# Changelog

## Score

- CAI 60 → 61 (+0.9)
- Rubric changed (rubric-2026.10.5 → rubric-2026.11.4) — scores are not directly comparable.

## Lenses

- Code Health 56 → 55 (-0.9)
- Architecture 97 → 97 (+0.0)
- Maturity 58 → 62 (+3.8)
- Readiness 74 → 75 (+1.1)
- Security 54 → 56 (+1.8)
- Performance 100 → 100 (+0.0)

## Resolved (119)

- Base-context workflow trigger runs with an unscoped token (.github/workflows/upload_pr_documentation.yml)
- CI installs an unverified third-party binary (.github/workflows/linting.yaml)
- Concentrated knowledge decay
- Duplicated block (10 lines × 2) (backends/candle/src/models/flash_modernbert.rs)
- Duplicated block (10 lines × 2) (router/src/http/server.rs)
- Duplicated block (10–11 lines × 4) (backends/candle/src/models/bert.rs)
- Duplicated block (11 lines × 10) (backends/candle/src/models/bert.rs)
- Duplicated block (11 lines × 2) (backends/candle/src/models/flash_qwen3.rs)
- Duplicated block (11 lines × 2) (router/src/grpc/server.rs)
- Duplicated block (11 lines × 3) (backends/candle/src/models/distilbert.rs)
- Duplicated block (11 lines × 3) (backends/candle/src/models/flash_qwen3.rs)
- Duplicated block (11–16 lines × 11) (backends/candle/src/models/bert.rs)
- Duplicated block (12 lines × 2) (backends/candle/src/models/flash_gte.rs)
- Duplicated block (12 lines × 4) (backends/candle/src/models/flash_bert.rs)
- Duplicated block (12 lines × 4) (backends/candle/src/models/flash_gte.rs)
- Duplicated block (12 lines × 7) (backends/candle/src/models/bert.rs)
- Duplicated block (13 lines × 2) (backends/candle/src/models/bert.rs)
- Duplicated block (13 lines × 6) (backends/candle/src/models/bert.rs)
- Duplicated block (14 lines × 2) (backends/candle/src/models/flash_mistral.rs)
- Duplicated block (14 lines × 3) (backends/candle/src/models/bert.rs)
- …and 99 more

## New (163)

- Critical CVE: [GHSA redacted] (backends/python/server/requirements-hpu.txt)
- Critical CVE: [GHSA redacted] (Cargo.lock)
- Critical vulnerability: [GHSA redacted] (backends/python/server/requirements-hpu.txt)
- Critical vulnerability: [GHSA redacted] (backends/python/server/requirements-intel.txt)
- Duplicated block (10 lines × 2) (router/src/http/server.rs)
- Duplicated block (10 lines × 3) (router/src/grpc/server.rs)
- Duplicated block (10 lines × 3) (router/src/http/server.rs)
- Duplicated block (10–11 lines × 4) (backends/candle/src/models/bert.rs)
- Duplicated block (11 lines × 10) (backends/candle/src/models/bert.rs)
- Duplicated block (11 lines × 2) (backends/candle/src/models/distilbert.rs)
- Duplicated block (11 lines × 2) (backends/candle/src/models/flash_qwen3.rs)
- Duplicated block (11 lines × 3) (backends/candle/src/models/flash_qwen3.rs)
- Duplicated block (11 lines × 5) (backends/candle/src/models/flash_bert.rs)
- Duplicated block (11–13 lines × 2) (backends/ort/src/lib.rs)
- Duplicated block (11–16 lines × 11) (backends/candle/src/models/bert.rs)
- Duplicated block (12 lines × 2) (backends/candle/src/models/flash_modernbert.rs)
- Duplicated block (12 lines × 4) (backends/candle/src/models/flash_gte.rs)
- Duplicated block (12 lines × 7) (backends/candle/src/models/bert.rs)
- Duplicated block (13 lines × 10) (backends/candle/src/models/bert.rs)
- Duplicated block (13 lines × 6) (backends/candle/src/models/bert.rs)
- …and 143 more

## Changes since last survey

- 26 commits — 20 feature/other, 6 fixes

## By area

- backends/candle — 8 commits
- router/src — 5 commits
- .github/workflows — 3 commits
- core/src — 3 commits
- (root) — 2 commits
- (repo) — 1 commit
- backends/python — 1 commit
- backends/src — 1 commit
- docs/source — 1 commit
- router/tests — 1 commit

## Notable commits

- fix: Fix CUDA_ERROR_INVALID_VALUE in Nomic MoE router for batches > 65535 tokens
- fix: Fix Gemma3 attention masks in unpadded batches (#943)
- fix: Fix Qwen3 causal masking for equal-length batches (#883)
- fix: Fix batched DeBERTa-v2 scores: mask padded keys in attention (#954)
- fix: Fix single embed_sparse and embed_all gRPC request metrics
- fix: fix: panic on NaN scores in predict softmax (infer.rs) (#932)
- change: Add HTTP rerank endpoint coverage (#878)
- change: Add OTLP HTTP/protobuf export with HTTPS support (#940)
- change: Add OpenTelemetry trace IDs to JSON logs (#935)
- change: Add a configurable HTTP route prefix (#894)
- change: Authenticate protoc setup in CI (#963)
- change: Bump doc-builder to get the docs machine-translated (#956)
- change: Clarify local installation and CPU portability (#781)
- change: Do not panic when the backend channel is full (#952)
- change: Improve Prometheus histogram resolution (#847)
- change: Keep requests in te_queue_size until their batch is dispatched
- change: Rebuild the docs when their workflow changes (#960)
- change: Reduce peak memory of the Qwen3 causal attention bias (#953)
- change: Respect truncation direction when cutting long inputs by characters (#955)
- change: Restore ModernBERT CUDA attention masks (#962)
- …and 6 more

## Architecture

- Unchanged — 0 containers · 1 contexts · 0 edges
