dsh-image-integrity 0.1.0 → 0.1.2

package/CHANGELOG.md

@@ -6,6 +6,39 @@ adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
 ## [Unreleased]
 
+### Added
+
+- `bench/run-images.mjs` and the `bench:images` script: score the plugin against
+  the frozen 100-pair synthetic image-reuse set in `dsh-research-bench`
+  (`data/images`) and write `bench/results-images.json`. The adapter predicts
+  `manipulated = matches.length > 0` and takes the first match's kind, using the
+  plugin defaults; the result has no timestamps and is byte-identical on rerun.
+- First reproducible benchmark numbers, replacing the "not yet claimed" status:
+  precision 1.0000, recall 1.0000, F1 1.0000 and false-positive rate 0.0000 on
+  50 manipulated + 50 clean pairs (25 duplicate, 25 flip, 25 unrelated, 25
+  hard-negative). Per-category accuracy and duplicate/flip subtype recall are
+  all 1.0000.
+
+### Changed
+
+- README evaluation sections now document the benchmark, its scope (grid-aligned
+  16×16 positive blocks; no shift/JPEG/scale robustness claims) and how to
+  reproduce it.
+
+## [0.1.2] - 2026-11-08
+
+### Added
+
+- 100-pair synthetic image-reuse benchmark and `pnpm run bench:images`; README reports the measured precision/recall/FPR and the grid-aligned, uncompressed-fixture limits.
+
+## [0.1.1] - 2026-11-08
+
+### Changed
+
+- Release-engineering refresh: CHANGELOG accuracy, package metadata and authors; no runtime changes.
+
+## [0.1.0] - 2026-11-08
+
 ### Fixed
 
 - Enforce `maxPixels` from the IHDR **before** inflating IDAT, and pass the
@@ -28,8 +61,6 @@ adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
   the O(n²) comparison loop.
 - Test count is now 22; README/CHANGELOG list the new failure branches.
 
-## [0.1.0] - 2026-11-08
-
 First release: block-level image reuse detection.
 
 ### Added

package/README-zh.md

@@ -82,9 +82,20 @@ dsh --profile <name> --dump-config | grep dsh-image-integrity
 
 ## 评测
 
-**22 个测试**：PNG 往返、非 PNG / 16 位 / 截断文件拒绝、解压前依据 IHDR 拒绝超像素上限（配置值与默认值）、汉明距离与 NCC 基础函数、重复块检测、水平镜像检测、平坦图与渐变图的误报控制、小图的有效块大小、成对匹配、尺寸不一致时拒绝、runtime（合法 REP 记录、每次比较生成两份快照、路径遍历/JPEG/字节与像素超限拒绝、非法阈值拒绝）、tools（schema、check→report、compare、格式错误的报告、失败分支）、包契约。
+**23 个测试**：PNG 往返、非 PNG / 16 位 / 截断文件拒绝、解压前依据 IHDR 拒绝超像素上限（配置值与默认值）、汉明距离与 NCC 基础函数、重复块检测、水平镜像检测、平坦图与渐变图的误报控制、小图的有效块大小、成对匹配、尺寸不一致时拒绝、runtime（合法 REP 记录、每次比较生成两份快照、路径遍历/JPEG/字节与像素超限拒绝、非法阈值拒绝）、tools（schema、check→report、compare、格式错误的报告、失败分支）、挂载契约（disabled 不挂载任何工具，enabled 注册三个唯一工具）、包契约。
 
