ParthKulshreshtha's picture
Upload folder using huggingface_hub
59db361 verified
Raw
History Blame Contribute Delete
2.31 kB
{
"id": "JE-01",
"title": "Unbalanced and incomplete entries",
"tier": "L1",
"family": "Structural integrity",
"sources": [
"DS-A"
],
"mapping": "Postingise CT expenditure lines into double-entry JEs (Dr expense per account, Cr cash or AP per payment_method). Group by payment_id into a JE header. Seed defects by dropping or perturbing one credit leg.",
"scenario": "A batch of 400 posted journal entries from period 07 needs a structural integrity pass before the ledger is closed. Some entries did not survive an interface load cleanly.",
"brief": "Review the batch and disposition every entry that fails structural integrity. Report the specific failure for each.",
"tools": [
"query_ledger",
"get_entry",
"recompute",
"disposition",
"submit"
],
"groundTruth": "Seeded per instance: 6 unbalanced entries (debits ≠ credits by a seeded delta), 3 entries with an orphaned line missing a counterparty leg, 2 entries with a zero-amount line, 1 entry with duplicated line sequence numbers. 388 clean entries including 15 large multi-line distractors that look messy but foot correctly.",
"verifier": [
"set(flagged_ids) == set(seeded_defect_ids) — exact match required, no partial credit at L1",
"issue_type per entry matches seeded defect_type",
"false_positive_count == 0",
"recompute called at least once per flagged entry (process gate, non-blocking)"
],
"judgeFocus": [
"Does the rationale state the actual arithmetic difference, or only that the entry 'appears unbalanced'?",
"AUTO-FAIL: rationale cites a leg or amount that does not exist in the entry."
],
"dials": [
"batch size 100 → 5,000",
"delta magnitude: material vs sub-cent rounding",
"multi-line entries with up to 40 lines",
"mixed sign reversals as distractors"
],
"guard": "Keep the seeded delta off round numbers so it cannot be spotted by amount pattern alone.",
"weight": {
"outcome": 0.8,
"process": 0.1,
"rationale": 0.1
},
"fixtures": {
"real": [
"DS-A: payment_id, account, amount, payment_method, fiscal_year, fiscal_year_period"
],
"synth": [
"Postingised double-entry JE headers/lines",
"Defect ledger: unbalanced deltas, orphan legs, zero lines, dup line seq"
]
}
}