artale
/
fable-agent
Template
1
1
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

docs: map adaptive tap pair red-team research

This commit is contained in:
artale 2026-07-02 01:10:36 +02:00
parent 7e6b624378
commit 85a8192810
1 changed files with 48 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

48
docs/receipts/adaptive-tap-pair-research-map-20260702.json Normal file
View File

@ -0,0 +1,48 @@
{
"schema": "fable.research_map.receipt.v1",
"createdAt": "2026-07-02T00:00:00.000Z",
"taskLabel": "adaptive-tap-pair-red-team-map",
"task": "Map arXiv:2606.18193v1 adaptive TAP-PAIR attacks into bounded Fable red-team reporting constraints",
"constraints": [
"arXiv:2606.18193v1 and adaptive TAP-PAIR attack results are external-paper advisory only until locally reproduced",
"do not report paper headline 94%/96% figures as Fable-verified attack success rate",
"represent 94%/96% only as paper-reported robustness/resistance or attempted-versus-confirmed distinction unless a local eval receipt proves otherwise",
"future Fable red-team reports must include denominators, attempted counts, confirmed counts, and confirmed-versus-attempted rates",
"do not make safety or robustness overclaims without a local eval receipt",
"static h4rm3l low-ASR statements are not local authority unless backed by a committed/local receipt",
"no external paper code execution, external repo vendoring, dashboards, new packages, or release changes are included in this map"
],
"existingPrimitives": [
"research-map-receipt",
"red-team-harness-receipt",
"jailbreak-assessment",
"falsification-receipt"
],
"recommendedPlan": [
{
"name": "Bound adaptive TAP-PAIR claims",
"goal": "Capture the paper as advisory red-team context and require future Fable reports to separate attempted attacks, confirmed bypasses, denominators, and local evidence.",
"acceptanceCriteria": [
"receipt records arXiv:2606.18193v1 as advisory only",
"94%/96% figures are not described as Fable-verified ASR",
"future local claims require denominators and confirmed versus attempted counts",
"no external execution or release action is implied"
],
"verification": [
"cat docs/receipts/adaptive-tap-pair-research-map-20260702.json",
"grep -R \"2606.18193\\|TAP-PAIR\\|Fable-verified attack success rate\" docs/receipts"
]
}
],
"skipped": [
"release changes",
"external paper code execution",
"external repo vendoring",
"dashboard changes",
"new packages",
"local ASR claim without eval receipt"
],
"decision": "ready",
"reasons": [],
"deployAttempted": false
}