Stretch item split out of the #39 scatter-gather epic when that epic was closed core-complete.
What
A dedicated TangleBrain-side delegate_reduce MCP tool that synthesizes/stitches delegate_many
results, instead of the orchestrator doing the synthesis itself.
Why deferred (and why it may never ship)
Slice 4 of #39 deliberately resolved as no-code: synthesis stays the orchestrator's job because
it holds the original task context and produces a better reduce than a fresh backend would. A purely
mechanical stitch is already offloadable with an ordinary delegate(task=…) call — so a dedicated
reducer tool is redundant with existing capability.
Entry criterion — DATA-GATED
Build this only if the delegate observability/metering shipped in v0.14.0 (slice 6) shows that
frontier-side synthesis is a real bottleneck in practice. Absent that data, this stays closed by
design. Do not build on reflex.
Stretch item split out of the #39 scatter-gather epic when that epic was closed core-complete.
What
A dedicated TangleBrain-side
delegate_reduceMCP tool that synthesizes/stitchesdelegate_manyresults, instead of the orchestrator doing the synthesis itself.
Why deferred (and why it may never ship)
Slice 4 of #39 deliberately resolved as no-code: synthesis stays the orchestrator's job because
it holds the original task context and produces a better reduce than a fresh backend would. A purely
mechanical stitch is already offloadable with an ordinary
delegate(task=…)call — so a dedicatedreducer tool is redundant with existing capability.
Entry criterion — DATA-GATED
Build this only if the delegate observability/metering shipped in v0.14.0 (slice 6) shows that
frontier-side synthesis is a real bottleneck in practice. Absent that data, this stays closed by
design. Do not build on reflex.