diff --git a/.github/workflows/benchmarks.yml b/.github/workflows/benchmarks.yml
index b0946c6..8071bfd 100644
--- a/.github/workflows/benchmarks.yml
+++ b/.github/workflows/benchmarks.yml
@@ -120,8 +120,10 @@ jobs:
           echo "NAME=$name" >> "$GITHUB_ENV"
           if [ "$LANGUAGE" = rust ]; then
             benchmark=(rust/target/release/sqlite-vs-doublets)
+            toolchain="$(rustc --version)"
           else
             benchmark=(dotnet csharp/SQLiteVSDoublets/bin/Release/net10.0/sqlite-vs-doublets.dll)
+            toolchain=".NET SDK $(dotnet --version)"
           fi
           "${benchmark[@]}" "$CATEGORY" "$BITS" "$SIZE" \
             --directory "$RUNNER_TEMP/storage" --output "$RUNNER_TEMP/measured.json"
@@ -133,6 +135,7 @@ jobs:
             --arg date "$(date -u +%Y-%m-%d)" --arg machine "$machine" \
             '. + {run_url: $run_url, source_sha: $source_sha, tree_sha: $tree_sha, date: $date, machine: $machine}' \
             "$RUNNER_TEMP/measured.json" > "results/$name.json"
+          python3 scripts/benchmark_provenance.py "results/$name.json" --toolchain "$toolchain"
       - uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7
         with:
           name: results-${{ env.NAME }}
diff --git a/.github/workflows/quality.yml b/.github/workflows/quality.yml
index 9582669..a04b4c9 100644
--- a/.github/workflows/quality.yml
+++ b/.github/workflows/quality.yml
@@ -4,9 +4,9 @@ on:
   workflow_dispatch:
   push:
     branches: [main]
-    paths: ['.github/**', 'scripts/**', 'experiments/**', 'pyproject.toml', 'README*.md', 'CONTRIBUTING.md', 'cpp/**', '.secretlintrc.json', '.pre-commit-config.yaml', '.markdownlint.json', '.codacy.yml']
+    paths: ['.github/**', 'scripts/**', 'experiments/**', 'pyproject.toml', 'README*.md', 'CONTRIBUTING.md', '.secretlintrc.json', '.pre-commit-config.yaml', '.markdownlint.json', '.codacy.yml']
   pull_request:
-    paths: ['.github/**', 'scripts/**', 'experiments/**', 'pyproject.toml', 'README*.md', 'CONTRIBUTING.md', 'cpp/**', '.secretlintrc.json', '.pre-commit-config.yaml', '.markdownlint.json', '.codacy.yml']
+    paths: ['.github/**', 'scripts/**', 'experiments/**', 'pyproject.toml', 'README*.md', 'CONTRIBUTING.md', '.secretlintrc.json', '.pre-commit-config.yaml', '.markdownlint.json', '.codacy.yml']
 
 permissions:
   contents: read
@@ -59,7 +59,7 @@ jobs:
       - run: python -m unittest discover -s experiments/ci -v
       - name: Verify preserved investigation evidence
         run: python experiments/ci/package_evidence.py dev/log/issues/110/pulls/111 --verify
-      - run: npx --yes markdownlint-cli@0.49.1 README.md README.ru.md CONTRIBUTING.md cpp/README.md
+      - run: npx --yes markdownlint-cli@0.49.1 README.md README.ru.md CONTRIBUTING.md
       - name: Scan maintained source for secrets
         run: git ls-files -z -- ':!dev/log/**' ':!docs/benchmarks/**' ':!*.png' | xargs -0 npx --yes -p secretlint@13.0.7 -p @secretlint/secretlint-rule-preset-recommend@13.0.7 secretlint
 
diff --git a/.gitignore b/.gitignore
index c5bd78a..14e95b9 100644
--- a/.gitignore
+++ b/.gitignore
@@ -1,12 +1,5 @@
 # Root level gitignore for the repository
 # C# specific gitignore is in csharp/.gitignore
-# C++ specific files
-cpp/build/
-cpp/*.o
-cpp/*.a
-cpp/*.so
-cpp/*.exe
-
 # Common temporary files
 __pycache__/
 .venv/
diff --git a/CONTRIBUTING.md b/CONTRIBUTING.md
index 3c4d45c..89cf057 100644
--- a/CONTRIBUTING.md
+++ b/CONTRIBUTING.md
@@ -31,7 +31,7 @@ dotnet format whitespace csharp/SQLiteVSDoublets.slnx --verify-no-changes
 dotnet build csharp/SQLiteVSDoublets.slnx -c Release -warnaserror
 (cd csharp && dotnet test --project SQLiteVSDoublets.Tests -c Release \
   --no-build -- --timeout 10m)
-npx --yes markdownlint-cli@0.49.1 README.md README.ru.md CONTRIBUTING.md cpp/README.md
+npx --yes markdownlint-cli@0.49.1 README.md README.ru.md CONTRIBUTING.md
 ```
 
 Workflow validation also runs actionlint with ShellCheck and zizmor. The pinned
@@ -60,7 +60,6 @@ reports from the newly fetched branch on every retry.
 This repository contains benchmark executables, without a package release or
 container deployment pipeline. Package changesets, registry credentials, and
 container build steps should be added with an actual distributable component.
-The C++ directory is a placeholder; see [its prerequisites](cpp/README.md).
 
 Immutable investigation evidence is preserved in
 [the issue 110 investigation](dev/log/issues/110/pulls/111/ANALYSIS.md).
diff --git a/README.md b/README.md
index b267257..54de54b 100644
--- a/README.md
+++ b/README.md
@@ -89,6 +89,12 @@ repetitions do not overlap and the medians differ by more than 5%, otherwise
 it is `≈ same`. The file size is measured after creation; Doublets files are
 preallocated memory-mapped files, so small stores show the preallocation size.
 
+Each results group includes linear and logarithmic charts. Linear charts raise
+bars below 0.5% of the panel maximum for visibility; the tables retain exact
+times. Provenance records the Doublets library and Rust compiler or .NET SDK
+versions used for that run. Older reports without this metadata identify it
+as missing rather than attributing current toolchains to past measurements.
+
 Every table is measured by its own GitHub Actions job, with all variants on
 the same runner, so that a difference is never a difference between machines,
 by the [Benchmarks workflow](.github/workflows/benchmarks.yml): links with
@@ -113,6 +119,10 @@ dotnet run -c Release --project csharp/SQLiteVSDoublets -- \
   links 64 1000 --repetitions 3 \
   --variants SQLite_Memory,SystemDataSQLite_Memory \
   --output results/links-csharp-64-1000.json
+python3 scripts/benchmark_provenance.py results/links-rust-64-100000.json \
+  --toolchain "$(rustc --version)"
+python3 scripts/benchmark_provenance.py results/links-csharp-64-1000.json \
+  --toolchain ".NET SDK $(dotnet --version)"
 # print the tables, add --readme README.md --charts docs/benchmarks to update
 python3 scripts/benchmark_report.py results
 ```
@@ -173,7 +183,7 @@ Notes:
 
 ##### 100,000 links
 
-_10 repetitions after a warm-up, median time per operation. SQLite 3.53.2, Ubuntu 24.04.5 LTS, AMD EPYC 7763 64-Core Processor, 4 cores, 16 GiB, [GitHub Actions run](https://github.com/linksplatform/Comparisons.SQLiteVSDoublets/actions/runs/37381004025) on 2026-10-05._
+_10 repetitions after a warm-up, median time per operation. SQLite 3.53.2; doublets 0.5.0; rustc 1.99.0 (b940084d7 2026-09-28). Ubuntu 24.04.5 LTS, AMD EPYC 7763 64-Core Processor, 4 cores, 16 GiB, [GitHub Actions run](https://github.com/linksplatform/Comparisons.SQLiteVSDoublets/actions/runs/37381004025) on 2026-10-05._
 
 | Storage | Create | Read all | Read by id | Search (from, to) | Read by from | Read by to | Update | Delete | File size |
 | --- | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
@@ -186,7 +196,7 @@ _10 repetitions after a warm-up, median time per operation. SQLite 3.53.2, Ubunt
 
 ##### 1,000,000 links
 
-_3 repetitions after a warm-up, median time per operation. SQLite 3.53.2, Ubuntu 24.04.5 LTS, AMD EPYC 7763 64-Core Processor, 4 cores, 16 GiB, [GitHub Actions run](https://github.com/linksplatform/Comparisons.SQLiteVSDoublets/actions/runs/37381004025) on 2026-10-05._
+_3 repetitions after a warm-up, median time per operation. SQLite 3.53.2; doublets 0.5.0; rustc 1.99.0 (b940084d7 2026-09-28). Ubuntu 24.04.5 LTS, AMD EPYC 7763 64-Core Processor, 4 cores, 16 GiB, [GitHub Actions run](https://github.com/linksplatform/Comparisons.SQLiteVSDoublets/actions/runs/37381004025) on 2026-10-05._
 
 | Storage | Create | Read all | Read by id | Search (from, to) | Read by from | Read by to | Update | Delete | File size |
 | --- | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
@@ -199,7 +209,7 @@ _3 repetitions after a warm-up, median time per operation. SQLite 3.53.2, Ubuntu
 
 ##### 10,000,000 links
 
-_1 repetition after a warm-up, median time per operation. SQLite 3.53.2, Ubuntu 24.04.5 LTS, AMD EPYC 9V74 80-Core Processor, 4 cores, 16 GiB, [GitHub Actions run](https://github.com/linksplatform/Comparisons.SQLiteVSDoublets/actions/runs/37381004025) on 2026-10-05._
+_1 repetition after a warm-up, median time per operation. SQLite 3.53.2; doublets 0.5.0; rustc 1.99.0 (b940084d7 2026-09-28). Ubuntu 24.04.5 LTS, AMD EPYC 9V74 80-Core Processor, 4 cores, 16 GiB, [GitHub Actions run](https://github.com/linksplatform/Comparisons.SQLiteVSDoublets/actions/runs/37381004025) on 2026-10-05._
 
 | Storage | Create | Read all | Read by id | Search (from, to) | Read by from | Read by to | Update | Delete | File size |
 | --- | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
@@ -210,13 +220,15 @@ _1 repetition after a warm-up, median time per operation. SQLite 3.53.2, Ubuntu
 | Doublets Split Volatile | 337 ns (9.32× faster) | 2.83 ns (34.7× faster) | 30.9 ns (44.9× faster) | 238 ns (5.79× faster) | 171 ns (8.23× faster) | 170 ns (8.18× faster) | 689 ns (12.3× faster) | 2.78 µs (2.09× faster) | — |
 | Doublets Split NonVolatile | 490 ns (17.9× faster) | 2.8 ns (35.7× faster) | 31.6 ns (67× faster) | 244 ns (9.81× faster) | 177 ns (13.1× faster) | 180 ns (12.9× faster) | 737 ns (31.5× faster) | 3.13 µs (4.39× faster) | 400.0 MiB |
 
-![Rust doublets vs SQLite, 32 bit, links](docs/benchmarks/links-rust-32.png)
+![Rust doublets vs SQLite, 32 bit, links, linear scale](docs/benchmarks/links-rust-32-linear.png)
+
+![Rust doublets vs SQLite, 32 bit, links, log scale](docs/benchmarks/links-rust-32.png)
 
 #### 64 bit address/id space benchmarks
 
 ##### 100,000 links
 
-_10 repetitions after a warm-up, median time per operation. SQLite 3.53.2, Ubuntu 24.04.5 LTS, AMD EPYC 7763 64-Core Processor, 4 cores, 16 GiB, [GitHub Actions run](https://github.com/linksplatform/Comparisons.SQLiteVSDoublets/actions/runs/37381004025) on 2026-10-05._
+_10 repetitions after a warm-up, median time per operation. SQLite 3.53.2; doublets 0.5.0; rustc 1.99.0 (b940084d7 2026-09-28). Ubuntu 24.04.5 LTS, AMD EPYC 7763 64-Core Processor, 4 cores, 16 GiB, [GitHub Actions run](https://github.com/linksplatform/Comparisons.SQLiteVSDoublets/actions/runs/37381004025) on 2026-10-05._
 
 | Storage | Create | Read all | Read by id | Search (from, to) | Read by from | Read by to | Update | Delete | File size |
 | --- | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
@@ -229,7 +241,7 @@ _10 repetitions after a warm-up, median time per operation. SQLite 3.53.2, Ubunt
 
 ##### 1,000,000 links
 
-_3 repetitions after a warm-up, median time per operation. SQLite 3.53.2, Ubuntu 24.04.5 LTS, AMD EPYC 7763 64-Core Processor, 4 cores, 16 GiB, [GitHub Actions run](https://github.com/linksplatform/Comparisons.SQLiteVSDoublets/actions/runs/37381004025) on 2026-10-05._
+_3 repetitions after a warm-up, median time per operation. SQLite 3.53.2; doublets 0.5.0; rustc 1.99.0 (b940084d7 2026-09-28). Ubuntu 24.04.5 LTS, AMD EPYC 7763 64-Core Processor, 4 cores, 16 GiB, [GitHub Actions run](https://github.com/linksplatform/Comparisons.SQLiteVSDoublets/actions/runs/37381004025) on 2026-10-05._
 
 | Storage | Create | Read all | Read by id | Search (from, to) | Read by from | Read by to | Update | Delete | File size |
 | --- | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
@@ -242,7 +254,7 @@ _3 repetitions after a warm-up, median time per operation. SQLite 3.53.2, Ubuntu
 
 ##### 10,000,000 links
 
-_1 repetition after a warm-up, median time per operation. SQLite 3.53.2, Ubuntu 24.04.5 LTS, AMD EPYC 9V74 80-Core Processor, 4 cores, 16 GiB, [GitHub Actions run](https://github.com/linksplatform/Comparisons.SQLiteVSDoublets/actions/runs/37381004025) on 2026-10-05._
+_1 repetition after a warm-up, median time per operation. SQLite 3.53.2; doublets 0.5.0; rustc 1.99.0 (b940084d7 2026-09-28). Ubuntu 24.04.5 LTS, AMD EPYC 9V74 80-Core Processor, 4 cores, 16 GiB, [GitHub Actions run](https://github.com/linksplatform/Comparisons.SQLiteVSDoublets/actions/runs/37381004025) on 2026-10-05._
 
 | Storage | Create | Read all | Read by id | Search (from, to) | Read by from | Read by to | Update | Delete | File size |
 | --- | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
@@ -253,7 +265,9 @@ _1 repetition after a warm-up, median time per operation. SQLite 3.53.2, Ubuntu
 | Doublets Split Volatile | 381 ns (8.71× faster) | 3.88 ns (24.7× faster) | 28 ns (54.4× faster) | 247 ns (5.91× faster) | 174 ns (8.47× faster) | 176 ns (8.31× faster) | 719 ns (12.1× faster) | 2.78 µs (2.19× faster) | — |
 | Doublets Split NonVolatile | 677 ns (13.4× faster) | 4.19 ns (23.7× faster) | 36.7 ns (58.3× faster) | 248 ns (9.71× faster) | 173 ns (13.4× faster) | 179 ns (13.1× faster) | 755 ns (32.6× faster) | 3.64 µs (4.18× faster) | 800.0 MiB |
 
-![Rust doublets vs SQLite, 64 bit, links](docs/benchmarks/links-rust-64.png)
+![Rust doublets vs SQLite, 64 bit, links, linear scale](docs/benchmarks/links-rust-64-linear.png)
+
+![Rust doublets vs SQLite, 64 bit, links, log scale](docs/benchmarks/links-rust-64.png)
 
 ### C# doublets vs SQLite
 
@@ -261,7 +275,7 @@ _1 repetition after a warm-up, median time per operation. SQLite 3.53.2, Ubuntu
 
 ##### 100,000 links
 
-_10 repetitions after a warm-up, median time per operation. SQLite 3.53.3, Ubuntu 24.04.5 LTS, AMD EPYC 7763 64-Core Processor, 4 cores, 16 GiB, [GitHub Actions run](https://github.com/linksplatform/Comparisons.SQLiteVSDoublets/actions/runs/37381004025) on 2026-10-05._
+_10 repetitions after a warm-up, median time per operation. SQLite 3.53.3; Platform.Data.Doublets 0.18.1; .NET SDK 10.0.401. Ubuntu 24.04.5 LTS, AMD EPYC 7763 64-Core Processor, 4 cores, 16 GiB, [GitHub Actions run](https://github.com/linksplatform/Comparisons.SQLiteVSDoublets/actions/runs/37381004025) on 2026-10-05._
 
 | Storage | Create | Read all | Read by id | Search (from, to) | Read by from | Read by to | Update | Delete | File size |
 | --- | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
@@ -276,7 +290,7 @@ _10 repetitions after a warm-up, median time per operation. SQLite 3.53.3, Ubunt
 
 ##### 1,000,000 links
 
-_3 repetitions after a warm-up, median time per operation. SQLite 3.53.3, Ubuntu 24.04.5 LTS, AMD EPYC 7763 64-Core Processor, 4 cores, 16 GiB, [GitHub Actions run](https://github.com/linksplatform/Comparisons.SQLiteVSDoublets/actions/runs/37381004025) on 2026-10-05._
+_3 repetitions after a warm-up, median time per operation. SQLite 3.53.3; Platform.Data.Doublets 0.18.1; .NET SDK 10.0.401. Ubuntu 24.04.5 LTS, AMD EPYC 7763 64-Core Processor, 4 cores, 16 GiB, [GitHub Actions run](https://github.com/linksplatform/Comparisons.SQLiteVSDoublets/actions/runs/37381004025) on 2026-10-05._
 
 | Storage | Create | Read all | Read by id | Search (from, to) | Read by from | Read by to | Update | Delete | File size |
 | --- | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
@@ -291,7 +305,7 @@ _3 repetitions after a warm-up, median time per operation. SQLite 3.53.3, Ubuntu
 
