chore(progress): mark follow-up #5 (cold-fallback torn-line) done

PROGRESS.md

@@ -402,13 +402,13 @@ deploy/scripts/run_e2e_sweep.sh --out-dir /tmp/sweep-$(date +%s) --soak-secs 120
    through to capability matching or the cold tier. Adding
    patterns is mechanical but expands what queries the warm
    tier can answer.
-5. **Cold reader is intolerant of torn last lines.** Under
-   concurrent producer write + reader scan, the §5.2
-   `parse_jsonl` path failed on a torn last line. A 5-line
-   change in
-   `asap-query-engine/src/drivers/query/fallback/cold_store/format.rs`
-   to drop a malformed trailing line + warn would unblock soaks
-   that don't pause writes before snapshotting.
+5. **~~Cold reader is intolerant of torn last lines.~~**
+   **Done 2026-05-05** ([ASAPQuery-backend#80](https://github.com/ProjectASAP/ASAPQuery-backend/pull/80)).
+   `parse_jsonl_at` tolerates a torn trailing line + warn-logs
+   the part-file path; mid-file corruption still hard-errors.
+   Two new pin tests (`parse_jsonl_ignores_torn_trailing_line`,
+   `parse_jsonl_errors_on_mid_file_corruption`) lock both
+   shapes against future regression.
 6. **Reducer runs offline; doesn't need the backend live.** That's
    fine for accuracy claims, but PromQL semantics are easy to
    drift from the engine. Add a self-check that runs the same

controller/build.rs

@@ -2,7 +2,7 @@ fn main() -> Result<(), Box<dyn std::error::Error>> {
     prost_build::compile_protos(&["proto/opamp.proto"], &["proto/"])?;
     // asap.runtime.v1.RuntimeSamples service — receives
     // PushExporter batches from agents. Must stay in lockstep
-    // with `sketchlib-bench/sketch-runtime/proto/feedback.proto`.
+    // with `sketch-bench/sketch-runtime/proto/feedback.proto`.
     tonic_build::configure()
         .build_server(true)
         .build_client(false)

controller/src/runtime_samples.rs

@@ -15,7 +15,7 @@
 //! the agent — critical when the decision loop is real-time.
 //! See the design discussion thread for the full trade-off;
 //! the summary is in
-//! [`sketch-runtime::exporter::grpc`](https://github.com/ProjectASAP/sketchlib-bench/blob/main/sketch-runtime/src/exporter/grpc.rs).
+//! [`sketch-runtime::exporter::grpc`](https://github.com/ProjectASAP/sketch-bench/blob/main/sketch-runtime/src/exporter/grpc.rs).
 //!
 //! ## Why a ring buffer, not a stream
 //!

docs/adr/adr-0001-retire-sketch-core.md

@@ -15,12 +15,12 @@ algorithm crates with overlapping responsibilities:
 
 - `asap_sketchlib` — the canonical algorithm crate, holding
   `DDSketch`, `KLL`, `HLL`, `Count`, `CountMin`, `CMSHeap`, etc.
-  Already shared with `sketchlib-bench` and external benchmark
+  Already shared with `sketch-bench` and external benchmark
   harnesses.
 - `sketch-core` — a wrapper crate vendored into
   `ASAPQuery-backend/asap-common/sketch-core/` (and its sibling
   forks in `ASAPQuery/asap-common/sketch-core/` and
-  `sketchlib-bench/sketch-core/`). It re-exposed the algorithms
+  `sketch-bench/sketch-core/`). It re-exposed the algorithms
   with ASAP-specific wire-format types (`DdSketchState`,
   `CountSketchState`, etc.), `apply_delta` methods, and the
   `Strategy::Legacy` vs `Strategy::Sketchlib` `ImplMode` switch.
@@ -66,7 +66,7 @@ exercised in production paths, the switch became dead weight.
      decision 3).
 
 2. **Drop the three on-disk `sketch-core` forks.** All consumers
-   (`ASAPQuery`, `ASAPQuery-backend`, `sketchlib-bench`) depend
+   (`ASAPQuery`, `ASAPQuery-backend`, `sketch-bench`) depend
    on `asap_sketchlib` directly via git URL.
 
 3. **Drop the `ImplMode` (Legacy / Sketchlib) dispatch.** Always
@@ -117,7 +117,7 @@ exercised in production paths, the switch became dead weight.
   `asap_sketchlib` are now the *only* defense against algorithm
   drift. The R1 risk in the design doc explicitly calls this
   out; mitigation (statistical-output cross-language harness in
-  `sketchlib-bench`) is tracked for follow-up.
+  `sketch-bench`) is tracked for follow-up.
 
 ### Compatibility
 

docs/design-asap-edge-framework.md

@@ -671,7 +671,7 @@ hash seeds, storage growth, and protobuf field ordering
 legitimately differ across runtimes). Same input stream into both
 runtimes; compare P50/P90/P99/count/sum/cardinality/top-K within
 each sketch's published error bound (DDSketch α, KLL rank error,
-HLL standard error). Harness in `sketchlib-bench` (the natural
+HLL standard error). Harness in `sketch-bench` (the natural
 cross-language home; already runs Go vs Rust microbenchmarks).
 Limit to two reference impls long-term: Go and Rust, no third
 language.
@@ -739,7 +739,7 @@ that's not how any of these platforms work.
 
 - [ ] Approve §3–§8 abstractions and naming.
 - [ ] Confirm R1 mitigation: statistical-output harness in
-      `sketchlib-bench`.
+      `sketch-bench`.
 - [ ] Decide repo layout for `asap-precompute-{go,rs}`: separate
       repos or subdirectories of `ASAPCollector`. Recommendation:
       subdirectories until a second consumer wants the runtime.

docs/paper-outline.md

@@ -67,7 +67,7 @@ workload + SLAs.
        answers vs. truth and computes per-row relative
        error / top-K recall; bound-derivation crib in
        `ASAPQuery-backend/TODO.md` "Accuracy-profile library
-       per sketch type" + `sketchlib-bench/docs/DESIGN.md`.
+       per sketch type" + `sketch-bench/docs/DESIGN.md`.
   5. **Fast query computation / short query latency.**
      Backend p50 / p99 query latency is production-usable.
      Headline target: ≤2× warm-hot for cold-fallback;