Task 13284: init/golden mismatch (Streets End filled with 9999) + instruction contradicts golden

[swenyang]

Task: 13284 (Verified-400)

The task is unsolvable as shipped: the golden answer depends on reference data that is not in the init file and is not in the scored range, and the instruction contradicts what the golden actually does.

1. The golden was computed against a modified Streets sheet not given to the agent

Base!E looks up Streets!F where Base!D falls within [Streets!C, Streets!D]. In the init, several Streets!D (End) cells are blank. In the golden, those same cells have been filled in (highlighted) with values that do not exist in the init:

Streets row	street	Start (C)	End (D) init → golden	note (col G)
5	Avenida Braz De Pina	1358	(blank) → 9999
8	Avenida Meriti	920	(blank) → 920	Conjunto Habitacional
10	Avenida Meriti	2301	(blank) → 9999
12	Avenida Oliveira Belo	701	(blank) → 9999
15	Avenida Vicente De Carvalho	1179	(blank) → 1179	Conjunto Habitacional
18	Avenida Vicente De Carvalho	1501	(blank) → 9999

So the open-ended "last segment" of each street was given an End sentinel (9999, i.e. "no upper limit"), and two rows were closed to a single number. The golden's Base!E values (E5=Sonia, E8=Mariza, E10=Marcy, E13=Carlos, E16=Fontana) only exist because of these fills.

But answer_position is 'Base'!E2:E26 — only column E is scored. The agent receives the init (blank End cells), is never told to fill Streets!D, and could not derive the 9999 sentinel (nor the two Conjunto Habitacional single-value fills). It therefore cannot reproduce the golden's E values.

2. The instruction contradicts the golden

"Only consider rows where both the Start (column C) and End (column D) values are populated."

The golden does the opposite: it makes the blank End cells populated (fills 9999) and includes those open-ended segments. Read against the init (where those cells are blank), the sentence tells a solver to exclude exactly the rows the golden includes. Any faithful implementation returns "" for the five queries that only match an open-ended segment (Base rows for house numbers 2354, 920, 820, 1179, 1691) and is scored wrong on E5, E8, E10, E13, E16.

Reproduction

Any correct INDEX/MATCH (or XLOOKUP) that honors "both Start and End populated" over the init data returns "" at Base!E5/E8/E10/E13/E16, while the golden has Sonia/Mariza/Marcy/Carlos/Fontana. 5-cell mismatch → fail, deterministically.

Suggested fix

Either ship the filled Streets!D values (the 9999 sentinels + the two single-value rows) in the init, or reword the instruction so it matches the golden (treat a blank End as open-ended; only exclude rows where both Start and End are blank). As written, init, golden, and the instruction are mutually inconsistent.