Correction: cash figures overstated 1000× on ~21 ticker pages

MRdata · correction · 2026-08-14

Between 2026-07-22 and 2026-08-14, the quarterly cash balance shown on roughly 21 ticker pages was overstated 1000×. Junior explorers appeared to hold billions: SGQ showed $80.65bn where the true figure — stated plainly in its own Appendix 4C — is $80.65M. NVX, ALR and ~18 others carried the same class of error. Every wrong figure was labelled high-confidence.

What happened

Appendix 4C/5B tables quote amounts in thousands, marked $A'000. The parser's "amounts in millions" detector accepted the digit pattern '000,000 without requiring the quote mark — so the tail of any large plain number on the page (say, "150,000,000 fully paid shares") read as a millions-declaration. A cash figure of 80,652 in a thousands-quoted table was then scaled ×1,000,000 instead of ×1,000.

The July quarterly wave (07-22 onward) was the first time the bug had a full season of 4C filings to bite; about 21 of 126 parsed cash rows came out 1000× high.

The fix (2026-08-14)

1. Units come from the form's declaration, not digit-count guesswork. The parser reads the declared marker — $A'000 → ×1,000; A$'m / "in millions" → ×1,000,000; no marker → raw dollars — and a units declaration now requires the quote mark, so a bare digit run can never masquerade as one.
2. A magnitude guard in front of the lake. A cash reading larger than 5× the company's market cap (or $1.5bn with no resolvable cap) is rejected before it is stored, loudly. Shells trading below cash backing stay legal — the guard catches unit-scale errors, not balance-sheet oddities.
3. Every stored row re-extracted from the original filings with the fixed parser. The affected rows were corrected on 2026-08-14; the rest were confirmed unchanged. One eye-catching figure survived because it is real: PLS genuinely holds ~$2.29bn.

The same session also began capturing quarterly operating outflow and implied funding runway from the same filings — validated against the forms' own "estimated quarters of funding available" line — so future unit errors have a second, independent number to disagree with.

Errors here get corrected in public, with the mechanism named. That is the deal this site runs on.