feat(orchestrator): the memory write path and the landscape (v0.16.0) - #11
Merged
Conversation
v0.15.0 took the survey's taxonomy. This takes the rest, as two references under the spine: how an entry is made, changed and discarded, and what exists to build on and measure with. The formation table states each operation's cost rather than implying it — summarization is lossy by design, structured construction pays schema rigidity, latent is a black box, parametric cannot be precisely removed. Updating is separated from consolidation, and temporal annotation replaces deletion as the recommended shape because it keeps the decision reversible. The landscape file refuses to be a list that expires: frameworks are compared by the axes that separate them, and the build-or-adopt framing is that you own the judgement in both columns. Frontiers are marked as frontiers, with one acted on — memory operations as tool calls, because that is what makes the decision NOT to retrieve visible in a trace. Three mechanisms are named as deliberately absent rather than quietly added. The validator refused the first version: three files each repeated the pinned citation. The spine holds it now and the siblings name it. Co-Authored-By: Claude Opus 5 (1M context) <[email protected]>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
v0.15.0 took the survey's taxonomy. This takes the rest, as two references under the spine.
memory-lifecycle.md— the write path. Five formation operations with each cost stated: summarization lossy by design, structured construction pays schema rigidity, latent is a black box, parametric cannot be precisely removed. Summarization's two shapes fail differently, and summarizing by fixed window is partitioned summarization with the worst partition rule. Updating is separated from consolidation; temporal annotation replaces deletion, which keeps the stability–plasticity decision reversible.memory-landscape.md— build or adopt, measure, frontiers. ~25 frameworks give you an index and leave you the judgement. Adopt for the index, not for the judgement. The first eval question is not which benchmark — it is what fails if memory is silently disabled?One frontier acted on: memory operations as tool calls, because that makes the decision not to retrieve visible in the trace. A background module cannot show you that.
Three mechanisms named as deliberately absent — frequency-based forgetting, temporal annotation, dual-phase updating — rather than quietly added from a survey.
The validator refused the first version: three files each repeated the pinned citation. The spine holds it; the siblings name it.
0 GAP, 14 PASS · body 4708/4750 · validator OK.
🤖 Generated with Claude Code