 ##### 10,000,000 links
 
-_1 repetition after a warm-up, median time per operation. SQLite 3.53.3, Ubuntu 24.04.5 LTS, INTEL(R) XEON(R) PLATINUM 8573C, 4 cores, 16 GiB, [GitHub Actions run](https://github.com/linksplatform/Comparisons.SQLiteVSDoublets/actions/runs/37381004025) on 2026-10-05._
+_1 repetition after a warm-up, median time per operation. SQLite 3.53.3; Platform.Data.Doublets 0.18.1; .NET SDK 10.0.401. Ubuntu 24.04.5 LTS, INTEL(R) XEON(R) PLATINUM 8573C, 4 cores, 16 GiB, [GitHub Actions run](https://github.com/linksplatform/Comparisons.SQLiteVSDoublets/actions/runs/37381004025) on 2026-10-05._
 
 | Storage | Create | Read all | Read by id | Search (from, to) | Read by from | Read by to | Update | Delete | File size |
 | --- | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
@@ -304,13 +318,15 @@ _1 repetition after a warm-up, median time per operation. SQLite 3.53.3, Ubuntu
 | Doublets Split Volatile | 486 ns (9.61× faster) | 10 ns (47.1× faster) | 176 ns (12.6× faster) | 431 ns (5.82× faster) | 265 ns (9.74× faster) | 257 ns (10× faster) | 694 ns (16.4× faster) | 563 ns (13.2× faster) | — |
 | Doublets Split NonVolatile | 1.27 µs (6.33× faster) | 10.1 ns (46.4× faster) | 205 ns (12.2× faster) | 507 ns (5.7× faster) | 309 ns (9.38× faster) | 305 ns (9.52× faster) | 979 ns (20.4× faster) | 817 ns (14.5× faster) | 400.0 MiB |
 
-![C# doublets vs SQLite, 32 bit, links](docs/benchmarks/links-csharp-32.png)
+![C# doublets vs SQLite, 32 bit, links, linear scale](docs/benchmarks/links-csharp-32-linear.png)
+
+![C# doublets vs SQLite, 32 bit, links, log scale](docs/benchmarks/links-csharp-32.png)
 
 #### 64 bit address/id space benchmarks
 
 ##### 100,000 links
 
-_10 repetitions after a warm-up, median time per operation. SQLite 3.53.3, Ubuntu 24.04.5 LTS, AMD EPYC 9V74 80-Core Processor, 4 cores, 16 GiB, [GitHub Actions run](https://github.com/linksplatform/Comparisons.SQLiteVSDoublets/actions/runs/37381004025) on 2026-10-05._
+_10 repetitions after a warm-up, median time per operation. SQLite 3.53.3; Platform.Data.Doublets 0.18.1; .NET SDK 10.0.401. Ubuntu 24.04.5 LTS, AMD EPYC 9V74 80-Core Processor, 4 cores, 16 GiB, [GitHub Actions run](https://github.com/linksplatform/Comparisons.SQLiteVSDoublets/actions/runs/37381004025) on 2026-10-05._
 
 | Storage | Create | Read all | Read by id | Search (from, to) | Read by from | Read by to | Update | Delete | File size |
 | --- | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
@@ -325,7 +341,7 @@ _10 repetitions after a warm-up, median time per operation. SQLite 3.53.3, Ubunt
 
 ##### 1,000,000 links
 
-_3 repetitions after a warm-up, median time per operation. SQLite 3.53.3, Ubuntu 24.04.5 LTS, AMD EPYC 9V74 80-Core Processor, 4 cores, 16 GiB, [GitHub Actions run](https://github.com/linksplatform/Comparisons.SQLiteVSDoublets/actions/runs/37381004025) on 2026-10-05._
+_3 repetitions after a warm-up, median time per operation. SQLite 3.53.3; Platform.Data.Doublets 0.18.1; .NET SDK 10.0.401. Ubuntu 24.04.5 LTS, AMD EPYC 9V74 80-Core Processor, 4 cores, 16 GiB, [GitHub Actions run](https://github.com/linksplatform/Comparisons.SQLiteVSDoublets/actions/runs/37381004025) on 2026-10-05._
 
 | Storage | Create | Read all | Read by id | Search (from, to) | Read by from | Read by to | Update | Delete | File size |
 | --- | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
@@ -340,7 +356,7 @@ _3 repetitions after a warm-up, median time per operation. SQLite 3.53.3, Ubuntu
 
 ##### 10,000,000 links
 
-_1 repetition after a warm-up, median time per operation. SQLite 3.53.3, Ubuntu 24.04.5 LTS, AMD EPYC 7763 64-Core Processor, 4 cores, 16 GiB, [GitHub Actions run](https://github.com/linksplatform/Comparisons.SQLiteVSDoublets/actions/runs/37381004025) on 2026-10-05._
+_1 repetition after a warm-up, median time per operation. SQLite 3.53.3; Platform.Data.Doublets 0.18.1; .NET SDK 10.0.401. Ubuntu 24.04.5 LTS, AMD EPYC 7763 64-Core Processor, 4 cores, 16 GiB, [GitHub Actions run](https://github.com/linksplatform/Comparisons.SQLiteVSDoublets/actions/runs/37381004025) on 2026-10-05._
 
 | Storage | Create | Read all | Read by id | Search (from, to) | Read by from | Read by to | Update | Delete | File size |
 | --- | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
@@ -353,7 +369,9 @@ _1 repetition after a warm-up, median time per operation. SQLite 3.53.3, Ubuntu
 | Doublets Split Volatile | 449 ns (10× faster) | 12.6 ns (36.6× faster) | 294 ns (7.12× faster) | 443 ns (5.32× faster) | 317 ns (7.65× faster) | 330 ns (7.26× faster) | 859 ns (11.3× faster) | 688 ns (8.95× faster) | — |
 | Doublets Split NonVolatile | 1.27 µs (8.22× faster) | 12.6 ns (36.8× faster) | 331 ns (8.88× faster) | 484 ns (7.17× faster) | 339 ns (9.86× faster) | 351 ns (9.55× faster) | 1.19 µs (21.4× faster) | 853 ns (18.1× faster) | 800.0 MiB |
 
-![C# doublets vs SQLite, 64 bit, links](docs/benchmarks/links-csharp-64.png)
+![C# doublets vs SQLite, 64 bit, links, linear scale](docs/benchmarks/links-csharp-64-linear.png)
+
+![C# doublets vs SQLite, 64 bit, links, log scale](docs/benchmarks/links-csharp-64.png)
 
 ## Doublets vs SQLite as storage for objects
 
@@ -363,7 +381,7 @@ _1 repetition after a warm-up, median time per operation. SQLite 3.53.3, Ubuntu
 
 ##### 100,000 blog posts
 
-_5 repetitions after a warm-up, median time per operation. SQLite 3.53.2, Ubuntu 24.04.5 LTS, AMD EPYC 9V45 96-Core Processor, 4 cores, 16 GiB, [GitHub Actions run](https://github.com/linksplatform/Comparisons.SQLiteVSDoublets/actions/runs/37381004025) on 2026-10-05._
+_5 repetitions after a warm-up, median time per operation. SQLite 3.53.2; doublets 0.5.0; rustc 1.99.0 (b940084d7 2026-09-28). Ubuntu 24.04.5 LTS, AMD EPYC 9V45 96-Core Processor, 4 cores, 16 GiB, [GitHub Actions run](https://github.com/linksplatform/Comparisons.SQLiteVSDoublets/actions/runs/37381004025) on 2026-10-05._
 
 | Storage | Create | Read all | Read by id | Delete | File size |
 | --- | ---: | ---: | ---: | ---: | ---: |
@@ -380,7 +398,7 @@ _5 repetitions after a warm-up, median time per operation. SQLite 3.53.2, Ubuntu
 
 ##### 1,000,000 blog posts
 
-_1 repetition after a warm-up, median time per operation. SQLite 3.53.2, Ubuntu 24.04.5 LTS, AMD EPYC 9V74 80-Core Processor, 4 cores, 16 GiB, [GitHub Actions run](https://github.com/linksplatform/Comparisons.SQLiteVSDoublets/actions/runs/37381004025) on 2026-10-05._
+_1 repetition after a warm-up, median time per operation. SQLite 3.53.2; doublets 0.5.0; rustc 1.99.0 (b940084d7 2026-09-28). Ubuntu 24.04.5 LTS, AMD EPYC 9V74 80-Core Processor, 4 cores, 16 GiB, [GitHub Actions run](https://github.com/linksplatform/Comparisons.SQLiteVSDoublets/actions/runs/37381004025) on 2026-10-05._
 
 | Storage | Create | Read all | Read by id | Delete | File size |
 | --- | ---: | ---: | ---: | ---: | ---: |
@@ -395,13 +413,15 @@ _1 repetition after a warm-up, median time per operation. SQLite 3.53.2, Ubuntu
 | Doublets Split NonVolatile Cached | 6.46 µs (1.89× slower) | 882 ns (1.34× slower) | 961 ns (4.23× faster) | 6.75 µs (1.68× faster) | 400.0 MiB |
 | Doublets Split NonVolatile Uncached | 36.6 µs (10.7× slower) | 14.5 µs (22.1× slower) | 14.8 µs (3.64× slower) | 6.98 µs (1.63× faster) | 400.0 MiB |
 
-![Rust doublets vs SQLite, 32 bit, objects](docs/benchmarks/objects-rust-32.png)
+![Rust doublets vs SQLite, 32 bit, objects, linear scale](docs/benchmarks/objects-rust-32-linear.png)
+
+![Rust doublets vs SQLite, 32 bit, objects, log scale](docs/benchmarks/objects-rust-32.png)
 
 #### 64 bit address/id space benchmarks
 
 ##### 100,000 blog posts
 
-_5 repetitions after a warm-up, median time per operation. SQLite 3.53.2, Ubuntu 24.04.5 LTS, AMD EPYC 9V45 96-Core Processor, 4 cores, 16 GiB, [GitHub Actions run](https://github.com/linksplatform/Comparisons.SQLiteVSDoublets/actions/runs/37381004025) on 2026-10-05._
+_5 repetitions after a warm-up, median time per operation. SQLite 3.53.2; doublets 0.5.0; rustc 1.99.0 (b940084d7 2026-09-28). Ubuntu 24.04.5 LTS, AMD EPYC 9V45 96-Core Processor, 4 cores, 16 GiB, [GitHub Actions run](https://github.com/linksplatform/Comparisons.SQLiteVSDoublets/actions/runs/37381004025) on 2026-10-05._
 
 | Storage | Create | Read all | Read by id | Delete | File size |
 | --- | ---: | ---: | ---: | ---: | ---: |
@@ -418,7 +438,7 @@ _5 repetitions after a warm-up, median time per operation. SQLite 3.53.2, Ubuntu
 
 ##### 1,000,000 blog posts
 
-_1 repetition after a warm-up, median time per operation. SQLite 3.53.2, Ubuntu 24.04.5 LTS, AMD EPYC 7763 64-Core Processor, 4 cores, 16 GiB, [GitHub Actions run](https://github.com/linksplatform/Comparisons.SQLiteVSDoublets/actions/runs/37381004025) on 2026-10-05._
+_1 repetition after a warm-up, median time per operation. SQLite 3.53.2; doublets 0.5.0; rustc 1.99.0 (b940084d7 2026-09-28). Ubuntu 24.04.5 LTS, AMD EPYC 7763 64-Core Processor, 4 cores, 16 GiB, [GitHub Actions run](https://github.com/linksplatform/Comparisons.SQLiteVSDoublets/actions/runs/37381004025) on 2026-10-05._
 
 | Storage | Create | Read all | Read by id | Delete | File size |
 | --- | ---: | ---: | ---: | ---: | ---: |
@@ -433,7 +453,9 @@ _1 repetition after a warm-up, median time per operation. SQLite 3.53.2, Ubuntu
 | Doublets Split NonVolatile Cached | 8.1 µs (2.3× slower) | 952 ns (1.5× slower) | 1.02 µs (3.45× faster) | 7.65 µs (1.32× faster) | 800.0 MiB |
 | Doublets Split NonVolatile Uncached | 55.9 µs (15.9× slower) | 14.7 µs (23.1× slower) | 15.4 µs (4.35× slower) | 8.04 µs (1.26× faster) | 800.0 MiB |
 
-![Rust doublets vs SQLite, 64 bit, objects](docs/benchmarks/objects-rust-64.png)
+![Rust doublets vs SQLite, 64 bit, objects, linear scale](docs/benchmarks/objects-rust-64-linear.png)
+
+![Rust doublets vs SQLite, 64 bit, objects, log scale](docs/benchmarks/objects-rust-64.png)
 
 ### C# doublets vs SQLite
 
@@ -441,7 +463,7 @@ _1 repetition after a warm-up, median time per operation. SQLite 3.53.2, Ubuntu
 
 ##### 100,000 blog posts
 
-_5 repetitions after a warm-up, median time per operation. SQLite 3.53.3, Ubuntu 24.04.5 LTS, AMD EPYC 7763 64-Core Processor, 4 cores, 16 GiB, [GitHub Actions run](https://github.com/linksplatform/Comparisons.SQLiteVSDoublets/actions/runs/37381004025) on 2026-10-05._ PostgreSQL 18.3 (Debian 18.3-1.pgdg13+1); Npgsql.EntityFrameworkCore.PostgreSQL 10.0.3.0; EF Core 10.0.4.0.
+_5 repetitions after a warm-up, median time per operation. SQLite 3.53.3; Platform.Data.Doublets 0.18.1; .NET SDK 10.0.401. Ubuntu 24.04.5 LTS, AMD EPYC 7763 64-Core Processor, 4 cores, 16 GiB, [GitHub Actions run](https://github.com/linksplatform/Comparisons.SQLiteVSDoublets/actions/runs/37381004025) on 2026-10-05. PostgreSQL 18.3 (Debian 18.3-1.pgdg13+1); Npgsql.EntityFrameworkCore.PostgreSQL 10.0.3.0; EF Core 10.0.4.0._
 
 | Storage | Create | Read all | Read by id | Delete | File size | Server relations |
 | --- | ---: | ---: | ---: | ---: | ---: | ---: |
@@ -461,7 +483,7 @@ _5 repetitions after a warm-up, median time per operation. SQLite 3.53.3, Ubuntu
 
 ##### 1,000,000 blog posts
 
-_1 repetition after a warm-up, median time per operation. SQLite 3.53.3, Ubuntu 24.04.5 LTS, INTEL(R) XEON(R) PLATINUM 8573C, 4 cores, 16 GiB, [GitHub Actions run](https://github.com/linksplatform/Comparisons.SQLiteVSDoublets/actions/runs/37381004025) on 2026-10-05._ PostgreSQL 18.3 (Debian 18.3-1.pgdg13+1); Npgsql.EntityFrameworkCore.PostgreSQL 10.0.3.0; EF Core 10.0.4.0.
+_1 repetition after a warm-up, median time per operation. SQLite 3.53.3; Platform.Data.Doublets 0.18.1; .NET SDK 10.0.401. Ubuntu 24.04.5 LTS, INTEL(R) XEON(R) PLATINUM 8573C, 4 cores, 16 GiB, [GitHub Actions run](https://github.com/linksplatform/Comparisons.SQLiteVSDoublets/actions/runs/37381004025) on 2026-10-05. PostgreSQL 18.3 (Debian 18.3-1.pgdg13+1); Npgsql.EntityFrameworkCore.PostgreSQL 10.0.3.0; EF Core 10.0.4.0._
 
 | Storage | Create | Read all | Read by id | Delete | File size | Server relations |
 | --- | ---: | ---: | ---: | ---: | ---: | ---: |
@@ -479,13 +501,15 @@ _1 repetition after a warm-up, median time per operation. SQLite 3.53.3, Ubuntu
 | Doublets Split NonVolatile Cached | 21 µs (2.32× slower) | 14 µs (3.79× slower) | 7.1 µs (≈ same) | 10.8 µs (≈ same) | 400.0 MiB | — |
 | Doublets Split NonVolatile Uncached | 142 µs (15.7× slower) | 460 µs (125× slower) | 460 µs (67.5× slower) | 11.2 µs (≈ same) | 400.0 MiB | — |
 
-![C# doublets vs SQLite, 32 bit, objects](docs/benchmarks/objects-csharp-32.png)
+![C# doublets vs SQLite, 32 bit, objects, linear scale](docs/benchmarks/objects-csharp-32-linear.png)
+
+![C# doublets vs SQLite, 32 bit, objects, log scale](docs/benchmarks/objects-csharp-32.png)
 
