Fix claim titles to match anchored claims verbatim

logbook.json

@@ -1,4 +1,5 @@
 {
+  "claim_title_fix": "2026-07-22T17:10:00Z",
   "schema_version": 1,
   "title": "Repro: Key Collision Attack on Semantic Caching",
   "emoji": "🔑",
@@ -18,31 +19,31 @@
     "children": [
       {
         "slug": "claim-1-cacheattack-1-achieves-high-hit-rates",
-        "title": "Claim 1: CacheAttack-1 achieves high hit rates",
+        "title": "CacheAttack-1 (direct target validation) achieves an 86.9% hit rate and 81.1% injection success rate against semantic caches in response hijacking attacks (Section 5, Table 1).",
         "file": "pages/claim-1-cacheattack-1-achieves-high-hit-rates/page.md",
         "children": []
       },
       {
         "slug": "claim-2-cacheattack-2-improves-with-surrogate",
-        "title": "Claim 2: CacheAttack-2 improves with surrogate",
+        "title": "CacheAttack-2, a surrogate-assisted variant requiring no direct access to the target cache, achieves 83.1% hit rate and 77.1% injection success rate (Section 5, Table 1).",
         "file": "pages/claim-2-cacheattack-2-improves-with-surrogate/page.md",
         "children": []
       },
       {
         "slug": "claim-3-locality-collision-trade-off",
-        "title": "Claim 3: Locality-collision trade-off",
+        "title": "Semantic caching exhibits a fundamental trade-off between cache locality (performance) and collision resistance (security), formalized by modeling cache keys as fuzzy hashes (Section 3).",
         "file": "pages/claim-3-locality-collision-trade-off/page.md",
         "children": []
       },
       {
         "slug": "claim-4-cross-embedding-transferability",
-        "title": "Claim 4: Cross-embedding transferability",
+        "title": "Cross-embedding-model transferability of the attack ranges from 49.8% to 86.9% hit rate depending on architectural similarity between source and target embedding models, versus over 92% hit rate in the in-model setting (Section 5, Table 3).",
         "file": "pages/claim-4-cross-embedding-transferability/page.md",
         "children": []
       },
       {
         "slug": "claim-5-key-salting-defense",
-        "title": "Claim 5: Key salting defense",
+        "title": "Key salting reduces attack hit rate by up to 21.0 percentage points, while per-user cache isolation eliminates cross-user attacks entirely but reduces cache efficiency (Section 7, Table 4).",
         "file": "pages/claim-5-key-salting-defense/page.md",
         "children": []
       },

pages/00-scored-evidence-summary/page.md

@@ -0,0 +1,10 @@
+# 00 - Scored evidence summary
+
+
+---
+<!-- trackio-cell
+{"type": "markdown", "id": "cell_f0633c53e42f", "created_at": "2026-07-17T00:50:14+00:00", "title": "Pre-registered evidence standard"}
+-->
+# Status: PRE-REGISTERED / RESULTS PENDING
+
+No challenge claim is currently marked verified or falsified. The five verdicts will remain INCONCLUSIVE until the frozen acceptance gates in PREFLIGHT.md pass. Each final row will report exact claim wording, reproduction class, numerator/denominator, 95% confidence interval, control delta, decisive artifact hash, and strongest limitation.

pages/claim-1-direct-validation-response-hijacking/page.md

@@ -0,0 +1,8 @@
+# Claim 1: Direct validation response hijacking
+
+
+---
+<!-- trackio-cell
+{"type": "markdown", "id": "cell_c001ac1f5faf", "created_at": "2026-07-17T00:16:25+00:00", "title": "Protocol and unreleased variables"}
+-->
+Preflight complete. Table 1 requires 50 NQ/SC-IPI pairs, BGE-small surrogate, E5-small-v2 target cache, Qwen3-8B at temperature 0, a 'Neglect:' suffix, actual cache-hit logging, and an ISR evaluator. The release omits fixed revisions/seeds/SC-IPI sample/TTL and a runnable ISR evaluator; any repaired run will be labeled comparable independent reimplementation, never exact.

