of sessions, and it doesn't catch the failure mode that does the most damage: confident reconstruction in late-session context.
CRAFT for Cowork takes a structural approach. The QA framework runs verification at four levels — individual claims, recipe execution, file integrity, and cross-session consistency — and treats trust as a measurable property rather than a vibe.
**The four-gate verification sub-routine** (RCP-CWK-024) runs before any recipe reports a result:
1. *File-pointability* — claim traceable to a specific file
2. *Read-vs-reconstructed* — was data actually read this session
3. *Lessons-Learned conflict* — contradicts documented prior truth
4. *Untested assumption* — verified vs. assumed
**Confidence scoring** grades every factual claim 0-100 against a source hierarchy: evidence read from files (80-100), tool-output observation (50-79), design intent (30-49), pure reasoning (0-29). A 10-point penalty applies past 70% token usage to correct for late-session reliability decay.
**Cross-session consistency** is enforced by a longitudinal audit recipe (RCP-CWK-036) run every 5-10 sessions. It has caught ~40% drift in tracking-file state tables — drift that would otherwise propagate as silent ground truth.
**Concrete result:** A factual claim validation pass caught nine pre-publication content files referencing the framework with an incorrect license descriptor. Single pass, all nine corrected.
This is week 5 of an 8-week capability spotlight. CRAFT for Cowork is a free public beta.
Repository: https://github.com/CRAFTFramework/craft-framework
License: https://craftframework.ai/craft-license/ (Spec under BSL 1.1, converts to Apache 2.0 on Jan 1, 2029; content proprietary) posted an update 29 days ago
# CRAFT Session Handoffs: Solving the Stateless AI Collaboration Problem
AI desktop tools like Claude Cowork are increasingly capable for multi-session projects — but they're architecturally stateless. Each session starts with an empty context window, losing all prior decisions, file states, and project context. The human becomes a manual bridge between sessions.
## Approach: Structured Handoff Protocol
CRAFT addresses this with a multi-format handoff protocol that captures complete session state in structured text files. At session end, a handoff recipe records accomplishments, decisions, file manifests, git state, active priorities, open questions, lessons learned, and persona configuration. At session start, an initialization recipe reads the latest handoff and restores full context.
The system uses four purpose-built recipes: Session Initialization (CWK-001), Direct Handoff (CWK-002), Device-Switch Handoff (CWK-002a), and Mid-Session Checkpoint (CWK-003). A two-layer initialization architecture — static CLAUDE.md bootstrap plus dynamic recipe state — provides redundancy against stale components.
A 4-layer propagation chain ensures persistent behavioral directives survive across sessions and framework updates: CLAUDE.md, project file, generator template, and enforcement recipe.
## Empirical Results
The system was validated during CRAFT's own development: 71 consecutive handoffs (H001–H071) with zero data loss. The handoff protocol survived 5+ context compaction events, multiple application crashes, and a cross-device transfer. Each session relied on the previous session's handoff for complete context restoration.
## Key Design Decisions
Handoff files are stored as structured plain text on the user's local machine — no cloud dependency, full data ownership. The handoff itself triggers an automatic git commit, providing version history of the project's session state over time.
Free public beta: [github.com/CRAFTFramework/craft-framework]Organizations
