Tool reference · Evaluation
memory_eval_contradictions
Checks that the contradiction detector marks a newer conflicting record as superseding an old one and leaves unrelated records alone.
local stdio server read-only
Caution. mode defaults to fast, which returns { status: "not_implemented" } immediately. Pass mode "balanced" or "deep" to actually run it.
When to use
- You enabled contradiction detection and want to confirm it behaves.
- After changing the contradiction detector or its thresholds.
- You want a two-case pass/fail check without a configured LLM provider.
Parameters
| Name | Type | Default | Description |
|---|---|---|---|
mode | "fast" | "balanced" | "deep" | "fast" | — |
fixture_path | string | — | — |
Example
Arguments
{
"mode": "balanced"
}Result shape
{
"scenarios_total": 2,
"scenarios_passed": 2,
"recall_at_5": 1,
"recall_at_10": 1,
"latency_ms": 0,
"mode": "balanced",
"llm_calls_during_eval": 0,
"network_calls_during_eval": 0,
"details": [
{
"label": "supersedes",
"expected": "superseded",
"decisions": [
"superseded"
],
"passed": true
},
{
"label": "unrelated",
"expected": "rejected",
"decisions": [
"rejected"
],
"passed": true
}
]
}The fixture uses built-in stand-in LLM answers, so no real LLM is contacted.
Values are illustrative; the keys follow the server's handler. MCP clients receive the result as JSON text content.
Server description
The description the server sends to your agent in tools/list, captured from the v14.7.0 source:
v11.0 Phase 8: runs contradiction_detector against a labelled fixture. Requires balanced/deep mode (LLM). Returns {status: 'not_implemented', ...} if module is unavailable.