Three Verified-400 tasks have malformed answer_position / answer_sheet values in dataset.json (independent of the parser bug in #33 — these are wrong data, not just naive parsing). Each makes the golden fail its own self-comparison, so the task is effectively unscoreable.
60-7 — answer_sheet lists every mentioned sheet instead of the destination
answer_position = "A3:E11"
answer_sheet = "Consolidated Tracker,Existing Task,Additions,Retired"
The prompt says the answer is written to the Consolidated Tracker tab starting at A3 (Consolidated Tracker = Existing Task + Additions - Retired). Only that one sheet is the answer location, but answer_sheet lists all four sheets named in the task. Combined with the single range A3:E11, the harness builds an invalid Consolidated Tracker,Existing Task,Additions,Retired!A3:E11.
Fix: answer_sheet = "Consolidated Tracker".
283-32 — answer_position missing the leading quote on the first sheet
answer_position = "Sheet3'!A:G,'Sheet4'!A:G" (Sheet3 has a trailing quote but no leading quote)
answer_sheet = "'Sheet3','Sheet4'"
Fix: answer_position = "'Sheet3'!A:G,'Sheet4'!A:G". (This task also needs the whole-column-range handling noted in #33.)
41-47 — same missing-leading-quote defect
answer_position = "OUT CAS'!A2:C1529,'OUT CAS'!E2:G586,'OUT CAS'!I2:K13,'OUT CAS'!L2:O8"
The first OUT CAS' is missing its leading quote (the other three are correctly 'OUT CAS'). It happens to still evaluate today because the naive per-part quote stripping tolerates it, but it is malformed and breaks any quote-aware parser.
Fix: answer_position = "'OUT CAS'!A2:C1529,'OUT CAS'!E2:G586,'OUT CAS'!I2:K13,'OUT CAS'!L2:O8".
All three golden files are otherwise correct — only the answer_position / answer_sheet metadata is wrong. After the fixes, each golden passes self-comparison and the tasks become scoreable.
Three Verified-400 tasks have malformed
answer_position/answer_sheetvalues indataset.json(independent of the parser bug in #33 — these are wrong data, not just naive parsing). Each makes the golden fail its own self-comparison, so the task is effectively unscoreable.60-7—answer_sheetlists every mentioned sheet instead of the destinationThe prompt says the answer is written to the Consolidated Tracker tab starting at A3 (
Consolidated Tracker = Existing Task + Additions - Retired). Only that one sheet is the answer location, butanswer_sheetlists all four sheets named in the task. Combined with the single rangeA3:E11, the harness builds an invalidConsolidated Tracker,Existing Task,Additions,Retired!A3:E11.Fix:
answer_sheet = "Consolidated Tracker".283-32—answer_positionmissing the leading quote on the first sheetFix:
answer_position = "'Sheet3'!A:G,'Sheet4'!A:G". (This task also needs the whole-column-range handling noted in #33.)41-47— same missing-leading-quote defectThe first
OUT CAS'is missing its leading quote (the other three are correctly'OUT CAS'). It happens to still evaluate today because the naive per-part quote stripping tolerates it, but it is malformed and breaks any quote-aware parser.Fix:
answer_position = "'OUT CAS'!A2:C1529,'OUT CAS'!E2:G586,'OUT CAS'!I2:K13,'OUT CAS'!L2:O8".All three golden files are otherwise correct — only the
answer_position/answer_sheetmetadata is wrong. After the fixes, each golden passes self-comparison and the tasks become scoreable.