 #### 64 bit address/id space benchmarks
 
 ##### 100,000 blog posts
 
-_5 repetitions after a warm-up, median time per operation. SQLite 3.53.3, Ubuntu 24.04.5 LTS, AMD EPYC 9V74 80-Core Processor, 4 cores, 16 GiB, [GitHub Actions run](https://github.com/linksplatform/Comparisons.SQLiteVSDoublets/actions/runs/37381004025) on 2026-10-05._ PostgreSQL 18.3 (Debian 18.3-1.pgdg13+1); Npgsql.EntityFrameworkCore.PostgreSQL 10.0.3.0; EF Core 10.0.4.0.
+_5 repetitions after a warm-up, median time per operation. SQLite 3.53.3; Platform.Data.Doublets 0.18.1; .NET SDK 10.0.401. Ubuntu 24.04.5 LTS, AMD EPYC 9V74 80-Core Processor, 4 cores, 16 GiB, [GitHub Actions run](https://github.com/linksplatform/Comparisons.SQLiteVSDoublets/actions/runs/37381004025) on 2026-10-05. PostgreSQL 18.3 (Debian 18.3-1.pgdg13+1); Npgsql.EntityFrameworkCore.PostgreSQL 10.0.3.0; EF Core 10.0.4.0._
 
 | Storage | Create | Read all | Read by id | Delete | File size | Server relations |
 | --- | ---: | ---: | ---: | ---: | ---: | ---: |
@@ -505,7 +529,7 @@ _5 repetitions after a warm-up, median time per operation. SQLite 3.53.3, Ubuntu
 
 ##### 1,000,000 blog posts
 
-_1 repetition after a warm-up, median time per operation. SQLite 3.53.3, Ubuntu 24.04.5 LTS, AMD EPYC 9V74 80-Core Processor, 4 cores, 16 GiB, [GitHub Actions run](https://github.com/linksplatform/Comparisons.SQLiteVSDoublets/actions/runs/37381004025) on 2026-10-05._ PostgreSQL 18.3 (Debian 18.3-1.pgdg13+1); Npgsql.EntityFrameworkCore.PostgreSQL 10.0.3.0; EF Core 10.0.4.0.
+_1 repetition after a warm-up, median time per operation. SQLite 3.53.3; Platform.Data.Doublets 0.18.1; .NET SDK 10.0.401. Ubuntu 24.04.5 LTS, AMD EPYC 9V74 80-Core Processor, 4 cores, 16 GiB, [GitHub Actions run](https://github.com/linksplatform/Comparisons.SQLiteVSDoublets/actions/runs/37381004025) on 2026-10-05. PostgreSQL 18.3 (Debian 18.3-1.pgdg13+1); Npgsql.EntityFrameworkCore.PostgreSQL 10.0.3.0; EF Core 10.0.4.0._
 
 | Storage | Create | Read all | Read by id | Delete | File size | Server relations |
 | --- | ---: | ---: | ---: | ---: | ---: | ---: |
@@ -523,123 +547,17 @@ _1 repetition after a warm-up, median time per operation. SQLite 3.53.3, Ubuntu
 | Doublets Split NonVolatile Cached | 27.6 µs (3.25× slower) | 11.3 µs (3.14× slower) | 7.63 µs (1.1× slower) | 13 µs (1.3× faster) | 800.0 MiB | — |
 | Doublets Split NonVolatile Uncached | 121 µs (14.2× slower) | 323 µs (90× slower) | 320 µs (46.2× slower) | 12.5 µs (1.35× faster) | 800.0 MiB | — |
 
-![C# doublets vs SQLite, 64 bit, objects](docs/benchmarks/objects-csharp-64.png)
-<!-- markdownlint-restore -->
-<!--BENCHMARK_RESULTS_END-->
-
-## Original comparison
-
-The original C# object comparison and its historical results.
-
-<!-- markdownlint-disable MD013 MD060 -->
-
-### SQLite
-
-```C#
-using System.Linq;
-using Comparisons.SQLiteVSDoublets.Model;
-
-namespace Comparisons.SQLiteVSDoublets.SQLite
-{
-    public class SQLiteTestRun : TestRun
-    {
-        public SQLiteTestRun(string dbFilename) : base(dbFilename) { }
-
-        public override void Prepare()
-        {
-            using var dbContext = new SQLiteDbContext(DbFilename);
-            dbContext.Database.EnsureCreated();
-        }
-
-        public override void CreateList()
-        {
-            using var dbContext = new SQLiteDbContext(DbFilename);
-            dbContext.BlogPosts.AddRange(BlogPosts.List);
-            dbContext.SaveChanges();
-        }
-
-        public override void ReadList()
-        {
-            using var dbContext = new SQLiteDbContext(DbFilename);
-            foreach (var blogPost in dbContext.BlogPosts)
-            {
-                ReadBlogPosts.Add(blogPost);
-            }
-        }
-
-        public override void DeleteList()
-        {
-            using var dbContext = new SQLiteDbContext(DbFilename);
-            var blogPostsToDelete = dbContext.BlogPosts.ToList();
-            dbContext.BlogPosts.RemoveRange(blogPostsToDelete);
-            dbContext.SaveChanges();
-        }
-    }
-}
-```
+![C# doublets vs SQLite, 64 bit, objects, linear scale](docs/benchmarks/objects-csharp-64-linear.png)
 
-### Doublets
-
-``` C#
-using System.IO;
-using Platform.IO;
-using Comparisons.SQLiteVSDoublets.Model;
-
-namespace Comparisons.SQLiteVSDoublets.Doublets
-{
-    public class DoubletsTestRun : TestRun
-    {
-        public string DbIndexFilename { get; }
-
-        public DoubletsTestRun(string dbFilename) : base(dbFilename) => DbIndexFilename = $"{Path.GetFileNameWithoutExtension(dbFilename)}.links.index";
-
-        public override void Prepare()
-        {
-            using var dbContext = new DoubletsDbContext(DbFilename, DbIndexFilename);
-        }
-
-        public override void CreateList()
-        {
-            using var dbContext = new DoubletsDbContext(DbFilename, DbIndexFilename);
-            foreach (var blogPost in BlogPosts.List)
-            {
-                dbContext.SaveBlogPost(blogPost);
-            }
-        }
-
-        public override void ReadList()
-        {
-            using var dbContext = new DoubletsDbContext(DbFilename, DbIndexFilename);
-            foreach (var blogPost in dbContext.BlogPosts)
-            {
-                ReadBlogPosts.Add(blogPost);
-            }
-        }
-
-        public override void DeleteList()
-        {
-            using var dbContext = new DoubletsDbContext(DbFilename, DbIndexFilename);
-            var blogPostsToDelete = dbContext.BlogPosts;
-            foreach (var blogPost in blogPostsToDelete)
-            {
-                dbContext.Delete((uint)blogPost.Id);
-            }
-        }
-
-        protected override void DeleteDatabase()
-        {
-            File.Delete(DbFilename);
-            File.Delete(DbIndexFilename);
-        }
-
-        protected override long GetDatabaseSizeInBytes() => FileHelpers.GetSize(DbFilename) + FileHelpers.GetSize(DbIndexFilename);
-    }
-}
-```
+![C# doublets vs SQLite, 64 bit, objects, log scale](docs/benchmarks/objects-csharp-64.png)
 
-<!-- markdownlint-restore -->
+## Conclusions
 
-### Conclusion
+Each comparison uses SQLite of matching durability and the same noise rule as the tables. Counts span the measured variants, sizes and id widths; they are not an overall speed ranking. RAM usage is not measured.
 
-In this particular comparison, Doublets are faster and use less memory on disk,
-but this comes with the cost of additional use of RAM (Sqlite uses it less).
+- Rust, links: 187 faster, 3 slower, 2 approximately equal Doublets operation comparisons with SQLite.
+- C#, links: 192 faster, 0 slower, 0 approximately equal Doublets operation comparisons with SQLite.
+- Rust, objects: 28 faster, 98 slower, 2 approximately equal Doublets operation comparisons with SQLite.
+- C#, objects: 10 faster, 109 slower, 9 approximately equal Doublets operation comparisons with SQLite.
+<!-- markdownlint-restore -->
+<!--BENCHMARK_RESULTS_END-->
diff --git a/README.ru.md b/README.ru.md
index 7518b21..b6d94fa 100644
--- a/README.ru.md
+++ b/README.ru.md
@@ -107,6 +107,12 @@ docker rm benchmark-postgres
 выделенные отображаемые в память файлы, поэтому для небольших хранилищ показан
 размер предварительного выделения.
 
+Для каждой группы результатов строятся линейный и логарифмический графики.
+На линейном графике столбцы ниже 0,5% максимума панели увеличены для видимости;
+точное время остаётся в таблицах. Происхождение результатов включает версии
+библиотеки Дуплетов и компилятора Rust или .NET SDK, использованных при запуске.
+Если старый отчёт не содержит этих данных, они отмечены как отсутствующие.
+
 Каждая таблица измеряется отдельной задачей GitHub Actions со всеми вариантами
 на одной машине, чтобы разница никогда не была разницей между машинами,
 в [workflow Benchmarks](.github/workflows/benchmarks.yml): связи на 100 000,
@@ -130,6 +136,10 @@ dotnet run -c Release --project csharp/SQLiteVSDoublets -- \
   links 64 1000 --repetitions 3 \
   --variants SQLite_Memory,SystemDataSQLite_Memory \
   --output results/links-csharp-64-1000.json
+python3 scripts/benchmark_provenance.py results/links-rust-64-100000.json \
+  --toolchain "$(rustc --version)"
+python3 scripts/benchmark_provenance.py results/links-csharp-64-1000.json \
+  --toolchain ".NET SDK $(dotnet --version)"
 # печатает таблицы, для обновления этого файла добавьте
 # --readme README.ru.md --charts docs/benchmarks
 python3 scripts/benchmark_report.py results
@@ -175,7 +185,7 @@ python3 scripts/benchmark_report.py results
 
 ##### 100,000 связей
 
-_10 повторов после прогрева, медианное время одной операции. SQLite 3.53.2, Ubuntu 24.04.5 LTS, AMD EPYC 7763 64-Core Processor, 4 cores, 16 GiB, [запуск GitHub Actions](https://github.com/linksplatform/Comparisons.SQLiteVSDoublets/actions/runs/37381004025) от 2026-10-05._
+_10 повторов после прогрева, медианное время одной операции. SQLite 3.53.2; doublets 0.5.0; rustc 1.99.0 (b940084d7 2026-09-28). Ubuntu 24.04.5 LTS, AMD EPYC 7763 64-Core Processor, 4 cores, 16 GiB, [запуск GitHub Actions](https://github.com/linksplatform/Comparisons.SQLiteVSDoublets/actions/runs/37381004025) от 2026-10-05._
 
 | Хранилище | Создание | Чтение всех | Чтение по id | Поиск (from, to) | Чтение по from | Чтение по to | Обновление | Удаление | Размер файлов |
 | --- | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
@@ -188,7 +198,7 @@ _10 повторов после прогрева, медианное время
 
 ##### 1,000,000 связей
 
-_3 повтора после прогрева, медианное время одной операции. SQLite 3.53.2, Ubuntu 24.04.5 LTS, AMD EPYC 7763 64-Core Processor, 4 cores, 16 GiB, [запуск GitHub Actions](https://github.com/linksplatform/Comparisons.SQLiteVSDoublets/actions/runs/37381004025) от 2026-10-05._
+_3 повтора после прогрева, медианное время одной операции. SQLite 3.53.2; doublets 0.5.0; rustc 1.99.0 (b940084d7 2026-09-28). Ubuntu 24.04.5 LTS, AMD EPYC 7763 64-Core Processor, 4 cores, 16 GiB, [запуск GitHub Actions](https://github.com/linksplatform/Comparisons.SQLiteVSDoublets/actions/runs/37381004025) от 2026-10-05._
 
 | Хранилище | Создание | Чтение всех | Чтение по id | Поиск (from, to) | Чтение по from | Чтение по to | Обновление | Удаление | Размер файлов |
 | --- | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
@@ -201,7 +211,7 @@ _3 повтора после прогрева, медианное время о
 
 ##### 10,000,000 связей
 
-_1 повтор после прогрева, медианное время одной операции. SQLite 3.53.2, Ubuntu 24.04.5 LTS, AMD EPYC 9V74 80-Core Processor, 4 cores, 16 GiB, [запуск GitHub Actions](https://github.com/linksplatform/Comparisons.SQLiteVSDoublets/actions/runs/37381004025) от 2026-10-05._
+_1 повтор после прогрева, медианное время одной операции. SQLite 3.53.2; doublets 0.5.0; rustc 1.99.0 (b940084d7 2026-09-28). Ubuntu 24.04.5 LTS, AMD EPYC 9V74 80-Core Processor, 4 cores, 16 GiB, [запуск GitHub Actions](https://github.com/linksplatform/Comparisons.SQLiteVSDoublets/actions/runs/37381004025) от 2026-10-05._
 
 | Хранилище | Создание | Чтение всех | Чтение по id | Поиск (from, to) | Чтение по from | Чтение по to | Обновление | Удаление | Размер файлов |
 | --- | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
@@ -212,13 +222,15 @@ _1 повтор после прогрева, медианное время од
 | Doublets Split Volatile | 337 ns (в 9.32× быстрее) | 2.83 ns (в 34.7× быстрее) | 30.9 ns (в 44.9× быстрее) | 238 ns (в 5.79× быстрее) | 171 ns (в 8.23× быстрее) | 170 ns (в 8.18× быстрее) | 689 ns (в 12.3× быстрее) | 2.78 µs (в 2.09× быстрее) | — |
 | Doublets Split NonVolatile | 490 ns (в 17.9× быстрее) | 2.8 ns (в 35.7× быстрее) | 31.6 ns (в 67× быстрее) | 244 ns (в 9.81× быстрее) | 177 ns (в 13.1× быстрее) | 180 ns (в 12.9× быстрее) | 737 ns (в 31.5× быстрее) | 3.13 µs (в 4.39× быстрее) | 400.0 MiB |
 
-![Дуплеты на Rust против SQLite, 32 бит, связи](docs/benchmarks/links-rust-32.png)
+![Дуплеты на Rust против SQLite, 32 бит, связи, линейная шкала](docs/benchmarks/links-rust-32-linear.png)
+
+![Дуплеты на Rust против SQLite, 32 бит, связи, логарифмическая шкала](docs/benchmarks/links-rust-32.png)
 
 #### Тесты с 64-битным пространством адресов/идентификаторов
 
 ##### 100,000 связей
 
-_10 повторов после прогрева, медианное время одной операции. SQLite 3.53.2, Ubuntu 24.04.5 LTS, AMD EPYC 7763 64-Core Processor, 4 cores, 16 GiB, [запуск GitHub Actions](https://github.com/linksplatform/Comparisons.SQLiteVSDoublets/actions/runs/37381004025) от 2026-10-05._
+_10 повторов после прогрева, медианное время одной операции. SQLite 3.53.2; doublets 0.5.0; rustc 1.99.0 (b940084d7 2026-09-28). Ubuntu 24.04.5 LTS, AMD EPYC 7763 64-Core Processor, 4 cores, 16 GiB, [запуск GitHub Actions](https://github.com/linksplatform/Comparisons.SQLiteVSDoublets/actions/runs/37381004025) от 2026-10-05._
 
 | Хранилище | Создание | Чтение всех | Чтение по id | Поиск (from, to) | Чтение по from | Чтение по to | Обновление | Удаление | Размер файлов |
 | --- | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
@@ -231,7 +243,7 @@ _10 повторов после прогрева, медианное время
 
 ##### 1,000,000 связей
 
-_3 повтора после прогрева, медианное время одной операции. SQLite 3.53.2, Ubuntu 24.04.5 LTS, AMD EPYC 7763 64-Core Processor, 4 cores, 16 GiB, [запуск GitHub Actions](https://github.com/linksplatform/Comparisons.SQLiteVSDoublets/actions/runs/37381004025) от 2026-10-05._
+_3 повтора после прогрева, медианное время одной операции. SQLite 3.53.2; doublets 0.5.0; rustc 1.99.0 (b940084d7 2026-09-28). Ubuntu 24.04.5 LTS, AMD EPYC 7763 64-Core Processor, 4 cores, 16 GiB, [запуск GitHub Actions](https://github.com/linksplatform/Comparisons.SQLiteVSDoublets/actions/runs/37381004025) от 2026-10-05._
 
 | Хранилище | Создание | Чтение всех | Чтение по id | Поиск (from, to) | Чтение по from | Чтение по to | Обновление | Удаление | Размер файлов |
 | --- | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
@@ -244,7 +256,7 @@ _3 повтора после прогрева, медианное время о
 
 ##### 10,000,000 связей
 
-_1 повтор после прогрева, медианное время одной операции. SQLite 3.53.2, Ubuntu 24.04.5 LTS, AMD EPYC 9V74 80-Core Processor, 4 cores, 16 GiB, [запуск GitHub Actions](https://github.com/linksplatform/Comparisons.SQLiteVSDoublets/actions/runs/37381004025) от 2026-10-05._
+_1 повтор после прогрева, медианное время одной операции. SQLite 3.53.2; doublets 0.5.0; rustc 1.99.0 (b940084d7 2026-09-28). Ubuntu 24.04.5 LTS, AMD EPYC 9V74 80-Core Processor, 4 cores, 16 GiB, [запуск GitHub Actions](https://github.com/linksplatform/Comparisons.SQLiteVSDoublets/actions/runs/37381004025) от 2026-10-05._
 
 | Хранилище | Создание | Чтение всех | Чтение по id | Поиск (from, to) | Чтение по from | Чтение по to | Обновление | Удаление | Размер файлов |
 | --- | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
@@ -255,7 +267,9 @@ _1 повтор после прогрева, медианное время од
 | Doublets Split Volatile | 381 ns (в 8.71× быстрее) | 3.88 ns (в 24.7× быстрее) | 28 ns (в 54.4× быстрее) | 247 ns (в 5.91× быстрее) | 174 ns (в 8.47× быстрее) | 176 ns (в 8.31× быстрее) | 719 ns (в 12.1× быстрее) | 2.78 µs (в 2.19× быстрее) | — |
 | Doublets Split NonVolatile | 677 ns (в 13.4× быстрее) | 4.19 ns (в 23.7× быстрее) | 36.7 ns (в 58.3× быстрее) | 248 ns (в 9.71× быстрее) | 173 ns (в 13.4× быстрее) | 179 ns (в 13.1× быстрее) | 755 ns (в 32.6× быстрее) | 3.64 µs (в 4.18× быстрее) | 800.0 MiB |
 
