Context reconstruction
How much useful work is lost when an agent has to relearn the project before every meaningful change?
Research
Substratia treats memory as infrastructure: explicit records, durable checks, and compact handoffs that reduce the cost of future work.
How much useful work is lost when an agent has to relearn the project before every meaningful change?
What shape should a record take when it needs to survive model changes, tool changes, and repo handoffs?
Which features are worth a hosted dependency, and which are better as files, scripts, and tests?