Public method disclosure · J. Ryan Clendenin
Living Context Model
An LCM keeps durable institutional meaning outside replaceable AI models and regenerates the minimum useful context needed for each computation.
The problem
Foundation models can generate, reason, search, summarize, critique, and act, but a model session is a poor place to put durable institutional continuity. Sessions end. Providers change. Search systems disagree. Fluent output can still be wrong in consequential ways.
The scarce resource is governed context: what the institution currently knows, why it believes it, which relationships and authorities apply, what remains unresolved, and what consequence an exterior computation is allowed to have.
The LCM cycle
governed state → bounded context projection → replaceable computation → clocked Capture → correspondence → pragmatic answer or validated delta → explicit admission when warranted → regenerated context
Method principles
- Reconstruct before asking. Start from current institutional state instead of a provider transcript.
- Project only useful context. Treat context dose, order, modality, evidence, and refresh as measurable treatment variables.
- Keep computation replaceable. The model or search engine does not become institutional memory because it produced an answer.
- Clock the return. Preserve what was observed and when before interpreting what it means.
- Cross-reference meaning. Compare claims and useful differentials to evidence and governing context; model consensus alone is not proof.
- Separate answers from canon. A useful answer may be warranted before an institutional admission is.
- Learn from failure without contamination. Rejected output can become Compost: learning-bearing, permanently non-governing residue.
- Regenerate and substitute. Preserve validated progress so a different provider can continue without inherited provider-local history.
What LCM is not
LCM is not defined by generic multi-agent orchestration, retrieval-augmented generation, prompt compression, external memory, confidence voting, AI-checking-AI, or a particular context-window size. Those techniques may be components; the method concerns governed reconstruction, consequence, correspondence, admission, and regenerative continuity.
Public / private boundary
This disclosure publishes method before mechanism. C&C does not publish private institutional records or the protected implementation learning behind the method.