-![Дуплеты на Rust против SQLite, 64 бит, связи](docs/benchmarks/links-rust-64.png)
+![Дуплеты на Rust против SQLite, 64 бит, связи, линейная шкала](docs/benchmarks/links-rust-64-linear.png)
+
+![Дуплеты на Rust против SQLite, 64 бит, связи, логарифмическая шкала](docs/benchmarks/links-rust-64.png)
 
 ### Дуплеты на C# против SQLite
 
@@ -263,7 +277,7 @@ _1 повтор после прогрева, медианное время од
 
 ##### 100,000 связей
 
-_10 повторов после прогрева, медианное время одной операции. SQLite 3.53.3, Ubuntu 24.04.5 LTS, AMD EPYC 7763 64-Core Processor, 4 cores, 16 GiB, [запуск GitHub Actions](https://github.com/linksplatform/Comparisons.SQLiteVSDoublets/actions/runs/37381004025) от 2026-10-05._
+_10 повторов после прогрева, медианное время одной операции. SQLite 3.53.3; Platform.Data.Doublets 0.18.1; .NET SDK 10.0.401. Ubuntu 24.04.5 LTS, AMD EPYC 7763 64-Core Processor, 4 cores, 16 GiB, [запуск GitHub Actions](https://github.com/linksplatform/Comparisons.SQLiteVSDoublets/actions/runs/37381004025) от 2026-10-05._
 
 | Хранилище | Создание | Чтение всех | Чтение по id | Поиск (from, to) | Чтение по from | Чтение по to | Обновление | Удаление | Размер файлов |
 | --- | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
@@ -278,7 +292,7 @@ _10 повторов после прогрева, медианное время
 
 ##### 1,000,000 связей
 
-_3 повтора после прогрева, медианное время одной операции. SQLite 3.53.3, Ubuntu 24.04.5 LTS, AMD EPYC 7763 64-Core Processor, 4 cores, 16 GiB, [запуск GitHub Actions](https://github.com/linksplatform/Comparisons.SQLiteVSDoublets/actions/runs/37381004025) от 2026-10-05._
+_3 повтора после прогрева, медианное время одной операции. SQLite 3.53.3; Platform.Data.Doublets 0.18.1; .NET SDK 10.0.401. Ubuntu 24.04.5 LTS, AMD EPYC 7763 64-Core Processor, 4 cores, 16 GiB, [запуск GitHub Actions](https://github.com/linksplatform/Comparisons.SQLiteVSDoublets/actions/runs/37381004025) от 2026-10-05._
 
 | Хранилище | Создание | Чтение всех | Чтение по id | Поиск (from, to) | Чтение по from | Чтение по to | Обновление | Удаление | Размер файлов |
 | --- | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
@@ -293,7 +307,7 @@ _3 повтора после прогрева, медианное время о
 
 ##### 10,000,000 связей
 
-_1 повтор после прогрева, медианное время одной операции. SQLite 3.53.3, Ubuntu 24.04.5 LTS, INTEL(R) XEON(R) PLATINUM 8573C, 4 cores, 16 GiB, [запуск GitHub Actions](https://github.com/linksplatform/Comparisons.SQLiteVSDoublets/actions/runs/37381004025) от 2026-10-05._
+_1 повтор после прогрева, медианное время одной операции. SQLite 3.53.3; Platform.Data.Doublets 0.18.1; .NET SDK 10.0.401. Ubuntu 24.04.5 LTS, INTEL(R) XEON(R) PLATINUM 8573C, 4 cores, 16 GiB, [запуск GitHub Actions](https://github.com/linksplatform/Comparisons.SQLiteVSDoublets/actions/runs/37381004025) от 2026-10-05._
 
 | Хранилище | Создание | Чтение всех | Чтение по id | Поиск (from, to) | Чтение по from | Чтение по to | Обновление | Удаление | Размер файлов |
 | --- | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
@@ -306,13 +320,15 @@ _1 повтор после прогрева, медианное время од
 | Doublets Split Volatile | 486 ns (в 9.61× быстрее) | 10 ns (в 47.1× быстрее) | 176 ns (в 12.6× быстрее) | 431 ns (в 5.82× быстрее) | 265 ns (в 9.74× быстрее) | 257 ns (в 10× быстрее) | 694 ns (в 16.4× быстрее) | 563 ns (в 13.2× быстрее) | — |
 | Doublets Split NonVolatile | 1.27 µs (в 6.33× быстрее) | 10.1 ns (в 46.4× быстрее) | 205 ns (в 12.2× быстрее) | 507 ns (в 5.7× быстрее) | 309 ns (в 9.38× быстрее) | 305 ns (в 9.52× быстрее) | 979 ns (в 20.4× быстрее) | 817 ns (в 14.5× быстрее) | 400.0 MiB |
 
-![Дуплеты на C# против SQLite, 32 бит, связи](docs/benchmarks/links-csharp-32.png)
+![Дуплеты на C# против SQLite, 32 бит, связи, линейная шкала](docs/benchmarks/links-csharp-32-linear.png)
+
+![Дуплеты на C# против SQLite, 32 бит, связи, логарифмическая шкала](docs/benchmarks/links-csharp-32.png)
 
 #### Тесты с 64-битным пространством адресов/идентификаторов
 
 ##### 100,000 связей
 
-_10 повторов после прогрева, медианное время одной операции. SQLite 3.53.3, Ubuntu 24.04.5 LTS, AMD EPYC 9V74 80-Core Processor, 4 cores, 16 GiB, [запуск GitHub Actions](https://github.com/linksplatform/Comparisons.SQLiteVSDoublets/actions/runs/37381004025) от 2026-10-05._
+_10 повторов после прогрева, медианное время одной операции. SQLite 3.53.3; Platform.Data.Doublets 0.18.1; .NET SDK 10.0.401. Ubuntu 24.04.5 LTS, AMD EPYC 9V74 80-Core Processor, 4 cores, 16 GiB, [запуск GitHub Actions](https://github.com/linksplatform/Comparisons.SQLiteVSDoublets/actions/runs/37381004025) от 2026-10-05._
 
 | Хранилище | Создание | Чтение всех | Чтение по id | Поиск (from, to) | Чтение по from | Чтение по to | Обновление | Удаление | Размер файлов |
 | --- | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
@@ -327,7 +343,7 @@ _10 повторов после прогрева, медианное время
 
 ##### 1,000,000 связей
 
-_3 повтора после прогрева, медианное время одной операции. SQLite 3.53.3, Ubuntu 24.04.5 LTS, AMD EPYC 9V74 80-Core Processor, 4 cores, 16 GiB, [запуск GitHub Actions](https://github.com/linksplatform/Comparisons.SQLiteVSDoublets/actions/runs/37381004025) от 2026-10-05._
+_3 повтора после прогрева, медианное время одной операции. SQLite 3.53.3; Platform.Data.Doublets 0.18.1; .NET SDK 10.0.401. Ubuntu 24.04.5 LTS, AMD EPYC 9V74 80-Core Processor, 4 cores, 16 GiB, [запуск GitHub Actions](https://github.com/linksplatform/Comparisons.SQLiteVSDoublets/actions/runs/37381004025) от 2026-10-05._
 
 | Хранилище | Создание | Чтение всех | Чтение по id | Поиск (from, to) | Чтение по from | Чтение по to | Обновление | Удаление | Размер файлов |
 | --- | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
@@ -342,7 +358,7 @@ _3 повтора после прогрева, медианное время о
 
 ##### 10,000,000 связей
 
-_1 повтор после прогрева, медианное время одной операции. SQLite 3.53.3, Ubuntu 24.04.5 LTS, AMD EPYC 7763 64-Core Processor, 4 cores, 16 GiB, [запуск GitHub Actions](https://github.com/linksplatform/Comparisons.SQLiteVSDoublets/actions/runs/37381004025) от 2026-10-05._
+_1 повтор после прогрева, медианное время одной операции. SQLite 3.53.3; Platform.Data.Doublets 0.18.1; .NET SDK 10.0.401. Ubuntu 24.04.5 LTS, AMD EPYC 7763 64-Core Processor, 4 cores, 16 GiB, [запуск GitHub Actions](https://github.com/linksplatform/Comparisons.SQLiteVSDoublets/actions/runs/37381004025) от 2026-10-05._
 
 | Хранилище | Создание | Чтение всех | Чтение по id | Поиск (from, to) | Чтение по from | Чтение по to | Обновление | Удаление | Размер файлов |
 | --- | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
@@ -355,7 +371,9 @@ _1 повтор после прогрева, медианное время од
 | Doublets Split Volatile | 449 ns (в 10× быстрее) | 12.6 ns (в 36.6× быстрее) | 294 ns (в 7.12× быстрее) | 443 ns (в 5.32× быстрее) | 317 ns (в 7.65× быстрее) | 330 ns (в 7.26× быстрее) | 859 ns (в 11.3× быстрее) | 688 ns (в 8.95× быстрее) | — |
 | Doublets Split NonVolatile | 1.27 µs (в 8.22× быстрее) | 12.6 ns (в 36.8× быстрее) | 331 ns (в 8.88× быстрее) | 484 ns (в 7.17× быстрее) | 339 ns (в 9.86× быстрее) | 351 ns (в 9.55× быстрее) | 1.19 µs (в 21.4× быстрее) | 853 ns (в 18.1× быстрее) | 800.0 MiB |
 
-![Дуплеты на C# против SQLite, 64 бит, связи](docs/benchmarks/links-csharp-64.png)
+![Дуплеты на C# против SQLite, 64 бит, связи, линейная шкала](docs/benchmarks/links-csharp-64-linear.png)
+
+![Дуплеты на C# против SQLite, 64 бит, связи, логарифмическая шкала](docs/benchmarks/links-csharp-64.png)
 
 ## Дуплеты против SQLite как хранилище объектов
 
@@ -365,7 +383,7 @@ _1 повтор после прогрева, медианное время од
 
 ##### 100,000 записей блога
 
-_5 повторов после прогрева, медианное время одной операции. SQLite 3.53.2, Ubuntu 24.04.5 LTS, AMD EPYC 9V45 96-Core Processor, 4 cores, 16 GiB, [запуск GitHub Actions](https://github.com/linksplatform/Comparisons.SQLiteVSDoublets/actions/runs/37381004025) от 2026-10-05._
+_5 повторов после прогрева, медианное время одной операции. SQLite 3.53.2; doublets 0.5.0; rustc 1.99.0 (b940084d7 2026-09-28). Ubuntu 24.04.5 LTS, AMD EPYC 9V45 96-Core Processor, 4 cores, 16 GiB, [запуск GitHub Actions](https://github.com/linksplatform/Comparisons.SQLiteVSDoublets/actions/runs/37381004025) от 2026-10-05._
 
 | Хранилище | Создание | Чтение всех | Чтение по id | Удаление | Размер файлов |
 | --- | ---: | ---: | ---: | ---: | ---: |
@@ -382,7 +400,7 @@ _5 повторов после прогрева, медианное время 
 
 ##### 1,000,000 записей блога
 
-_1 повтор после прогрева, медианное время одной операции. SQLite 3.53.2, Ubuntu 24.04.5 LTS, AMD EPYC 9V74 80-Core Processor, 4 cores, 16 GiB, [запуск GitHub Actions](https://github.com/linksplatform/Comparisons.SQLiteVSDoublets/actions/runs/37381004025) от 2026-10-05._
+_1 повтор после прогрева, медианное время одной операции. SQLite 3.53.2; doublets 0.5.0; rustc 1.99.0 (b940084d7 2026-09-28). Ubuntu 24.04.5 LTS, AMD EPYC 9V74 80-Core Processor, 4 cores, 16 GiB, [запуск GitHub Actions](https://github.com/linksplatform/Comparisons.SQLiteVSDoublets/actions/runs/37381004025) от 2026-10-05._
 
 | Хранилище | Создание | Чтение всех | Чтение по id | Удаление | Размер файлов |
 | --- | ---: | ---: | ---: | ---: | ---: |
@@ -397,13 +415,15 @@ _1 повтор после прогрева, медианное время од
 | Doublets Split NonVolatile Cached | 6.46 µs (в 1.89× медленнее) | 882 ns (в 1.34× медленнее) | 961 ns (в 4.23× быстрее) | 6.75 µs (в 1.68× быстрее) | 400.0 MiB |
 | Doublets Split NonVolatile Uncached | 36.6 µs (в 10.7× медленнее) | 14.5 µs (в 22.1× медленнее) | 14.8 µs (в 3.64× медленнее) | 6.98 µs (в 1.63× быстрее) | 400.0 MiB |
 
-![Дуплеты на Rust против SQLite, 32 бит, объекты](docs/benchmarks/objects-rust-32.png)
+![Дуплеты на Rust против SQLite, 32 бит, объекты, линейная шкала](docs/benchmarks/objects-rust-32-linear.png)
+
+![Дуплеты на Rust против SQLite, 32 бит, объекты, логарифмическая шкала](docs/benchmarks/objects-rust-32.png)
 
 #### Тесты с 64-битным пространством адресов/идентификаторов
 
 ##### 100,000 записей блога
 
-_5 повторов после прогрева, медианное время одной операции. SQLite 3.53.2, Ubuntu 24.04.5 LTS, AMD EPYC 9V45 96-Core Processor, 4 cores, 16 GiB, [запуск GitHub Actions](https://github.com/linksplatform/Comparisons.SQLiteVSDoublets/actions/runs/37381004025) от 2026-10-05._
+_5 повторов после прогрева, медианное время одной операции. SQLite 3.53.2; doublets 0.5.0; rustc 1.99.0 (b940084d7 2026-09-28). Ubuntu 24.04.5 LTS, AMD EPYC 9V45 96-Core Processor, 4 cores, 16 GiB, [запуск GitHub Actions](https://github.com/linksplatform/Comparisons.SQLiteVSDoublets/actions/runs/37381004025) от 2026-10-05._
 
 | Хранилище | Создание | Чтение всех | Чтение по id | Удаление | Размер файлов |
 | --- | ---: | ---: | ---: | ---: | ---: |
@@ -420,7 +440,7 @@ _5 повторов после прогрева, медианное время 
 
 ##### 1,000,000 записей блога
 
-_1 повтор после прогрева, медианное время одной операции. SQLite 3.53.2, Ubuntu 24.04.5 LTS, AMD EPYC 7763 64-Core Processor, 4 cores, 16 GiB, [запуск GitHub Actions](https://github.com/linksplatform/Comparisons.SQLiteVSDoublets/actions/runs/37381004025) от 2026-10-05._
+_1 повтор после прогрева, медианное время одной операции. SQLite 3.53.2; doublets 0.5.0; rustc 1.99.0 (b940084d7 2026-09-28). Ubuntu 24.04.5 LTS, AMD EPYC 7763 64-Core Processor, 4 cores, 16 GiB, [запуск GitHub Actions](https://github.com/linksplatform/Comparisons.SQLiteVSDoublets/actions/runs/37381004025) от 2026-10-05._
 
 | Хранилище | Создание | Чтение всех | Чтение по id | Удаление | Размер файлов |
 | --- | ---: | ---: | ---: | ---: | ---: |
@@ -435,7 +455,9 @@ _1 повтор после прогрева, медианное время од
 | Doublets Split NonVolatile Cached | 8.1 µs (в 2.3× медленнее) | 952 ns (в 1.5× медленнее) | 1.02 µs (в 3.45× быстрее) | 7.65 µs (в 1.32× быстрее) | 800.0 MiB |
 | Doublets Split NonVolatile Uncached | 55.9 µs (в 15.9× медленнее) | 14.7 µs (в 23.1× медленнее) | 15.4 µs (в 4.35× медленнее) | 8.04 µs (в 1.26× быстрее) | 800.0 MiB |
 
-![Дуплеты на Rust против SQLite, 64 бит, объекты](docs/benchmarks/objects-rust-64.png)
+![Дуплеты на Rust против SQLite, 64 бит, объекты, линейная шкала](docs/benchmarks/objects-rust-64-linear.png)
+
+![Дуплеты на Rust против SQLite, 64 бит, объекты, логарифмическая шкала](docs/benchmarks/objects-rust-64.png)
 