-**基准测试状态：尚未声称。** 计划中的 `dsh-research-bench` 数据集是 100 对合成 PNG（重复 / 镜像 / 无关 / 困难负样本）。在该数据集冻结并评分之前，上述单元 fixture 是唯一的证据，README 也如实说明这一点。
+**基准测试 —— 100 对合成图像。** `pnpm run bench:images` 会在配套的 [`dsh-research-bench`](https://github.com/PerryLink/dsh-research-bench) 仓库（`data/images`，数据集版本 0.1.0）上评分：25 对 duplicate + 25 对 flip + 25 对 unrelated + 25 对 hard-negative，均为 64×64 8 位灰度 PNG，固定种子，正样本复制的是与网格对齐的 16×16 块。适配器针对构建后的 `lib/` 运行，写入 [`bench/results-images.json`](bench/results-images.json)，不含时间戳（同一数据复跑字节一致）；默认读取同级 `../dsh-research-bench` 检出。其他位置可用 `BENCH_DIR`，或用 `node bench/run-images.mjs --data <dir>` 指定。
+
+| 指标（50 对 manipulated / 50 对 clean） | 值 |
+|---|---|
+| Manipulated precision | 1.0000 |
+| Manipulated recall | 1.0000 |
+| Manipulated F1 | 1.0000 |
+| 误报率（unrelated + hard-negative） | 0.0000 |
+| Subtype recall —— duplicate | 1.0000 |
+| Subtype recall —— flip（水平 13 / 垂直 12） | 1.0000 |
+
+混淆计数：TP 50、FP 0、TN 50、FN 0；没有误分类的图对。四类逐类准确率均为 1.0000（各 25/25）。范围：正样本复制的是**与网格对齐**的 16×16 块（即检测器自身的处理单元）；亚块偏移、JPEG 重压缩、缩放、旋转不在本集合内，仍是已记录的限制。这些是冻结合成 fixture 上的数字，不代表真实论文图中复用现象的比例。数据集与生成器维护在配套仓库中，不随本包发布。
 
 ## 已知限制（v0.1）
 

package/README.md

@@ -85,9 +85,20 @@ dsh --profile <name> --dump-config | grep dsh-image-integrity
 
 ## Evaluation
 
-**22 tests**: PNG round-trip, non-PNG / 16-bit / truncated rejection, pixel-cap rejection from the IHDR before inflation (configured and default), Hamming and NCC primitives, duplicate block detection, horizontal mirror detection, flat-image and gradient false-positive controls, small-image effective block size, pair matching, dimension-mismatch refusal, runtime (valid REP record, two snapshots per comparison, traversal/JPEG/byte-and-pixel oversize rejection, invalid threshold rejection), tools (schemas, check→report, compare, malformed report, failure branch), package contract.
+**23 tests**: PNG round-trip, non-PNG / 16-bit / truncated rejection, pixel-cap rejection from the IHDR before inflation (configured and default), Hamming and NCC primitives, duplicate block detection, horizontal mirror detection, flat-image and gradient false-positive controls, small-image effective block size, pair matching, dimension-mismatch refusal, runtime (valid REP record, two snapshots per comparison, traversal/JPEG/byte-and-pixel oversize rejection, invalid threshold rejection), tools (schemas, check→report, compare, malformed report, failure branch), mount contract (disabled mounts nothing, enabled registers three unique tools), package contract.
 
-**Benchmark status: not yet claimed.** The planned `dsh-research-bench` set is 100 synthetic PNG pairs (duplicate / mirrored / unrelated / hard negative). Until that set is frozen and scored, the unit fixtures above are the only evidence, and the README says so.
+**Benchmark — 100 synthetic pairs.** `pnpm run bench:images` scores the plugin against the frozen set (dataset version 0.1.0) in the companion [`dsh-research-bench`](https://github.com/PerryLink/dsh-research-bench) repository (`data/images`): 25 duplicate + 25 flip + 25 unrelated + 25 hard-negative pairs of 64×64 8-bit grey PNGs, fixed seed, with positives copying grid-aligned 16×16 blocks. The adapter runs against the built `lib/`, writes [`bench/results-images.json`](bench/results-images.json) with no timestamps (rerunning on the same data is byte-identical), and reads a sibling `../dsh-research-bench` checkout by default. Set `BENCH_DIR` or pass `node bench/run-images.mjs --data <dir>` for another location.
+
+| Metric (50 manipulated / 50 clean) | Value |
+|---|---|
+| Manipulated precision | 1.0000 |
+| Manipulated recall | 1.0000 |
+| Manipulated F1 | 1.0000 |
+| False-positive rate (unrelated + hard-negative) | 0.0000 |
+| Subtype recall — duplicate | 1.0000 |
+| Subtype recall — flip (horizontal 13 / vertical 12) | 1.0000 |
+
+Confusion: TP 50, FP 0, TN 50, FN 0; no pair was misclassified. Per-category accuracy is 1.0000 in all four categories (25/25 each). Scope: the positives copy **grid-aligned** 16×16 blocks, i.e. the detector's own unit; sub-grid shifts, JPEG recompression, scaling and rotation are outside this set and remain documented limitations. These are synthetic numbers on a frozen fixture, not a claim about how often real figures reuse panels. The dataset and generator live in the companion repository and are not published with this package.
 
