You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Confidence: 95% · Found by: architecture review (slice S7). Doc-only ADR — no code change.
The entity-coreference + synthesis subsystem (~5,500 LOC) is a faithful, well-built port of real, wiredlossless-claw v4.1 functionality — NOT speculative scope. But a 540★ production sibling (hermes-lcm) ships with neither subsystem and works. The subsystem is load-bearing for the port's verbatim-pr-613 contract (ADR-030) yet deferrable for a minimal LCM product — an unexamined assumption.
Decision for ADR-036: record that v0.1.0 keeps both subsystems for lossless-claw port fidelity, AND define a falsifiable kill-criterion: if Epic-09 eval shows entity recall contributes < X% measured retrieval lift on real Hermes traffic, demote both to an optional plugin-extra in v0.3. Converts an unexamined assumption into a tracked, falsifiable decision.
Confidence: 95% · Found by: architecture review (slice S7). Doc-only ADR — no code change.
The entity-coreference + synthesis subsystem (~5,500 LOC) is a faithful, well-built port of real, wired
lossless-clawv4.1 functionality — NOT speculative scope. But a 540★ production sibling (hermes-lcm) ships with neither subsystem and works. The subsystem is load-bearing for the port's verbatim-pr-613 contract (ADR-030) yet deferrable for a minimal LCM product — an unexamined assumption.Decision for ADR-036: record that v0.1.0 keeps both subsystems for
lossless-clawport fidelity, AND define a falsifiable kill-criterion: if Epic-09 eval shows entity recall contributes < X% measured retrieval lift on real Hermes traffic, demote both to an optional plugin-extra in v0.3. Converts an unexamined assumption into a tracked, falsifiable decision.