 ### Дуплеты на C# против SQLite
 
@@ -443,7 +465,7 @@ _1 повтор после прогрева, медианное время од
 
 ##### 100,000 записей блога
 
-_5 повторов после прогрева, медианное время одной операции. SQLite 3.53.3, Ubuntu 24.04.5 LTS, AMD EPYC 7763 64-Core Processor, 4 cores, 16 GiB, [запуск GitHub Actions](https://github.com/linksplatform/Comparisons.SQLiteVSDoublets/actions/runs/37381004025) от 2026-10-05._ PostgreSQL 18.3 (Debian 18.3-1.pgdg13+1); Npgsql.EntityFrameworkCore.PostgreSQL 10.0.3.0; EF Core 10.0.4.0.
+_5 повторов после прогрева, медианное время одной операции. SQLite 3.53.3; Platform.Data.Doublets 0.18.1; .NET SDK 10.0.401. Ubuntu 24.04.5 LTS, AMD EPYC 7763 64-Core Processor, 4 cores, 16 GiB, [запуск GitHub Actions](https://github.com/linksplatform/Comparisons.SQLiteVSDoublets/actions/runs/37381004025) от 2026-10-05. PostgreSQL 18.3 (Debian 18.3-1.pgdg13+1); Npgsql.EntityFrameworkCore.PostgreSQL 10.0.3.0; EF Core 10.0.4.0._
 
 | Хранилище | Создание | Чтение всех | Чтение по id | Удаление | Размер файлов | Размер на сервере |
 | --- | ---: | ---: | ---: | ---: | ---: | ---: |
@@ -463,7 +485,7 @@ _5 повторов после прогрева, медианное время 
 
 ##### 1,000,000 записей блога
 
-_1 повтор после прогрева, медианное время одной операции. SQLite 3.53.3, Ubuntu 24.04.5 LTS, INTEL(R) XEON(R) PLATINUM 8573C, 4 cores, 16 GiB, [запуск GitHub Actions](https://github.com/linksplatform/Comparisons.SQLiteVSDoublets/actions/runs/37381004025) от 2026-10-05._ PostgreSQL 18.3 (Debian 18.3-1.pgdg13+1); Npgsql.EntityFrameworkCore.PostgreSQL 10.0.3.0; EF Core 10.0.4.0.
+_1 повтор после прогрева, медианное время одной операции. SQLite 3.53.3; Platform.Data.Doublets 0.18.1; .NET SDK 10.0.401. Ubuntu 24.04.5 LTS, INTEL(R) XEON(R) PLATINUM 8573C, 4 cores, 16 GiB, [запуск GitHub Actions](https://github.com/linksplatform/Comparisons.SQLiteVSDoublets/actions/runs/37381004025) от 2026-10-05. PostgreSQL 18.3 (Debian 18.3-1.pgdg13+1); Npgsql.EntityFrameworkCore.PostgreSQL 10.0.3.0; EF Core 10.0.4.0._
 
 | Хранилище | Создание | Чтение всех | Чтение по id | Удаление | Размер файлов | Размер на сервере |
 | --- | ---: | ---: | ---: | ---: | ---: | ---: |
@@ -481,13 +503,15 @@ _1 повтор после прогрева, медианное время од
 | Doublets Split NonVolatile Cached | 21 µs (в 2.32× медленнее) | 14 µs (в 3.79× медленнее) | 7.1 µs (≈ так же) | 10.8 µs (≈ так же) | 400.0 MiB | — |
 | Doublets Split NonVolatile Uncached | 142 µs (в 15.7× медленнее) | 460 µs (в 125× медленнее) | 460 µs (в 67.5× медленнее) | 11.2 µs (≈ так же) | 400.0 MiB | — |
 
-![Дуплеты на C# против SQLite, 32 бит, объекты](docs/benchmarks/objects-csharp-32.png)
+![Дуплеты на C# против SQLite, 32 бит, объекты, линейная шкала](docs/benchmarks/objects-csharp-32-linear.png)
+
+![Дуплеты на C# против SQLite, 32 бит, объекты, логарифмическая шкала](docs/benchmarks/objects-csharp-32.png)
 
 #### Тесты с 64-битным пространством адресов/идентификаторов
 
 ##### 100,000 записей блога
 
-_5 повторов после прогрева, медианное время одной операции. SQLite 3.53.3, Ubuntu 24.04.5 LTS, AMD EPYC 9V74 80-Core Processor, 4 cores, 16 GiB, [запуск GitHub Actions](https://github.com/linksplatform/Comparisons.SQLiteVSDoublets/actions/runs/37381004025) от 2026-10-05._ PostgreSQL 18.3 (Debian 18.3-1.pgdg13+1); Npgsql.EntityFrameworkCore.PostgreSQL 10.0.3.0; EF Core 10.0.4.0.
+_5 повторов после прогрева, медианное время одной операции. SQLite 3.53.3; Platform.Data.Doublets 0.18.1; .NET SDK 10.0.401. Ubuntu 24.04.5 LTS, AMD EPYC 9V74 80-Core Processor, 4 cores, 16 GiB, [запуск GitHub Actions](https://github.com/linksplatform/Comparisons.SQLiteVSDoublets/actions/runs/37381004025) от 2026-10-05. PostgreSQL 18.3 (Debian 18.3-1.pgdg13+1); Npgsql.EntityFrameworkCore.PostgreSQL 10.0.3.0; EF Core 10.0.4.0._
 
 | Хранилище | Создание | Чтение всех | Чтение по id | Удаление | Размер файлов | Размер на сервере |
 | --- | ---: | ---: | ---: | ---: | ---: | ---: |
@@ -507,7 +531,7 @@ _5 повторов после прогрева, медианное время 
 
 ##### 1,000,000 записей блога
 
-_1 повтор после прогрева, медианное время одной операции. SQLite 3.53.3, Ubuntu 24.04.5 LTS, AMD EPYC 9V74 80-Core Processor, 4 cores, 16 GiB, [запуск GitHub Actions](https://github.com/linksplatform/Comparisons.SQLiteVSDoublets/actions/runs/37381004025) от 2026-10-05._ PostgreSQL 18.3 (Debian 18.3-1.pgdg13+1); Npgsql.EntityFrameworkCore.PostgreSQL 10.0.3.0; EF Core 10.0.4.0.
+_1 повтор после прогрева, медианное время одной операции. SQLite 3.53.3; Platform.Data.Doublets 0.18.1; .NET SDK 10.0.401. Ubuntu 24.04.5 LTS, AMD EPYC 9V74 80-Core Processor, 4 cores, 16 GiB, [запуск GitHub Actions](https://github.com/linksplatform/Comparisons.SQLiteVSDoublets/actions/runs/37381004025) от 2026-10-05. PostgreSQL 18.3 (Debian 18.3-1.pgdg13+1); Npgsql.EntityFrameworkCore.PostgreSQL 10.0.3.0; EF Core 10.0.4.0._
 
 | Хранилище | Создание | Чтение всех | Чтение по id | Удаление | Размер файлов | Размер на сервере |
 | --- | ---: | ---: | ---: | ---: | ---: | ---: |
@@ -525,162 +549,17 @@ _1 повтор после прогрева, медианное время од
 | Doublets Split NonVolatile Cached | 27.6 µs (в 3.25× медленнее) | 11.3 µs (в 3.14× медленнее) | 7.63 µs (в 1.1× медленнее) | 13 µs (в 1.3× быстрее) | 800.0 MiB | — |
 | Doublets Split NonVolatile Uncached | 121 µs (в 14.2× медленнее) | 323 µs (в 90× медленнее) | 320 µs (в 46.2× медленнее) | 12.5 µs (в 1.35× быстрее) | 800.0 MiB | — |
 
-![Дуплеты на C# против SQLite, 64 бит, объекты](docs/benchmarks/objects-csharp-64.png)
-<!-- markdownlint-restore -->
-<!--BENCHMARK_RESULTS_END-->
-
-## Исходное сравнение
-
-Исходное сравнение объектов на C# и его исторические результаты.
-
-<!-- markdownlint-disable MD013 MD060 -->
-
-### SQLite
-
-```C#
-using System.Linq;
-using Comparisons.SQLiteVSDoublets.Model;
-
-namespace Comparisons.SQLiteVSDoublets.SQLite
-{
-    public class SQLiteTestRun : TestRun
-    {
-        public SQLiteTestRun(string dbFilename) : base(dbFilename) { }
-
-        public override void Prepare()
-        {
-            using var dbContext = new SQLiteDbContext(DbFilename);
-            dbContext.Database.EnsureCreated();
-        }
-
-        public override void CreateList()
-        {
-            using var dbContext = new SQLiteDbContext(DbFilename);
-            dbContext.BlogPosts.AddRange(BlogPosts.List);
-            dbContext.SaveChanges();
-        }
-
-        public override void ReadList()
-        {
-            using var dbContext = new SQLiteDbContext(DbFilename);
-            foreach (var blogPost in dbContext.BlogPosts)
-            {
-                ReadBlogPosts.Add(blogPost);
-            }
-        }
-
-        public override void DeleteList()
-        {
-            using var dbContext = new SQLiteDbContext(DbFilename);
-            var blogPostsToDelete = dbContext.BlogPosts.ToList();
-            dbContext.BlogPosts.RemoveRange(blogPostsToDelete);
-            dbContext.SaveChanges();
-        }
-    }
-}
-```
-
-### Дуплеты
-
-``` C#
-using System.IO;
-using Platform.IO;
-using Comparisons.SQLiteVSDoublets.Model;
-
-namespace Comparisons.SQLiteVSDoublets.Doublets
-{
-    public class DoubletsTestRun : TestRun
-    {
-        public string DbIndexFilename { get; }
-
-        public DoubletsTestRun(string dbFilename) : base(dbFilename) => DbIndexFilename = $"{Path.GetFileNameWithoutExtension(dbFilename)}.links.index";
-
-        public override void Prepare()
-        {
-            using var dbContext = new DoubletsDbContext(DbFilename, DbIndexFilename);
-        }
-
-        public override void CreateList()
-        {
-            using var dbContext = new DoubletsDbContext(DbFilename, DbIndexFilename);
-            foreach (var blogPost in BlogPosts.List)
-            {
-                dbContext.SaveBlogPost(blogPost);
-            }
-        }
-
-        public override void ReadList()
-        {
-            using var dbContext = new DoubletsDbContext(DbFilename, DbIndexFilename);
-            foreach (var blogPost in dbContext.BlogPosts)
-            {
-                ReadBlogPosts.Add(blogPost);
-            }
-        }
-
-        public override void DeleteList()
-        {
-            using var dbContext = new DoubletsDbContext(DbFilename, DbIndexFilename);
-            var blogPostsToDelete = dbContext.BlogPosts;
-            foreach (var blogPost in blogPostsToDelete)
-            {
-                dbContext.Delete((uint)blogPost.Id);
-            }
-        }
-
-        protected override void DeleteDatabase()
-        {
-            File.Delete(DbFilename);
-            File.Delete(DbIndexFilename);
-        }
-
-        protected override long GetDatabaseSizeInBytes() => FileHelpers.GetSize(DbFilename) + FileHelpers.GetSize(DbIndexFilename);
-    }
-}
-```
-
-### [Результат](https://www.icloud.com/keynote/0cYVNWkWD5RLU0k-XIBs3qWkA#Sqlite_vs_Doublets)
-
-#### Производительность
-
-![Изображение с результатом сравнения производительности SQLite и Дуплетов.](https://raw.githubusercontent.com/linksplatform/Documentation/master/doc/Examples/sqlite_vs_doublets_performance.png "Результат сравнения производительности SQLite и Дуплетов")
-
-#### Использование пространства на диске
-
-![Изображение с результатом сравнения использования пространства на диске SQLite и Дуплетов.](https://raw.githubusercontent.com/linksplatform/Documentation/master/doc/Examples/sqlite_vs_doublets_disk_usage.png "Результат сравнения использования пространства на диске SQLite и Дуплетов")
-
-#### Использование оперативной памяти
-
-![Изображение с результатом сравнения использования оперативной памяти SQLite и Дуплетов.](https://raw.githubusercontent.com/linksplatform/Documentation/master/doc/Examples/sqlite_vs_doublets_ram_usage.png "Результат сравнения использования оперативной памяти SQLite и Дуплетов")
+![Дуплеты на C# против SQLite, 64 бит, объекты, линейная шкала](docs/benchmarks/objects-csharp-64-linear.png)
 
-#### Исходные данные
+![Дуплеты на C# против SQLite, 64 бит, объекты, логарифмическая шкала](docs/benchmarks/objects-csharp-64.png)
 
-``` ini
+## Выводы
 
-BenchmarkDotNet=v0.12.0, OS=Windows 10.0.18362
-Intel Core i7-6700K CPU 4.00GHz (Skylake), 1 CPU, 8 logical and 4 physical cores
-.NET Core SDK=3.0.100
-  [Host]     : .NET Core 3.0.0 (CoreCLR 4.700.19.46205, CoreFX 4.700.19.46214), X64 RyuJIT
-  Job-AYEMIX : .NET Core 3.0.0 (CoreCLR 4.700.19.46205, CoreFX 4.700.19.46214), X64 RyuJIT
-
-InvocationCount=1  IterationCount=1  UnrollFactor=1  
-WarmupCount=2  
-
-```
-
-|   Method |      N |        Mean | Error |        Gen 0 |       Gen 1 | Gen 2 |   Allocated | SizeAfterCreation |
-|--------- |------- |------------:|------:|-------------:|------------:|------:|------------:|------------------:|
-|   **SQLite** |   **1000** |    **719.1 ms** |    **NA** |    **5000.0000** |           **-** |     **-** |    **30.67 MB** |            **925696** |
-| Doublets |   1000 |    145.0 ms |    NA |   34000.0000 |   1000.0000 |     - |   139.37 MB |            767616 |
-|   **SQLite** |  **10000** |  **2,770.3 ms** |    **NA** |   **64000.0000** |  **19000.0000** |     **-** |   **315.71 MB** |           **9056256** |
-| Doublets |  10000 |  1,003.8 ms |    NA |  304000.0000 |  31000.0000 |     - |  1220.55 MB |           6528256 |
-|   **SQLite** | **100000** | **35,853.8 ms** |    **NA** |  **680000.0000** | **151000.0000** |     **-** |  **3234.09 MB** |          **90890240** |
-| Doublets | 100000 | 13,083.4 ms |    NA | 3088000.0000 | 328000.0000 |     - | 12356.33 MB |          64192256 |
+Каждое сравнение использует SQLite той же надёжности хранения и то же правило учёта шума, что и таблицы. Числа охватывают измеренные варианты, размеры и разрядности id и не являются общим рейтингом скорости. Использование ОЗУ не измеряется.
 