 ## Known limitations (v0.1)
 

package/SECURITY.md

@@ -0,0 +1,19 @@
+# Security policy
+
+## Supported versions
+
+The latest published `0.1.x` release of this package receives security fixes.
+
+## Reporting a vulnerability
+
+Please use GitHub's private vulnerability reporting (Security → Advisories) on
+the repository, or open a minimal issue asking for a private channel if
+advisories are unavailable. Do not include credentials, personal data or
+working exploits in a public issue.
+
+## Scope notes
+
+- This package performs local file analysis only unless its README documents a
+  network provider; when it does, only the documented hosts are contacted.
+- Inputs are treated as untrusted data: document text is never executed.
+- No secrets are read or written by the package.

package/THIRD_PARTY_NOTICES.md

@@ -0,0 +1,10 @@
+# Third-party notices
+
+This package contains **no third-party runtime code**: `dependencies` is empty.
+
+Development-only tooling (TypeScript, Vitest, `@deepseek-ai/*` type packages and
+the shared `@perrylink/dsh-research-core` peer, when present) is not bundled and
+is not distributed in the published tarball.
+
+If that ever changes, the affected license texts must be added here before a
+release.

package/lib/version.d.ts

@@ -3,5 +3,5 @@
  * @module dsh-image-integrity/version
  */
 /** The package version. */
-export declare const VERSION = "0.1.0";
+export declare const VERSION = "0.1.2";
 //# sourceMappingURL=version.d.ts.map

package/lib/version.js

@@ -3,5 +3,5 @@
  * @module dsh-image-integrity/version
  */
 /** The package version. */
-export const VERSION = '0.1.0';
+export const VERSION = '0.1.2';
 //# sourceMappingURL=version.js.map

package/package.json

@@ -1,8 +1,8 @@
 {
   "name": "dsh-image-integrity",
-  "version": "0.1.0",
+  "version": "0.1.2",
   "description": "Image integrity checks for DeepSeek Harness: zero-dependency 8-bit PNG decoding (grey/RGB/RGBA) with mean-threshold block fingerprints and NCC confirmation to find duplicated or mirrored regions inside one figure or shared between two figures, plus REP evidence and Markdown/SARIF reports. Finds reused pixels, not manipulation intent; JPEG is out of scope in v0.1.",
-  "author": "dsh-statcheck contributors",
+  "author": "PerryLink contributors",
   "repository": {
     "type": "git",
     "url": "git+https://github.com/PerryLink/dsh-image-integrity.git"
@@ -29,7 +29,9 @@
     "CHANGELOG.md",
     "README.md",
     "README-zh.md",
-    "LICENSE"
+    "LICENSE",
+    "SECURITY.md",
+    "THIRD_PARTY_NOTICES.md"
   ],
   "sideEffects": false,
   "dsh": {
@@ -75,7 +77,8 @@
     "typecheck": "tsc -p tsconfig.json --noEmit && tsc -p tsconfig.test.json",
     "test": "vitest run",
     "test:watch": "vitest",
-    "verify:package": "node scripts/verify-package.mjs"
+    "verify:package": "node scripts/verify-package.mjs",
+    "bench:images": "node bench/run-images.mjs"
   },
   "license": "Apache-2.0"
 }

package/src/version.ts

@@ -4,4 +4,4 @@
  */
 
 /** The package version. */
-export const VERSION = '0.1.0'
+export const VERSION = '0.1.2'