fix(statistical): an all-blank array is 0, not three different answers

[hhimanshu]

closes #775

Summary

Four functions that should agree gave three different answers for an array where every element is blank:

	before	Sheets
MAX	#REF!	0
MIN	0 ✓	0
MAXA	#N/A	0
MINA	#N/A	0

MIN was the only one already right. A blank-only array is what a range over empty cells materialises as, so =MAX(A1:A3) on an untouched column is the everyday form of this.

Getting the evidence took fixing the probe harness first

An earlier attempt to capture this measured the harness instead of Sheets. With M4 holding 1, =MAX(Data!M4) returned #REF! while =COUNTA(Data!M1:M4) returned 1 — impossible as real behaviour, and only visible because a control asserted a boring known value.

Root cause: the evaluator resolves the setup sidecar from the path it actually reads rows from. These functions read from common/, so the sidecar had to be at test-inputs/common/{FN}.setup.json. It had been placed under google_sheets/statistical/, where it was never read — so no Data sheet was created and every reference correctly returned #REF!.

(SPARKLINE's probes worked only by accident: it is a Sheets-only function with no common/ file, so its rows and sidecar happen to share a directory. The two paths coincided by luck, not by rule.)

The rule generalises past one shape — and that is now captured, not assumed

The predicate is "every argument blank AND at least one arrived as an array", which is broader than the single shape first captured. At the Value layer the engine cannot see range geometry, so a narrower predicate is not expressible.

Rather than ship that as a judgement call, all seven shapes were probed — each with a populated control proving the range actually resolved:

shape	value	control
A1:A1 single-cell	0	7
A1:A3 single-column	0	7
A1:B1 multi-column	0	7
A1:B2 2-D	0	7
A1:A100 far past the used area	0	7
A1,A1:A3 scalar then range	0	7
A1:A3,A1 reversed	0	7

Identical for all four functions. Blankness asserted rather than assumed (=COUNTA(Data!A1:B100) → 0, =COUNTBLANK(Data!A1:A100) → 100), and the type read back through a real cell → 0 number, so it is a plain zero and not a date-formatted one.

⚠️ None of those rows are in this repo yet

They live on the conformance-fixtures pipeline branch and land in a separate fixtures-only PR. Two reasons, both forced: three of the four fail until this code exists, and CI rejects any PR touching both fixture TSVs and code.

A reviewer working from this repo alone cannot verify the Sheets answer. The nearest in-repo rows are =MAX({}) → #REF! and the sparkline Data!K1:K1 rows; neither covers a blank-only array. The doc on is_blank_only_array says this explicitly, and each test comment notes that read from this repo alone it pins the behaviour, not the Sheets answer.

Review

• Independent review with a 249,910-case differential built from scratch: zero movement outside the blank-only-array class. The predicate is exhaustive over all 10 Value variants with no catch-all — a new variant is a compile error there — it fires strictly after every error path, and costs nothing on the hot path.
• Two findings, both about claims rather than behaviour, both fixed: the generalisation was undisclosed (now captured and stated), and comments implied local backing they lacked (now explicit about what is and is not checkable here).

How to test

cargo test -p truecalc-core statistical::max
cargo test -p truecalc-core statistical::min
cargo test --workspace

Directly — =MAX(A1:A3), =MIN(A1:A3), =MAXA(A1:A3), =MINA(A1:A3) over three empty cells all return 0. Before: #REF!, 0, #N/A, #N/A.

Test plan

• cargo test --workspace — 3766 passed, 89 suites
• cargo clippy --workspace -- -D warnings — clean
• cargo nextest run --workspace --profile ci — 3759/3759
• No fixture TSV touched
• cargo fmt --check fails only on files this branch never touched (main is not fmt-clean; fmt is not a CI gate)
• CI green

Related

• fix(statistical): MAX/MIN ignore dates inside arrays — a date column returns #REF! (MAX) or 0 (MIN) instead of the extreme date #776 — dates in the same four functions; conflicts with this branch by design and rebases onto it after this lands
• fix(statistical): COUNT does not count dates, so COUNT(MAX(<date range>)) is 0 #780 — COUNT does not count dates
• fix(statistical): MAXA/MINA ignore a Zoned value that MAX/MIN treat as an error #781 — MAXA/MINA drop a Zoned that MAX/MIN error on

🤖 Generated with Claude Code

---

^{Need help on this PR? Tag @codesmith-bot with what you need. Autofix is disabled.}

[github-actions]

Test Coverage by Category

Category	Unit Tests	Google Sheets Conformance	Property Cases	Total
Array	42	547/547 ✓	1,000 (2×500)	1,589
Database	35	182/182 ✓	3,500 (7×500)	3,717
Date	369	418/418 ✓	2,500 (5×500)	3,287
Engineering	245	886/888 ⚠	5,500 (11×500)	6,633
Filter	11	80/80 ✓	4,500 (9×500)	4,591
Financial	149	1,208/1,208 ✓	2,000 (4×500)	3,357
Info	0	256/256 ✓	4,500 (9×500)	4,756
Logical	121	263/263 ✓	3,500 (7×500)	3,884
Lookup	69	392/392 ✓	1,000 (2×500)	1,461
Math	536	2,006/2,006 ✓	8,000 (16×500)	10,542
Operator	87	250/250 ✓	7,500 (15×500)	7,837
Parser	83	92/92 ✓	4,000 (8×500)	4,175
Query	37	—	—	37
Statistical	501	3,156/3,156 ✓	5,000 (10×500)	8,657
Text	298	729/733 ⚠	4,000 (8×500)	5,031
Timezone	47	—	—	47
Volatile	0	—	3,500 (7×500)	3,500
Web	29	56/56 ✓	6,000 (12×500)	6,085
Total	2,915	10,521/10,527	66,000 (132×500)	~79,442

_{✓ = 100% passing · ⚠ = known deviation · The ~79,442 total counts formula evaluations (each conformance row and each property case = 1). GitHub Checks reports 3,759 Rust test functions: 2,915 unit + 159 property functions (shown as cases above) + 685 conformance/integration.}