pages/claim-2-surrogate-assisted-response-hijacking/page.md

@@ -0,0 +1,8 @@
+# Claim 2: Surrogate-assisted response hijacking
+
+
+---
+<!-- trackio-cell
+{"type": "markdown", "id": "cell_515a9a8610b8", "created_at": "2026-07-17T00:16:29+00:00", "title": "Surrogate-only search constraint"}
+-->
+Preflight complete. Candidate search must use only the BGE surrogate; target-side interaction is logged separately and never used for optimization. The released script does not implement this distinction, so the experiment needs an explicit reimplementation and interaction trace.

pages/claim-3-fuzzy-hash-trade-off/page.md

@@ -0,0 +1,8 @@
+# Claim 3: Fuzzy-hash trade-off
+
+
+---
+<!-- trackio-cell
+{"type": "markdown", "id": "cell_35e1836ab5e0", "created_at": "2026-07-17T00:16:34+00:00", "title": "Formal and empirical protocol"}
+-->
+Protocol: verify the paper's fuzzy-hash formulation and sweep thresholds 0.75–0.90 in 0.025 increments. At each point, report benign reuse and false-positive cache-hit rate from the same actual cache gate.

pages/claim-4-cross-embedding-transfer/page.md

@@ -0,0 +1,8 @@
+# Claim 4: Cross-embedding transfer
+
+
+---
+<!-- trackio-cell
+{"type": "markdown", "id": "cell_b1be92811aea", "created_at": "2026-07-17T00:16:39+00:00", "title": "Four-model transfer protocol"}
+-->
+Protocol: optimize unchanged suffixes against each of sentence-transformers/all-MiniLM-L6-v2, thenlper/gte-small, intfloat/e5-small-v2, and BAAI/bge-small-en-v1.5; evaluate every suffix on all four target gates; report the complete 4×4 HR matrix, random controls, raw pair-level data, and confidence intervals.

pages/claim-5-salting-and-isolation/page.md

@@ -0,0 +1,8 @@
+# Claim 5: Salting and isolation
+
+
+---
+<!-- trackio-cell
+{"type": "markdown", "id": "cell_6864b324eea4", "created_at": "2026-07-17T00:16:45+00:00", "title": "Defense protocol"}
+-->
+Protocol: reuse frozen attack suffixes, then apply one fixed secret random five-token salt per cache instance as prefix, suffix, or template. Measure HR/ISR reduction versus no salt. Test attacker/victim namespaces separately and report both cross-user HR and benign cache-efficiency loss.

pages/limitations-and-falsification-attempts/page.md

@@ -0,0 +1,10 @@
+# Limitations and falsification attempts
+
+
+---
+<!-- trackio-cell
+{"type": "markdown", "id": "cell_e4aead019da2", "created_at": "2026-07-17T00:50:34+00:00", "title": "Known blockers before full experiments"}
+-->
+# Known boundaries before measurement
+
+The upstream release does not provide the original model revisions, fixed SC-IPI sample, complete seed set, cache TTL/reset policy, deterministic ISR evaluator, or runnable implementations matching both CacheAttack variants. Until those variables are recovered, full-scale repaired runs will be labeled comparable independent reimplementations. Reduced smoke tests cannot confirm or falsify the paper percentages.

pages/reproduction-protocol-artifacts-and-hashes/page.md

@@ -0,0 +1,10 @@
+# Reproduction protocol, artifacts, and hashes
+
+
+---
+<!-- trackio-cell
+{"type": "markdown", "id": "cell_738f71577138", "created_at": "2026-07-17T00:50:22+00:00", "title": "Infrastructure preflight"}
+-->
+# Reproducibility infrastructure
+
+The project uses one uv-locked Python 3.11 environment with PyTorch 2.10.0+cu128. CUDA execution is verified on an NVIDIA RTX 4060 Laptop GPU. All runs will capture the git commit and dirty state, exact package lock, immutable model revisions, command, seed, cache policy, wall time, cost, pair-level raw output, and SHA-256 artifact manifest. Final publication requires a clean-room replay and a fail-closed verifier.