Rearm consolidation drift and bound Natural active work - #6
Conversation
There was a problem hiding this comment.
Pull request overview
This PR updates the internal consolidation-throughput hint latch so that, after a maintenance acknowledgment disarms the hint, it can be rearmed not only by recovery, but also by a material sustained throughput drawdown (with an F/S/T-derived “material range” guard) to prevent slow degradation from staying permanently silent.
Changes:
- Add an F/S/T-derived drift rearm threshold (
p_drift = 1 - p) and require a material relative throughput range before rearming. - Update the disarmed rearm logic to allow rearming on either recovery (
z >= p_rearm) or sustained degradation (z <= p) once the excursion is material. - Extend the consolidation throughput tests to cover drift rearming and jitter suppression, and update the paper contract + regenerated manuscript.
Reviewed changes
Copilot reviewed 5 out of 5 changed files in this pull request and generated no comments.
Show a summary per file
| File | Description |
|---|---|
| tests/operations_consolidation.test.cpp | Adds monotonicity assertions for the new drift fraction and adds coverage for drift-based rearming vs. stable jitter remaining disarmed. |
| src/operations/consolidation_throughput_state_internal.hpp | Introduces DriftRearmFraction() and updates Observe() to gate rearming on material relative-range and allow both recovery-edge and drift-edge rearming. |
| docs/paper/sections/7_consolidation.qmd | Updates the consolidation contract to document p_drift and the new material-range + dual-edge rearm semantics. |
| docs/paper/sections/10_implementation.qmd | Documents the new drift-based rearm behavior and the jitter-suppression guard in the implementation section. |
| docs/paper/_manuscript/index.md | Regenerates the manuscript to reflect the updated consolidation contract and implementation description. |
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: d3df85e3b1
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
|
Implementation/review checkpoint for exact head
Nonclaims remain whole-engine raw-time reset, bounded whole-engine restart, Durable plateau, production-wide boundedness, merge, release, deployment, and publication. Exact-head CI is still in progress. |
|
Pushed exact-head CI/review repair as f7f0fc5. What changed:
Current proof:
The RIF review thread is replied to and resolved. Replacement exact-head CI is now the remaining gate, including Linux sanitizer proof. This does not claim whole-engine reset, bounded restart, Durable plateau, production-wide boundedness, merge, release, or deployment. |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: b7f9e27e01
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
|
Pushed follow-up native/sanitizer repair as 764e4af. The three failed jobs shared one test issue: the regression assumed a platform-specific HNSW random-level outcome. The sanitizer job additionally found pinned hnswlib issuing a zero-byte memcpy with a potentially null empty-vector destination. Repair:
Proof:
Replacement exact-head CI now owns Linux native and sanitizer confirmation. No review requests or unresolved threads exist. No merge, release, deployment, bounded-restart, Durable-plateau, or production-wide boundedness claim is made. |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 764e4af202
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 89a06084dd
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 25cf7b130e
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: cca844d758
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: d2213ea42f
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
|
@codex review |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 3f4d2e5e7c
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
|
@codex review |
|
Codex Review: Didn't find any major issues. Bravo. Reviewed commit: ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
If Codex has suggestions, it will comment; otherwise it will react with 👍. Codex can also answer questions or update the PR. Try commenting "@codex address that feedback". |
Rearm consolidation drift and bound Natural active work
Summary
source_id-agnosticProduction retrieval contract
C = lround(256 + 256F + 128S + 128T),B = lround(64 + 64F + 32S + 32T),A = 2C + 2B,R = max(2, B / 16)8C; retrieval-active queries advance byRto9CA; total query rows are bounded by9C + A; restart opens at9C.5/.5/.5:C=512,B=128,A=1280,R=8, floor4096, peak4608, total5888;129is only logicalB+1Retrieval quality
>=511/512)>=0.999; recall@16>=0.998; semantic coverage>=0.95Result: 4,608/4,608 top-1 controls, recall@16
1.0, semantic coverage1.0, deterministic ties, zero miss clusters, and passing text/audio/image plus shared/opaque-source invariance.Verification
artifacts/flat_storage_cost/sqlite-hnsw-production-sawtooth-knob-ablation-v2.jsongit diff --checkpassThe 30,380-packet maturity run passed 511/512 top-1, recall@16
0.999265, semantic coverage0.998571, deterministic ties, 19/20 material cycle resets, and no work-bound violations. The 15,695-packet Natural and 2,016-message Durable replays each passed 512/512 controls; the Durable horizon is insufficient for plateau proof.Review and boundaries