_FINDING_PUBLIC PROMPT PROBE — 4ce0212286e0
Evidence: 3004 chars; quote: 2004 chars (caps: 2000 / 1000).

{
 "prompt_type": "re-ground",
 "full_evidence_in_prompt": false,
 "visible_evidence_chars": 1995,
 "ends_mid_word": false,
 "truncation_marker_present": true,
 "full_sha256": "fbad05d2e2435a37ee5b74acfca72ac7d1ae470e6de48f2a3803b7034756ff50"
}
{
 "prompt_type": "cross-family confirmation",
 "full_evidence_in_prompt": false,
 "visible_evidence_chars": 1995,
 "ends_mid_word": false,
 "truncation_marker_present": true,
 "full_sha256": "fbad05d2e2435a37ee5b74acfca72ac7d1ae470e6de48f2a3803b7034756ff50"
}
{
 "prompt_type": "Tech Lead adjudication",
 "full_evidence_in_prompt": false,
 "visible_evidence_chars": 1995,
 "ends_mid_word": false,
 "truncation_marker_present": true,
 "full_sha256": "fbad05d2e2435a37ee5b74acfca72ac7d1ae470e6de48f2a3803b7034756ff50"
}

All three adjudication prompt types route the finding through _finding_public: True
Silent slice (no marker) somewhere: False
Mid-word cut somewhere: False

Note (context, outside this item's fix scope):
the Tech Lead prompt also caps the dispute
positions inline (raiser_quote 400, refuter_reason
300, refuter_quote 400 chars) from the live dispute
object; storage keeps full text in sidecars.

OUTCOME: prompts now carry the explicit truncation marker — the hardening is in place; see the suite tests for the boundary guarantees.
