ADR-0023: delta-aware adjudication (state = context, delta = question) — Proposed#205
Merged
Merged
Conversation
…question) [Proposed] Draft for review (JEF-391). From the 77-min churn capture: 85% of re-judges are genuinely -new fingerprints (new reachable objectives as ephemeral pods cycle) — correct to judge, but the model re-derives the whole world each time instead of judging the change. The engine already computes the delta (graph::delta, first_seen, prev_posture) but it never reaches the prompt. Proposes: full state as context + a "what's NEW since the last decisive verdict" section as the question; re-judge on ADDITIVE deltas only (subtractive flicker stops re-judging); full state always present (correctness guard). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01VtjoJttCvBY4dzCoE4f9vP
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Draft ADR for JEF-391, opened for review to pin the design before building.
The insight (from the 77-min churn capture, JEF-387): 85% of re-judges are genuinely-new fingerprints — new reachable objectives as ephemeral pods (CI runners) cycle. The churn is correct (new surface = re-judge), but the model re-derives the entire world each pass instead of judging the change. The engine already computes the delta (
graph::delta,first_seen,prev_posture) — it just never reaches the adjudication prompt.The proposal: full state stays as the context; add a
Changes since the last decisive verdictsection as the question. Re-judge on additive deltas only (subtractive flicker — a peer aging out — stops re-judging at the root). Correctness guard: the full state is always in the prompt (a new node can light up an old chain), so we never judge a diff in isolation.Please review the Decision + the open questions (baseline location, delta projection, cache/verdict semantics). This is Proposed — not to merge as Accepted until the design's agreed.
🤖 Generated with Claude Code