Document deterministic search and LLM boundary - #176
Draft
PhysShell wants to merge 2 commits into
Draft
Conversation
|
Important Review skippedDraft detected. Please check the settings in the CodeRabbit UI or the ⚙️ Run configurationConfiguration used: defaults Review profile: CHILL Plan: Pro Plus Run ID: You can disable this status message by setting the Use the checkbox below for a quick retry:
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
PhysShell
pushed a commit
that referenced
this pull request
Aug 17, 2026
A caller that shows three alternatives and then asks for five must not watch the first three renumber themselves. The accepted engine already satisfies this — neither the enumeration order nor the greedy diversity filter reads `k` — so this is a **characterization test**, not a red→green fix: no production code changes, and the acceptance given at `7d0c0cb` is not reopened. The property was specified and left unasserted. The deterministic-search proposal (#176, still a draft) asks for it in as many words — "requesting more results never changes the prefix already returned" — and "true by construction" is exactly the kind of claim that quietly stops holding one refactor later without anything failing. Compared per rank, not merely per path list: ordinals, `total_cost` bits, and the full `Debug` rendering, which carries every step and edge with its axes, rationale entries, weights and provenance. `f64` renders in shortest round-trip form, so a cost differing in one bit renders differently — "the same path" cannot quietly become "the same indices, differently computed". Swept over all 120 width vectors for 1–4 layers over widths 1–3, every `min_distance`, and every `k` below the widest request, with an anti-vacuity floor on the number of ranks actually compared. `exhausted` is deliberately excluded from the compared prefix: it describes the *search*, not the paths, and a request for fewer alternatives can legitimately stop before the space runs out. Verification: `cargo test --workspace` 1454 passed / 0 failed (1453 before, +1); clippy `--workspace --all-targets -D warnings` clean; `cargo fmt --all --check` clean. The diff touches one test file and no production code. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01TvcPGrQsgY4HqzmdDcRqvu
This was referenced Aug 17, 2026
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.
What changed
ArrangementIntentV1as the only optional LLM-facing generation contract;docs/proposals/README.md.Why
Griff already has deterministic layered DP, explainable scoring, a hard-constraint direction, and a functioning offline oracle. The missing architectural bridge is how to produce reproducible alternatives, introduce diversity without disguising randomness as search, and permit an LLM to describe intent without giving it authority over canonical score material or validity.
This proposal records that bridge without assigning a roadmap stage or silently accepting implementation decisions.
Impact
Docs only. No production behaviour, dependencies, workspace membership, score schema, or roadmap stage changes.
Validation
Scoredconsumers;lab/oracle and its July 2026 audit report;docs/proposals/README.mdlifecycle rules.