+- Rust, связи: Дуплеты быстрее в 187, медленнее в 3, примерно равны SQLite в 2 сравнениях операций.
+- C#, связи: Дуплеты быстрее в 192, медленнее в 0, примерно равны SQLite в 0 сравнениях операций.
+- Rust, объекты: Дуплеты быстрее в 28, медленнее в 98, примерно равны SQLite в 2 сравнениях операций.
+- C#, объекты: Дуплеты быстрее в 10, медленнее в 109, примерно равны SQLite в 9 сравнениях операций.
 <!-- markdownlint-restore -->
-
-### Заключение
-
-В этом конкретном сравнении Дуплеты работают быстрее и используют меньше памяти
-на диске, но это достигается за счёт дополнительного использования оперативной
-памяти (Sqlite использует её меньше).
+<!--BENCHMARK_RESULTS_END-->
diff --git a/cpp/README.md b/cpp/README.md
deleted file mode 100644
index 0b0ecfa..0000000
--- a/cpp/README.md
+++ /dev/null
@@ -1,10 +0,0 @@
-<!-- markdownlint-disable MD043 -->
-# C++ placeholder
-
-This directory contains an empty Conan manifest and no C++ implementation,
-`CMakeLists.txt`, tests, or package metadata. The obsolete build and release
-workflow was retired in issue #110; it could not build or publish this tree.
-
-A future C++ implementation must add a real build target and tests before
-enabling CI. Use a supported Clang version on a pinned runner, Conan 2 profiles
-and generators, pull request validation, and release jobs gated on passing tests.
diff --git a/cpp/conanfile.txt b/cpp/conanfile.txt
deleted file mode 100644
index a25eecc..0000000
--- a/cpp/conanfile.txt
+++ /dev/null
@@ -1,4 +0,0 @@
-[requires]
-
-[generators]
-cmake
diff --git a/docs/benchmarks/links-csharp-32-linear.png b/docs/benchmarks/links-csharp-32-linear.png
new file mode 100644
index 0000000..b3908e9
Binary files /dev/null and b/docs/benchmarks/links-csharp-32-linear.png differ
diff --git a/docs/benchmarks/links-csharp-32.png b/docs/benchmarks/links-csharp-32.png
index aff9d3f..a408c70 100644
Binary files a/docs/benchmarks/links-csharp-32.png and b/docs/benchmarks/links-csharp-32.png differ
diff --git a/docs/benchmarks/links-csharp-64-linear.png b/docs/benchmarks/links-csharp-64-linear.png
new file mode 100644
index 0000000..ddc31b4
Binary files /dev/null and b/docs/benchmarks/links-csharp-64-linear.png differ
diff --git a/docs/benchmarks/links-csharp-64.png b/docs/benchmarks/links-csharp-64.png
index dae5a23..b431558 100644
Binary files a/docs/benchmarks/links-csharp-64.png and b/docs/benchmarks/links-csharp-64.png differ
diff --git a/docs/benchmarks/links-rust-32-linear.png b/docs/benchmarks/links-rust-32-linear.png
new file mode 100644
index 0000000..615ca54
Binary files /dev/null and b/docs/benchmarks/links-rust-32-linear.png differ
diff --git a/docs/benchmarks/links-rust-32.png b/docs/benchmarks/links-rust-32.png
index 005d45e..a0648cf 100644
Binary files a/docs/benchmarks/links-rust-32.png and b/docs/benchmarks/links-rust-32.png differ
diff --git a/docs/benchmarks/links-rust-64-linear.png b/docs/benchmarks/links-rust-64-linear.png
new file mode 100644
index 0000000..e0e9a65
Binary files /dev/null and b/docs/benchmarks/links-rust-64-linear.png differ
diff --git a/docs/benchmarks/links-rust-64.png b/docs/benchmarks/links-rust-64.png
index 1c2277c..a14d6c2 100644
Binary files a/docs/benchmarks/links-rust-64.png and b/docs/benchmarks/links-rust-64.png differ
diff --git a/docs/benchmarks/objects-csharp-32-linear.png b/docs/benchmarks/objects-csharp-32-linear.png
new file mode 100644
index 0000000..8fc6e03
Binary files /dev/null and b/docs/benchmarks/objects-csharp-32-linear.png differ
diff --git a/docs/benchmarks/objects-csharp-32.png b/docs/benchmarks/objects-csharp-32.png
index 5ca27b3..00345c0 100644
Binary files a/docs/benchmarks/objects-csharp-32.png and b/docs/benchmarks/objects-csharp-32.png differ
diff --git a/docs/benchmarks/objects-csharp-64-linear.png b/docs/benchmarks/objects-csharp-64-linear.png
new file mode 100644
index 0000000..49a1d26
Binary files /dev/null and b/docs/benchmarks/objects-csharp-64-linear.png differ
diff --git a/docs/benchmarks/objects-csharp-64.png b/docs/benchmarks/objects-csharp-64.png
index 5bd50ef..d1c6502 100644
Binary files a/docs/benchmarks/objects-csharp-64.png and b/docs/benchmarks/objects-csharp-64.png differ
diff --git a/docs/benchmarks/objects-rust-32-linear.png b/docs/benchmarks/objects-rust-32-linear.png
new file mode 100644
index 0000000..7046c58
Binary files /dev/null and b/docs/benchmarks/objects-rust-32-linear.png differ
diff --git a/docs/benchmarks/objects-rust-32.png b/docs/benchmarks/objects-rust-32.png
index a600e78..66328d0 100644
Binary files a/docs/benchmarks/objects-rust-32.png and b/docs/benchmarks/objects-rust-32.png differ
diff --git a/docs/benchmarks/objects-rust-64-linear.png b/docs/benchmarks/objects-rust-64-linear.png
new file mode 100644
index 0000000..fbd006a
Binary files /dev/null and b/docs/benchmarks/objects-rust-64-linear.png differ
diff --git a/docs/benchmarks/objects-rust-64.png b/docs/benchmarks/objects-rust-64.png
index 7d8bada..326c1b7 100644
Binary files a/docs/benchmarks/objects-rust-64.png and b/docs/benchmarks/objects-rust-64.png differ
diff --git a/docs/benchmarks/results/links-csharp-32-100000.json b/docs/benchmarks/results/links-csharp-32-100000.json
index 0299704..94dc7ca 100644
--- a/docs/benchmarks/results/links-csharp-32-100000.json
+++ b/docs/benchmarks/results/links-csharp-32-100000.json
@@ -1159,5 +1159,11 @@
   "source_sha": "4e269b47eb3e6d0a264b57f27b9a51832562a99b",
   "tree_sha": "412060b55aeed6f627c46da833f68ae21c71a6aa",
   "date": "2026-10-05",
-  "machine": "Ubuntu 24.04.5 LTS, AMD EPYC 7763 64-Core Processor, 4 cores, 16 GiB"
+  "machine": "Ubuntu 24.04.5 LTS, AMD EPYC 7763 64-Core Processor, 4 cores, 16 GiB",
+  "software": {
+    "libraries": {
+      "Platform.Data.Doublets": "0.18.1"
+    },
+    "toolchain": ".NET SDK 10.0.401"
+  }
 }
diff --git a/docs/benchmarks/results/links-csharp-32-1000000.json b/docs/benchmarks/results/links-csharp-32-1000000.json
index 6119abb..2bbec2d 100644
--- a/docs/benchmarks/results/links-csharp-32-1000000.json
+++ b/docs/benchmarks/results/links-csharp-32-1000000.json
@@ -711,5 +711,11 @@
   "source_sha": "4e269b47eb3e6d0a264b57f27b9a51832562a99b",
   "tree_sha": "412060b55aeed6f627c46da833f68ae21c71a6aa",
   "date": "2026-10-05",
-  "machine": "Ubuntu 24.04.5 LTS, AMD EPYC 7763 64-Core Processor, 4 cores, 16 GiB"
+  "machine": "Ubuntu 24.04.5 LTS, AMD EPYC 7763 64-Core Processor, 4 cores, 16 GiB",
+  "software": {
+    "libraries": {
+      "Platform.Data.Doublets": "0.18.1"
+    },
+    "toolchain": ".NET SDK 10.0.401"
+  }
 }
diff --git a/docs/benchmarks/results/links-csharp-32-10000000.json b/docs/benchmarks/results/links-csharp-32-10000000.json
index 790a0cd..c54f015 100644
--- a/docs/benchmarks/results/links-csharp-32-10000000.json
+++ b/docs/benchmarks/results/links-csharp-32-10000000.json
@@ -583,5 +583,11 @@
   "source_sha": "4e269b47eb3e6d0a264b57f27b9a51832562a99b",
   "tree_sha": "412060b55aeed6f627c46da833f68ae21c71a6aa",
   "date": "2026-10-05",
-  "machine": "Ubuntu 24.04.5 LTS, INTEL(R) XEON(R) PLATINUM 8573C, 4 cores, 16 GiB"
+  "machine": "Ubuntu 24.04.5 LTS, INTEL(R) XEON(R) PLATINUM 8573C, 4 cores, 16 GiB",
+  "software": {
+    "libraries": {
+      "Platform.Data.Doublets": "0.18.1"
+    },
+    "toolchain": ".NET SDK 10.0.401"
+  }
 }
diff --git a/docs/benchmarks/results/links-csharp-64-100000.json b/docs/benchmarks/results/links-csharp-64-100000.json
index c7449f1..9a80a0e 100644
--- a/docs/benchmarks/results/links-csharp-64-100000.json
+++ b/docs/benchmarks/results/links-csharp-64-100000.json
@@ -1159,5 +1159,11 @@
   "source_sha": "4e269b47eb3e6d0a264b57f27b9a51832562a99b",
   "tree_sha": "412060b55aeed6f627c46da833f68ae21c71a6aa",
   "date": "2026-10-05",
-  "machine": "Ubuntu 24.04.5 LTS, AMD EPYC 9V74 80-Core Processor, 4 cores, 16 GiB"
+  "machine": "Ubuntu 24.04.5 LTS, AMD EPYC 9V74 80-Core Processor, 4 cores, 16 GiB",
+  "software": {
+    "libraries": {
+      "Platform.Data.Doublets": "0.18.1"
+    },
+    "toolchain": ".NET SDK 10.0.401"
+  }
 }
diff --git a/docs/benchmarks/results/links-csharp-64-1000000.json b/docs/benchmarks/results/links-csharp-64-1000000.json
index e929eb7..5d7f84d 100644
--- a/docs/benchmarks/results/links-csharp-64-1000000.json
+++ b/docs/benchmarks/results/links-csharp-64-1000000.json
@@ -711,5 +711,11 @@
   "source_sha": "4e269b47eb3e6d0a264b57f27b9a51832562a99b",
   "tree_sha": "412060b55aeed6f627c46da833f68ae21c71a6aa",
   "date": "2026-10-05",
-  "machine": "Ubuntu 24.04.5 LTS, AMD EPYC 9V74 80-Core Processor, 4 cores, 16 GiB"
+  "machine": "Ubuntu 24.04.5 LTS, AMD EPYC 9V74 80-Core Processor, 4 cores, 16 GiB",
+  "software": {
+    "libraries": {
+      "Platform.Data.Doublets": "0.18.1"
+    },
+    "toolchain": ".NET SDK 10.0.401"
+  }
 }
diff --git a/docs/benchmarks/results/links-csharp-64-10000000.json b/docs/benchmarks/results/links-csharp-64-10000000.json
index 23b5bd2..1066ce7 100644
--- a/docs/benchmarks/results/links-csharp-64-10000000.json
+++ b/docs/benchmarks/results/links-csharp-64-10000000.json
@@ -583,5 +583,11 @@
   "source_sha": "4e269b47eb3e6d0a264b57f27b9a51832562a99b",
   "tree_sha": "412060b55aeed6f627c46da833f68ae21c71a6aa",
   "date": "2026-10-05",
-  "machine": "Ubuntu 24.04.5 LTS, AMD EPYC 7763 64-Core Processor, 4 cores, 16 GiB"
+  "machine": "Ubuntu 24.04.5 LTS, AMD EPYC 7763 64-Core Processor, 4 cores, 16 GiB",
+  "software": {
+    "libraries": {
+      "Platform.Data.Doublets": "0.18.1"
+    },
+    "toolchain": ".NET SDK 10.0.401"
+  }
 }
diff --git a/docs/benchmarks/results/links-rust-32-100000.json b/docs/benchmarks/results/links-rust-32-100000.json
index 1068928..e4cd651 100644
--- a/docs/benchmarks/results/links-rust-32-100000.json
+++ b/docs/benchmarks/results/links-rust-32-100000.json
@@ -865,5 +865,11 @@
   "source_sha": "4e269b47eb3e6d0a264b57f27b9a51832562a99b",
   "tree_sha": "412060b55aeed6f627c46da833f68ae21c71a6aa",
   "date": "2026-10-05",
-  "machine": "Ubuntu 24.04.5 LTS, AMD EPYC 7763 64-Core Processor, 4 cores, 16 GiB"
+  "machine": "Ubuntu 24.04.5 LTS, AMD EPYC 7763 64-Core Processor, 4 cores, 16 GiB",
+  "software": {
+    "libraries": {
+      "doublets": "0.5.0"
+    },
+    "toolchain": "rustc 1.99.0 (b940084d7 2026-09-28)"
+  }
 }
diff --git a/docs/benchmarks/results/links-rust-32-1000000.json b/docs/benchmarks/results/links-rust-32-1000000.json
index 35318f7..ff0afb5 100644
--- a/docs/benchmarks/results/links-rust-32-1000000.json
+++ b/docs/benchmarks/results/links-rust-32-1000000.json
@@ -529,5 +529,11 @@
   "source_sha": "4e269b47eb3e6d0a264b57f27b9a51832562a99b",
   "tree_sha": "412060b55aeed6f627c46da833f68ae21c71a6aa",
   "date": "2026-10-05",
-  "machine": "Ubuntu 24.04.5 LTS, AMD EPYC 7763 64-Core Processor, 4 cores, 16 GiB"
+  "machine": "Ubuntu 24.04.5 LTS, AMD EPYC 7763 64-Core Processor, 4 cores, 16 GiB",
+  "software": {
+    "libraries": {
+      "doublets": "0.5.0"
+    },
+    "toolchain": "rustc 1.99.0 (b940084d7 2026-09-28)"
+  }
 }
diff --git a/docs/benchmarks/results/links-rust-32-10000000.json b/docs/benchmarks/results/links-rust-32-10000000.json
index 1edff3a..2ef23df 100644
--- a/docs/benchmarks/results/links-rust-32-10000000.json
+++ b/docs/benchmarks/results/links-rust-32-10000000.json
@@ -433,5 +433,11 @@
   "source_sha": "4e269b47eb3e6d0a264b57f27b9a51832562a99b",
   "tree_sha": "412060b55aeed6f627c46da833f68ae21c71a6aa",
   "date": "2026-10-05",
-  "machine": "Ubuntu 24.04.5 LTS, AMD EPYC 9V74 80-Core Processor, 4 cores, 16 GiB"
+  "machine": "Ubuntu 24.04.5 LTS, AMD EPYC 9V74 80-Core Processor, 4 cores, 16 GiB",
+  "software": {
+    "libraries": {
+      "doublets": "0.5.0"
+    },
+    "toolchain": "rustc 1.99.0 (b940084d7 2026-09-28)"
+  }
 }
diff --git a/docs/benchmarks/results/links-rust-64-100000.json b/docs/benchmarks/results/links-rust-64-100000.json
index 5b9f627..38b1129 100644
--- a/docs/benchmarks/results/links-rust-64-100000.json
+++ b/docs/benchmarks/results/links-rust-64-100000.json
@@ -865,5 +865,11 @@
   "source_sha": "4e269b47eb3e6d0a264b57f27b9a51832562a99b",
   "tree_sha": "412060b55aeed6f627c46da833f68ae21c71a6aa",
   "date": "2026-10-05",
-  "machine": "Ubuntu 24.04.5 LTS, AMD EPYC 7763 64-Core Processor, 4 cores, 16 GiB"
+  "machine": "Ubuntu 24.04.5 LTS, AMD EPYC 7763 64-Core Processor, 4 cores, 16 GiB",
+  "software": {
+    "libraries": {
+      "doublets": "0.5.0"
+    },
+    "toolchain": "rustc 1.99.0 (b940084d7 2026-09-28)"
+  }
 }
diff --git a/docs/benchmarks/results/links-rust-64-1000000.json b/docs/benchmarks/results/links-rust-64-1000000.json
index 4d821df..4c18236 100644
--- a/docs/benchmarks/results/links-rust-64-1000000.json
+++ b/docs/benchmarks/results/links-rust-64-1000000.json
@@ -529,5 +529,11 @@
   "source_sha": "4e269b47eb3e6d0a264b57f27b9a51832562a99b",
   "tree_sha": "412060b55aeed6f627c46da833f68ae21c71a6aa",
   "date": "2026-10-05",
-  "machine": "Ubuntu 24.04.5 LTS, AMD EPYC 7763 64-Core Processor, 4 cores, 16 GiB"
+  "machine": "Ubuntu 24.04.5 LTS, AMD EPYC 7763 64-Core Processor, 4 cores, 16 GiB",
+  "software": {
+    "libraries": {
+      "doublets": "0.5.0"
+    },
+    "toolchain": "rustc 1.99.0 (b940084d7 2026-09-28)"
+  }
 }
diff --git a/docs/benchmarks/results/links-rust-64-10000000.json b/docs/benchmarks/results/links-rust-64-10000000.json
index 440fbe5..f7addfa 100644
--- a/docs/benchmarks/results/links-rust-64-10000000.json
+++ b/docs/benchmarks/results/links-rust-64-10000000.json
@@ -433,5 +433,11 @@
   "source_sha": "4e269b47eb3e6d0a264b57f27b9a51832562a99b",
   "tree_sha": "412060b55aeed6f627c46da833f68ae21c71a6aa",
   "date": "2026-10-05",
-  "machine": "Ubuntu 24.04.5 LTS, AMD EPYC 9V74 80-Core Processor, 4 cores, 16 GiB"
+  "machine": "Ubuntu 24.04.5 LTS, AMD EPYC 9V74 80-Core Processor, 4 cores, 16 GiB",
+  "software": {
+    "libraries": {
+      "doublets": "0.5.0"
+    },
+    "toolchain": "rustc 1.99.0 (b940084d7 2026-09-28)"
+  }
 }
diff --git a/docs/benchmarks/results/objects-csharp-32-100000.json b/docs/benchmarks/results/objects-csharp-32-100000.json
index 66776a8..651072e 100644
--- a/docs/benchmarks/results/objects-csharp-32-100000.json
+++ b/docs/benchmarks/results/objects-csharp-32-100000.json
@@ -732,5 +732,11 @@
   "source_sha": "4e269b47eb3e6d0a264b57f27b9a51832562a99b",
   "tree_sha": "412060b55aeed6f627c46da833f68ae21c71a6aa",
   "date": "2026-10-05",
-  "machine": "Ubuntu 24.04.5 LTS, AMD EPYC 7763 64-Core Processor, 4 cores, 16 GiB"
+  "machine": "Ubuntu 24.04.5 LTS, AMD EPYC 7763 64-Core Processor, 4 cores, 16 GiB",
+  "software": {
+    "libraries": {
+      "Platform.Data.Doublets": "0.18.1"
+    },
+    "toolchain": ".NET SDK 10.0.401"
+  }
 }
diff --git a/docs/benchmarks/results/objects-csharp-32-1000000.json b/docs/benchmarks/results/objects-csharp-32-1000000.json
index 91004d8..7a4e962 100644
--- a/docs/benchmarks/results/objects-csharp-32-1000000.json
+++ b/docs/benchmarks/results/objects-csharp-32-1000000.json
@@ -524,5 +524,11 @@
   "source_sha": "4e269b47eb3e6d0a264b57f27b9a51832562a99b",
   "tree_sha": "412060b55aeed6f627c46da833f68ae21c71a6aa",
   "date": "2026-10-05",
-  "machine": "Ubuntu 24.04.5 LTS, INTEL(R) XEON(R) PLATINUM 8573C, 4 cores, 16 GiB"
+  "machine": "Ubuntu 24.04.5 LTS, INTEL(R) XEON(R) PLATINUM 8573C, 4 cores, 16 GiB",
+  "software": {
+    "libraries": {
+      "Platform.Data.Doublets": "0.18.1"
+    },
+    "toolchain": ".NET SDK 10.0.401"
+  }
 }
diff --git a/docs/benchmarks/results/objects-csharp-64-100000.json b/docs/benchmarks/results/objects-csharp-64-100000.json
index 0666978..75783d1 100644
--- a/docs/benchmarks/results/objects-csharp-64-100000.json
+++ b/docs/benchmarks/results/objects-csharp-64-100000.json
@@ -732,5 +732,11 @@
   "source_sha": "4e269b47eb3e6d0a264b57f27b9a51832562a99b",
   "tree_sha": "412060b55aeed6f627c46da833f68ae21c71a6aa",
   "date": "2026-10-05",
-  "machine": "Ubuntu 24.04.5 LTS, AMD EPYC 9V74 80-Core Processor, 4 cores, 16 GiB"
+  "machine": "Ubuntu 24.04.5 LTS, AMD EPYC 9V74 80-Core Processor, 4 cores, 16 GiB",
+  "software": {
+    "libraries": {
+      "Platform.Data.Doublets": "0.18.1"
+    },
+    "toolchain": ".NET SDK 10.0.401"
+  }
 }
diff --git a/docs/benchmarks/results/objects-csharp-64-1000000.json b/docs/benchmarks/results/objects-csharp-64-1000000.json
index cd27699..7bb4544 100644
--- a/docs/benchmarks/results/objects-csharp-64-1000000.json
+++ b/docs/benchmarks/results/objects-csharp-64-1000000.json
@@ -524,5 +524,11 @@
   "source_sha": "4e269b47eb3e6d0a264b57f27b9a51832562a99b",
   "tree_sha": "412060b55aeed6f627c46da833f68ae21c71a6aa",
   "date": "2026-10-05",
-  "machine": "Ubuntu 24.04.5 LTS, AMD EPYC 9V74 80-Core Processor, 4 cores, 16 GiB"
+  "machine": "Ubuntu 24.04.5 LTS, AMD EPYC 9V74 80-Core Processor, 4 cores, 16 GiB",
+  "software": {
+    "libraries": {
+      "Platform.Data.Doublets": "0.18.1"
+    },
+    "toolchain": ".NET SDK 10.0.401"
+  }
 }
diff --git a/docs/benchmarks/results/objects-rust-32-100000.json b/docs/benchmarks/results/objects-rust-32-100000.json
index 54c4f90..3db654d 100644
--- a/docs/benchmarks/results/objects-rust-32-100000.json
+++ b/docs/benchmarks/results/objects-rust-32-100000.json
@@ -553,5 +553,11 @@
   "source_sha": "4e269b47eb3e6d0a264b57f27b9a51832562a99b",
   "tree_sha": "412060b55aeed6f627c46da833f68ae21c71a6aa",
   "date": "2026-10-05",
-  "machine": "Ubuntu 24.04.5 LTS, AMD EPYC 9V45 96-Core Processor, 4 cores, 16 GiB"
+  "machine": "Ubuntu 24.04.5 LTS, AMD EPYC 9V45 96-Core Processor, 4 cores, 16 GiB",
+  "software": {
+    "libraries": {
+      "doublets": "0.5.0"
+    },
+    "toolchain": "rustc 1.99.0 (b940084d7 2026-09-28)"
+  }
 }
diff --git a/docs/benchmarks/results/objects-rust-32-1000000.json b/docs/benchmarks/results/objects-rust-32-1000000.json
index 41a9a7e..529efae 100644
--- a/docs/benchmarks/results/objects-rust-32-1000000.json
+++ b/docs/benchmarks/results/objects-rust-32-1000000.json
@@ -393,5 +393,11 @@
   "source_sha": "4e269b47eb3e6d0a264b57f27b9a51832562a99b",
   "tree_sha": "412060b55aeed6f627c46da833f68ae21c71a6aa",
   "date": "2026-10-05",
-  "machine": "Ubuntu 24.04.5 LTS, AMD EPYC 9V74 80-Core Processor, 4 cores, 16 GiB"
+  "machine": "Ubuntu 24.04.5 LTS, AMD EPYC 9V74 80-Core Processor, 4 cores, 16 GiB",
+  "software": {
+    "libraries": {
+      "doublets": "0.5.0"
+    },
+    "toolchain": "rustc 1.99.0 (b940084d7 2026-09-28)"
+  }
 }
diff --git a/docs/benchmarks/results/objects-rust-64-100000.json b/docs/benchmarks/results/objects-rust-64-100000.json
index 553c4bb..a80e09b 100644
--- a/docs/benchmarks/results/objects-rust-64-100000.json
+++ b/docs/benchmarks/results/objects-rust-64-100000.json
@@ -553,5 +553,11 @@
   "source_sha": "4e269b47eb3e6d0a264b57f27b9a51832562a99b",
   "tree_sha": "412060b55aeed6f627c46da833f68ae21c71a6aa",
   "date": "2026-10-05",
-  "machine": "Ubuntu 24.04.5 LTS, AMD EPYC 9V45 96-Core Processor, 4 cores, 16 GiB"
+  "machine": "Ubuntu 24.04.5 LTS, AMD EPYC 9V45 96-Core Processor, 4 cores, 16 GiB",
+  "software": {
+    "libraries": {
+      "doublets": "0.5.0"
+    },
+    "toolchain": "rustc 1.99.0 (b940084d7 2026-09-28)"
+  }
 }
diff --git a/docs/benchmarks/results/objects-rust-64-1000000.json b/docs/benchmarks/results/objects-rust-64-1000000.json
index b406af2..4f19373 100644
--- a/docs/benchmarks/results/objects-rust-64-1000000.json
+++ b/docs/benchmarks/results/objects-rust-64-1000000.json
@@ -393,5 +393,11 @@
   "source_sha": "4e269b47eb3e6d0a264b57f27b9a51832562a99b",
   "tree_sha": "412060b55aeed6f627c46da833f68ae21c71a6aa",
   "date": "2026-10-05",
-  "machine": "Ubuntu 24.04.5 LTS, AMD EPYC 7763 64-Core Processor, 4 cores, 16 GiB"
+  "machine": "Ubuntu 24.04.5 LTS, AMD EPYC 7763 64-Core Processor, 4 cores, 16 GiB",
+  "software": {
+    "libraries": {
+      "doublets": "0.5.0"
+    },
+    "toolchain": "rustc 1.99.0 (b940084d7 2026-09-28)"
+  }
 }
diff --git a/experiments/backfill_benchmark_provenance.py b/experiments/backfill_benchmark_provenance.py
new file mode 100644
index 0000000..700b949
--- /dev/null
+++ b/experiments/backfill_benchmark_provenance.py
@@ -0,0 +1,70 @@
+"""Recover provenance for committed results from their original CI logs and source commit."""
+
+import argparse
+import json
+import re
+import subprocess  # nosec B404 # Only local git object reads, with argument lists and no shell.
+import sys
+import tempfile
+from pathlib import Path
+
+sys.path.insert(0, str(Path(__file__).resolve().parents[1] / "scripts"))
+from benchmark_provenance import record  # noqa: E402
+from ci_process import executable  # noqa: E402
+
+
+def backfill(directory, log):
+    versions = {}
+    source_shas = {}
+    for line in log.read_text(encoding="utf-8").splitlines():
+        parts = line.split("\t", 2)
+        if len(parts) != 3:
+            continue
+        job, _, timed_message = parts
+        message = timed_message.partition(" ")[2]
+        source = re.search(r"\bSOURCE_SHA: ([0-9a-f]{40})\b", message)
+        if source:
+            source_shas[job] = source[1]
+        if re.fullmatch(r"rustc \d+\.\d+\.\d+ .*", message):
+            versions[job] = message
+        match = re.search(r"\.NET Core SDK with version '([^']+)'", message)
+        if match:
+            versions[job] = f".NET SDK {match[1]}"
+    for path in sorted(directory.glob("*.json")):
+        data = json.loads(path.read_text(encoding="utf-8"))
+        language = "rust" if data["language"] == "Rust" else "csharp"
+        job = f"{language} {data['category']} {data['bits']} bit {data['size']}"
+        toolchain = versions[job]
+        sha = data["source_sha"]
+        if not re.fullmatch(r"[0-9a-f]{40}", sha):
+            raise ValueError(f"Invalid measured source SHA: {sha}")
+        if source_shas.get(job) != sha:
+            raise ValueError(f"{path}: logs do not match the measured source commit")
+        with tempfile.TemporaryDirectory() as temporary:
+            root = Path(temporary)
+            manifest = (
+                "rust/Cargo.lock" if language == "rust" else "csharp/SQLiteVSDoublets/SQLiteVSDoublets.csproj"
+            )
+            original = subprocess.run(  # nosec B603 # Validated SHA and fixed manifest paths; no shell.
+                [executable("git"), "show", f"{sha}:{manifest}"],
+                capture_output=True,
+                text=True,
+                check=True,
+            ).stdout
+            target = root / manifest
+            target.parent.mkdir(parents=True)
+            target.write_text(original, encoding="utf-8")
+            record(path, toolchain, root)
+        print(f"{path.name}: {toolchain}, libraries from {sha}")
+
+
+def main():
+    parser = argparse.ArgumentParser(description=__doc__)
+    parser.add_argument("results", type=Path)
+    parser.add_argument("log", type=Path, help="gh run view RUN --log output from the measured run")
+    args = parser.parse_args()
+    backfill(args.results, args.log)
+
+
+if __name__ == "__main__":
+    main()
diff --git a/scripts/benchmark_provenance.py b/scripts/benchmark_provenance.py
new file mode 100644
index 0000000..8b7964e
--- /dev/null
+++ b/scripts/benchmark_provenance.py
@@ -0,0 +1,48 @@
+"""Record the library and toolchain versions used by a benchmark job."""
+
+import argparse
+import json
+import re
+import tomllib
+from pathlib import Path
+
+ROOT = Path(__file__).resolve().parents[1]
+
+
+def library_versions(language, root=ROOT):
+    """Read the Rust lockfile or C# package reference for the measured Doublets library."""
+    if language == "Rust":
+        lock = tomllib.loads((root / "rust/Cargo.lock").read_text(encoding="utf-8"))
+        versions = {entry["version"] for entry in lock["package"] if entry["name"] == "doublets"}
+        if len(versions) != 1:
+            raise ValueError("Cargo.lock must resolve exactly one doublets version")
+        return {"doublets": versions.pop()}
+    if language != "C#":
+        raise ValueError(f"Unknown benchmark language: {language}")
+    project = (root / "csharp/SQLiteVSDoublets/SQLiteVSDoublets.csproj").read_text(encoding="utf-8")
+    for reference in re.findall(r"<PackageReference\b[^>]*>", project):
+        attributes = dict(re.findall(r'([\w]+)\s*=\s*"([^"]*)"', reference))
+        if attributes.get("Include") == "Platform.Data.Doublets" and attributes.get("Version"):
+            return {"Platform.Data.Doublets": attributes["Version"]}
+    raise ValueError("Missing Platform.Data.Doublets version in the C# project")
+
+
+def record(path, toolchain, root=ROOT):
+    """Add a software snapshot while preserving every measurement and existing metadata field."""
+    if not toolchain.strip():
+        raise ValueError("Missing measured toolchain version")
+    data = json.loads(path.read_text(encoding="utf-8"))
+    data["software"] = {"libraries": library_versions(data["language"], root), "toolchain": toolchain}
+    path.write_text(json.dumps(data, indent=2, ensure_ascii=False) + "\n", encoding="utf-8")
+
+
+def main():
+    parser = argparse.ArgumentParser(description=__doc__)
+    parser.add_argument("report", type=Path)
+    parser.add_argument("--toolchain", required=True, help="rustc --version or the measured .NET SDK version")
+    args = parser.parse_args()
+    record(args.report, args.toolchain)
+
+
+if __name__ == "__main__":
+    main()
diff --git a/scripts/benchmark_report.py b/scripts/benchmark_report.py
index 7508196..3a73810 100644
--- a/scripts/benchmark_report.py
+++ b/scripts/benchmark_report.py
@@ -16,6 +16,8 @@
 from pathlib import Path
 from typing import Any
 
+from benchmark_provenance import library_versions
+
 START_MARKER = "<!--BENCHMARK_RESULTS_START-->"
 END_MARKER = "<!--BENCHMARK_RESULTS_END-->"
 # The wide tables cannot be wrapped, and the hierarchy repeats headings under different parents.
@@ -75,7 +77,14 @@ def russian_plural(count, one, few, many):
         "faster": "{ratio}× faster",
         "slower": "{ratio}× slower",
         "same": "≈ same",
-        "provenance": "_{repetitions} after a warm-up, median time per operation. SQLite {sqlite}, {machine}, {source}._",
+        "provenance": "{repetitions} after a warm-up, median time per operation. SQLite {sqlite}; {software}. {machine}, {source}.",
+        "unrecorded": "version at measurement not recorded",
+        "toolchain_missing": "toolchain version not recorded",
+        "log_scale": "log scale",
+        "linear_scale": "linear scale",
+        "conclusions": "Conclusions",
+        "summary": "{language}, {category}: {faster} faster, {slower} slower, {same} approximately equal Doublets operation comparisons with SQLite.",
+        "summary_note": "Each comparison uses SQLite of matching durability and the same noise rule as the tables. Counts span the measured variants, sizes and id widths; they are not an overall speed ranking. RAM usage is not measured.",
         "repetitions": lambda count: f"{count} repetition" + ("" if count == 1 else "s"),
         "machine": "unknown machine",
         "local": "a local run",
@@ -110,7 +119,14 @@ def russian_plural(count, one, few, many):
         "faster": "в {ratio}× быстрее",
         "slower": "в {ratio}× медленнее",
         "same": "≈ так же",
-        "provenance": "_{repetitions} после прогрева, медианное время одной операции. SQLite {sqlite}, {machine}, {source}._",
+        "provenance": "{repetitions} после прогрева, медианное время одной операции. SQLite {sqlite}; {software}. {machine}, {source}.",
+        "unrecorded": "версия при измерении не записана",
+        "toolchain_missing": "версия среды разработки не записана",
+        "log_scale": "логарифмическая шкала",
+        "linear_scale": "линейная шкала",
+        "conclusions": "Выводы",
+        "summary": "{language}, {category}: Дуплеты быстрее в {faster}, медленнее в {slower}, примерно равны SQLite в {same} сравнениях операций.",
+        "summary_note": "Каждое сравнение использует SQLite той же надёжности хранения и то же правило учёта шума, что и таблицы. Числа охватывают измеренные варианты, размеры и разрядности id и не являются общим рейтингом скорости. Использование ОЗУ не измеряется.",
         "repetitions": lambda count: f"{count} повтор" + russian_plural(count, "", "а", "ов"),
         "machine": "неизвестная машина",
         "local": "локальный запуск",
@@ -161,6 +177,18 @@ def validate_metadata(report):
         raise ValueError("invalid repetitions")
     if not report["sqlite_version"]:
         raise ValueError("missing SQLite version")
+    if "software" in report:
+        software = report["software"]
+        library = "doublets" if report["language"] == "Rust" else "Platform.Data.Doublets"
+        if (
+            not isinstance(software, dict)
+            or not isinstance(software.get("libraries"), dict)
+            or not isinstance(software["libraries"].get(library), str)
+            or not software["libraries"][library].strip()
+            or not isinstance(software.get("toolchain"), str)
+            or not software["toolchain"].strip()
+        ):
+            raise ValueError("missing or invalid software provenance")
 
 
 def validate_variants(results):
@@ -291,9 +319,21 @@ def table(report, text):
 
 def provenance(report, text):
     run = report.get("run_url")
+    software = report.get("software")
+    if software:
+        libraries = software["libraries"]
+        versions = "; ".join(f"{name} {version}" for name, version in libraries.items())
+        versions += f"; {software['toolchain']}"
+    else:
+        # Older reports remain readable, but today's manifests are not measured provenance.
+        versions = "; ".join(
+            f"{name} {version}" for name, version in library_versions(report["language"]).items()
+        )
+        versions += f" ({text['unrecorded']}); {text['toolchain_missing']}"
     value = text["provenance"].format(
         repetitions=text["repetitions"](report["repetitions"]),
         sqlite=report["sqlite_version"],
+        software=versions,
         machine=report.get("machine") or text["machine"],
         source=text["run"].format(url=run, date=report.get("date")) if run else text["local"],
     )
@@ -304,11 +344,44 @@ def provenance(report, text):
             f"Npgsql.EntityFrameworkCore.PostgreSQL {postgres['provider_version']}; "
             f"EF Core {postgres['ef_core_version']}."
         )
-    return value
+    return f"_{value}_"
+
 
+def chart_name(category, language, bits, scale="log"):
+    suffix = "-linear" if scale == "linear" else ""
+    return f"{category}-{language.lower().replace('#', 'sharp')}-{bits}{suffix}.png"
 
-def chart_name(category, language, bits):
-    return f"{category}-{language.lower().replace('#', 'sharp')}-{bits}.png"
+
+def conclusions(reports, text):
+    """Summarize the measured Doublets comparisons with the table's durability and noise rules."""
+    lines = [f"## {text['conclusions']}", "", text["summary_note"], ""]
+    for category in CATEGORIES:
+        for language in LANGUAGES:
+            counts = {"faster": 0, "slower": 0, "same": 0}
+            for key, data in reports.items():
+                if key[:2] != (category, language):
+                    continue
+                variants = {result["variant"]: result for result in data["results"]}
+                for variant, result in variants.items():
+                    if not variant.startswith("Doublets_"):
+                        continue
+                    reference = variants[baseline(variant)]
+                    for operation, measured in result["operations"].items():
+                        other = reference["operations"][operation]
+                        outcome = (
+                            "same"
+                            if comparison(measured, other, text) == text["same"]
+                            else "faster"
+                            if measured["median_ns"] < other["median_ns"]
+                            else "slower"
+                        )
+                        counts[outcome] += 1
+            if sum(counts.values()):
+                lines.append(
+                    "- "
+                    + text["summary"].format(language=language, category=text["nouns"][category], **counts)
+                )
+    return lines
 
 
 def section(reports, language_code, charts_link):
@@ -336,11 +409,13 @@ def section(reports, language_code, charts_link):
                         "",
                     ]
                 if charts_link is not None:
-                    name = chart_name(category, language, bits)
                     title = text["chart"].format(
                         language=language, bits=bits, category=text["nouns"][category]
                     )
-                    lines += [f"![{title}]({charts_link}/{name})", ""]
+                    for scale in ("linear", "log"):
+                        name = chart_name(category, language, bits, scale)
+                        lines += [f"![{title}, {text[f'{scale}_scale']}]({charts_link}/{name})", ""]
+    lines += conclusions(reports, text)
     return "\n".join(lines).rstrip() + "\n"
 
 
@@ -352,52 +427,77 @@ def replace_section(document, generated):
 
 
 def charts(reports, directory):
-    """One logarithmic bar chart per category, language and bits, with a panel per size."""
+    """Linear and log charts per category, language and bits, with a panel per size."""
     import matplotlib
 
     matplotlib.use("Agg")
-    import matplotlib.pyplot as plt
-
     directory.mkdir(parents=True, exist_ok=True)
     written = []
     for category in CATEGORIES:
-        operations = OPERATIONS[category]
-        labels = [TEXT["en"]["operations"][operation] for operation in operations]
         for language in LANGUAGES:
             for bits in BITS:
                 sizes = sorted(key[3] for key in reports if key[:3] == (category, language, bits))
                 if not sizes:
                     continue
-                figure, axes = plt.subplots(
-                    1, len(sizes), figsize=(6 * len(sizes), 5), squeeze=False, sharey=False
-                )
-                for axis, count in zip(axes[0], sizes, strict=True):
-                    results = reports[(category, language, bits, count)]["results"]
-                    width = 0.8 / len(results)
-                    for index, result in enumerate(results):
-                        axis.bar(
-                            [position + index * width for position in range(len(operations))],
-                            [result["operations"][operation]["median_ns"] for operation in operations],
-                            width,
-                            label=result["variant"].replace("_", " "),
-                            color=plt.get_cmap("tab10")(index % 10),
-                        )
-                    axis.set_yscale("log")
-                    axis.set_title(TEXT["en"]["size"][category].format(size=size(count)))
-                    axis.set_ylabel("median ns per operation (log scale)")
-                    axis.set_xticks([position + 0.4 - width / 2 for position in range(len(operations))])
-                    axis.set_xticklabels(labels, rotation=30, ha="right")
-                    axis.grid(axis="y", which="major", alpha=0.3)
-                axes[0][0].legend(fontsize="small")
-                figure.suptitle(TEXT["en"]["chart"].format(language=language, bits=bits, category=category))
-                figure.tight_layout()
-                path = directory / chart_name(category, language, bits)
-                figure.savefig(path, dpi=80)
-                plt.close(figure)
-                written.append(path)
+                for scale in ("linear", "log"):
+                    path = directory / chart_name(category, language, bits, scale)
+                    draw_chart(reports, category, language, bits, sizes, scale, path)
+                    written.append(path)
     return written
 
 
+def draw_chart(reports, category, language, bits, sizes, scale, path):
+    """Draw each size in its own panel, with a visibility floor only on the linear scale."""
+    import matplotlib.pyplot as plt
+
+    operations = OPERATIONS[category]
+    labels = [TEXT["en"]["operations"][operation] for operation in operations]
+    figure, axes = plt.subplots(1, len(sizes), figsize=(6 * len(sizes), 6), squeeze=False, sharey=False)
+    try:
+        for axis, count in zip(axes[0], sizes, strict=True):
+            results = reports[(category, language, bits, count)]["results"]
+            width = 0.8 / len(results)
+            maximum = max(
+                result["operations"][operation]["median_ns"] for result in results for operation in operations
+            )
+            minimum = maximum * 0.005 if scale == "linear" else 0
+            for index, result in enumerate(results):
+                axis.bar(
+                    [position + index * width for position in range(len(operations))],
+                    [max(result["operations"][operation]["median_ns"], minimum) for operation in operations],
+                    width,
+                    label=result["variant"].replace("_", " "),
+                    color=plt.get_cmap("tab20" if len(results) > 10 else "tab10")(index),
+                )
+            axis.set_yscale(scale)
+            if scale == "linear":
+                axis.set_ylim(bottom=0)
+            axis.set_title(TEXT["en"]["size"][category].format(size=size(count)))
+            axis.set_ylabel(f"median ns per operation ({scale} scale)")
+            axis.set_xticks([position + 0.4 - width / 2 for position in range(len(operations))])
+            axis.set_xticklabels(labels, rotation=30, ha="right")
+            axis.grid(axis="y", which="major", alpha=0.3)
+        handles, names = axes[0][0].get_legend_handles_labels()
+        footer = 0.04 if scale == "linear" else 0
+        figure.legend(
+            handles, names, loc="lower center", ncol=3, fontsize="small", bbox_to_anchor=(0.5, footer)
+        )
+        figure.suptitle(TEXT["en"]["chart"].format(language=language, bits=bits, category=category))
+        if scale == "linear":
+            figure.text(
+                0.5,
+                0.01,
+                "Bars below 0.5% of each panel's maximum are raised for visibility.\nSee tables for exact times.",
+                ha="center",
+                fontsize="small",
+            )
+        legend_height = 0.05 + math.ceil(len(names) / 3) * 0.035
+        figure.tight_layout(rect=(0, legend_height + footer, 1, 1))
+        figure.savefig(path, dpi=80)
+    finally:
+        plt.close(figure)
+
+
 def main(arguments=None):
     parser = argparse.ArgumentParser(description=__doc__)
     parser.add_argument("results", type=Path, help="directory with the JSON reports")
diff --git a/scripts/ci_checks.py b/scripts/ci_checks.py
index 5af8607..5a3eeac 100644
--- a/scripts/ci_checks.py
+++ b/scripts/ci_checks.py
@@ -10,18 +10,17 @@
 DOCUMENT_SECTIONS = {
     "README.md": (
         "## Benchmarks",
-        "## Original comparison",
+        "## Conclusions",
         "<!--BENCHMARK_RESULTS_START-->",
         "<!--BENCHMARK_RESULTS_END-->",
     ),
     "README.ru.md": (
         "## Тесты производительности",
-        "## Исходное сравнение",
+        "## Выводы",
         "<!--BENCHMARK_RESULTS_START-->",
         "<!--BENCHMARK_RESULTS_END-->",
     ),
     "CONTRIBUTING.md": ("## Local checks", "## Dependencies and diagnostics"),
-    "cpp/README.md": ("# C++ placeholder",),
 }
 
 
@@ -70,8 +69,6 @@ def main():
     for path in sorted((ROOT / ".github/workflows").glob("*.yml")):
         workflow = load_workflow(path.read_text(encoding="utf-8"))
         errors += [f"{path.name}: {error}" for error in workflow_errors(workflow)]
-    if (ROOT / ".github/workflows/cpp.yml").exists() and not (ROOT / "cpp/CMakeLists.txt").exists():
-        errors.append("C++ build workflow has no CMake project")
     for directory in ("scripts", "rust/src", "rust/tests", "csharp", "experiments"):
         for path in (ROOT / directory).rglob("*"):
             if path.suffix not in (".py", ".rs", ".cs", ".sh") or any(
diff --git a/scripts/test_benchmark_provenance.py b/scripts/test_benchmark_provenance.py
new file mode 100644
index 0000000..c4ebce0
--- /dev/null
+++ b/scripts/test_benchmark_provenance.py
@@ -0,0 +1,49 @@
+"""Snapshot measured software versions rather than attributing today's versions to old runs."""
+
+import json
+import tempfile
+import unittest
+from pathlib import Path
+
+import benchmark_provenance as provenance
+
+
+class SoftwareTests(unittest.TestCase):
+    def setUp(self):
+        temporary = tempfile.TemporaryDirectory()
+        self.addCleanup(temporary.cleanup)
+        self.root = Path(temporary.name)
+        (self.root / "rust").mkdir()
+        (self.root / "rust/Cargo.lock").write_text(
+            'version = 4\n[[package]]\nname = "doublets"\nversion = "0.5.0"\n'
+        )
+        (self.root / "csharp/SQLiteVSDoublets").mkdir(parents=True)
+        (self.root / "csharp/SQLiteVSDoublets/SQLiteVSDoublets.csproj").write_text(
+            '<Project><ItemGroup><PackageReference Version="0.18.1" '
+            'Include="Platform.Data.Doublets" /></ItemGroup></Project>'
+        )
+
+    def test_reads_resolved_rust_and_csharp_library_versions(self):
+        self.assertEqual(provenance.library_versions("Rust", self.root), {"doublets": "0.5.0"})
+        self.assertEqual(provenance.library_versions("C#", self.root), {"Platform.Data.Doublets": "0.18.1"})
+
+    def test_recording_preserves_measurements_and_captures_toolchain(self):
+        path = self.root / "report.json"
+        data = {"language": "Rust", "results": [{"median_ns": 1234}]}
+        path.write_text(json.dumps(data))
+        provenance.record(path, "rustc 1.98.1", self.root)
+        recorded = json.loads(path.read_text())
+        self.assertEqual(recorded["results"], data["results"])
+        self.assertEqual(
+            recorded["software"],
+            {"libraries": {"doublets": "0.5.0"}, "toolchain": "rustc 1.98.1"},
+        )
+
+    def test_missing_manifests_cannot_silently_record_unknown_versions(self):
+        (self.root / "rust/Cargo.lock").unlink()
+        with self.assertRaises((ValueError, FileNotFoundError)):
+            provenance.library_versions("Rust", self.root)
+
+
+if __name__ == "__main__":
+    unittest.main()
diff --git a/scripts/test_benchmark_report.py b/scripts/test_benchmark_report.py
index a82eb11..2bc9d67 100644
--- a/scripts/test_benchmark_report.py
+++ b/scripts/test_benchmark_report.py
@@ -8,6 +8,7 @@
 import tempfile
 import unittest
 from pathlib import Path
+from unittest.mock import patch
 
 import benchmark_report as report
 
@@ -120,6 +121,23 @@ def test_table_compares_every_operation_with_the_baseline(self):
 
 
 class SectionTests(unittest.TestCase):
+    def test_recorded_software_versions_are_inside_the_provenance_line(self):
+        data = sample(software={"libraries": {"doublets": "0.4.0"}, "toolchain": "rustc 1.97.0"})
+        text = report.provenance(data, report.TEXT["en"])
+        self.assertIn("doublets 0.4.0", text)
+        self.assertIn("rustc 1.97.0", text)
+        self.assertEqual(text.count("_"), 2)
+
+    def test_conclusions_are_generated_from_the_current_measurements(self):
+        text = report.section({("links", "Rust", 64, 100_000): sample()}, "en", None)
+        self.assertIn("## Conclusions", text)
+        self.assertIn("8 faster, 0 slower, 0 approximately equal", text)
+        data = sample()
+        for measured in data["results"][-1]["operations"].values():
+            measured.update(measurement(10_000))
+        text = report.section({("links", "Rust", 64, 100_000): data}, "en", None)
+        self.assertIn("0 faster, 8 slower, 0 approximately equal", text)
+
     def test_hierarchy_is_category_language_bits_size(self):
         reports = {
             ("links", "Rust", 64, 1_000_000): sample(size=1_000_000),
@@ -149,6 +167,7 @@ def test_hierarchy_is_category_language_bits_size(self):
                 "#### 32 bit address/id space benchmarks",
                 "##### 100,000 blog posts",
                 "#### 64 bit address/id space benchmarks",
+                "## Conclusions",
             ],
         )
 
@@ -160,10 +179,10 @@ def test_missing_groups_and_provenance(self):
         text = report.section(reports, "en", None)
         self.assertEqual(text.count("_No results yet._"), 7)
         self.assertIn(
-            f"_3 repetitions after a warm-up, median time per operation. SQLite 3.53.2, ubuntu-24.04, "
             f"[GitHub Actions run]({url}) on 2026-10-04._",
             text,
         )
+        self.assertIn("toolchain version not recorded", text)
         self.assertNotIn("![", text)
 
     def test_charts_are_linked_per_category_language_and_bits(self):
@@ -224,6 +243,36 @@ def test_incomplete_and_nonfinite_measurements_are_rejected(self):
                 with self.assertRaises(ValueError):
                     report.load(self.root)
 
+    def test_recorded_provenance_requires_the_measured_library_and_toolchain(self):
+        for software in (
+            {},
+            {"libraries": {"doublets": "0.5.0"}, "toolchain": ""},
+            {"libraries": {"other": "0.5.0"}, "toolchain": "rustc 1.99.0"},
+        ):
+            with self.subTest(software=software), self.assertRaisesRegex(ValueError, "software provenance"):
+                report.validate(sample(software=software), "report.json")
+
+    @unittest.skipUnless(importlib.util.find_spec("matplotlib"), "matplotlib is not installed")
+    def test_linear_charts_keep_small_bars_visible_and_log_charts_keep_actual_values(self):
+        data = sample()
+        data["results"][-1]["operations"]["create"] = measurement(1)
+        captured = {}
+
+        def capture(figure, path, **kwargs):
+            captured[path.name] = figure
+
+        with patch("matplotlib.figure.Figure.savefig", autospec=True, side_effect=capture):
+            paths = report.charts({("links", "Rust", 64, 100_000): data}, self.root)
+        self.assertEqual({path.name for path in paths}, {"links-rust-64.png", "links-rust-64-linear.png"})
+        for name, scale, minimum in (
+            ("links-rust-64.png", "log", 1),
+            ("links-rust-64-linear.png", "linear", 10),
+        ):
+            axis = captured[name].axes[0]
+            self.assertEqual(axis.get_yscale(), scale)
+            self.assertEqual(min(bar.get_height() for bar in axis.patches), minimum)
+        self.assertEqual(data["results"][-1]["operations"]["create"]["median_ns"], 1)
+
     @unittest.skipUnless(importlib.util.find_spec("matplotlib"), "matplotlib is not installed")
     def test_main_updates_readmes_and_writes_charts(self):
         results_directory = self.root / "results"
@@ -247,12 +296,34 @@ def test_main_updates_readmes_and_writes_charts(self):
         )
         self.assertEqual(
             sorted(path.name for path in (self.root / "docs").iterdir()),
-            ["links-rust-64.png", "objects-csharp-32.png"],
+            [
+                "links-rust-64-linear.png",
+                "links-rust-64.png",
+                "objects-csharp-32-linear.png",
+                "objects-csharp-32.png",
+            ],
         )
         english = (self.root / "README.md").read_text(encoding="utf-8")
-        self.assertIn("![Rust doublets vs SQLite, 64 bit, links](docs/links-rust-64.png)", english)
+        self.assertIn("![Rust doublets vs SQLite, 64 bit, links, log scale](docs/links-rust-64.png)", english)
+        self.assertIn(
+            "![Rust doublets vs SQLite, 64 bit, links, linear scale](docs/links-rust-64-linear.png)", english
+        )
         self.assertIn("Дуплеты на C# против SQLite", (self.root / "README.ru.md").read_text(encoding="utf-8"))
 
 
+class RepositoryReportsTests(unittest.TestCase):
+    def test_both_committed_readmes_match_the_generator_without_a_stale_tail(self):
+        root = Path(__file__).resolve().parents[1]
+        reports = report.load(root / "docs/benchmarks/results")
+        for name, language in (("README.md", "en"), ("README.ru.md", "ru")):
+            with self.subTest(name=name):
+                document = (root / name).read_text(encoding="utf-8")
+                self.assertTrue(document.rstrip().endswith(report.END_MARKER))
+                self.assertEqual(
+                    document,
+                    report.replace_section(document, report.section(reports, language, "docs/benchmarks")),
+                )
+
+
 if __name__ == "__main__":
     unittest.main()
diff --git a/scripts/test_postgresql_report.py b/scripts/test_postgresql_report.py
index 1c76863..ef3e1ac 100644
--- a/scripts/test_postgresql_report.py
+++ b/scripts/test_postgresql_report.py
@@ -7,6 +7,21 @@
 
 
 class PostgreSQLReportTests(unittest.TestCase):
+    def test_all_postgresql_versions_stay_inside_the_italic_provenance(self):
+        data = sample(
+            "objects",
+            "C#",
+            32,
+            1000,
+            postgresql={"server_version": "18.3", "provider_version": "10.0.3", "ef_core_version": "10.0.4"},
+        )
+        for language in ("en", "ru"):
+            text = report.provenance(data, report.TEXT[language])
+            self.assertIn("PostgreSQL 18.3", text)
+            self.assertIn("EF Core 10.0.4", text)
+            self.assertEqual(text.count("_"), 2)
+            self.assertTrue(text.endswith("._"))
+
     def test_server_provider_has_no_embedded_durability_baseline(self):
         self.assertIsNone(report.baseline("PostgreSQL_EFCore"))