Every figure our AI states, checked against the figure our code derived
Each stored AI read on this site (the research brief, the investment committee, a model answer to Ask this ticker, the chart explainer and the tool explainers) is stored with the figures it was given: the revenue, EPS, net margin, F-Score, Form 4 count and filing dates our code derived from the company's SEC filings for that period, or the numbered facts of a code-written read. So every sentence that states a figure can be compared, by code, with the figure our code derived. This page runs that comparison over every stored read, checked 2026-11-05. Each read also prints its own count beneath it.
The count
14 stored reads; 202 sentences state a figure. 196 agree with the figure our code derived, 0 disagree, and 6 are untestable because they state a figure we hold no derivation for.
| Surface | Reads | Sentences with a figure | Agree | Disagree | Untestable |
|---|---|---|---|---|---|
| Research brief | 5 | 52 | 52 | 0 | 0 |
| Investment committee | 5 | 106 | 100 | 0 | 6 |
| Ask this ticker (model answers) | 1 | 2 | 2 | 0 | 0 |
| Chart explainer | 2 | 27 | 27 | 0 | 0 |
| Tool explainer | 1 | 15 | 15 | 0 | 0 |
Disagreements
None in the 202 sentences checked. When a read states a figure our code did not derive for the metric it names, it is listed here with both figures.
Untestable sentences
- PLTR, investment committee written 2026-07-21, bear claim 2: “A $323B market cap resting on just $4.48B of revenue implies a lofty ~72× sales-type premium that dwarfs the underlying business scale.” We hold no derivation for 72×.
- PLTR, investment committee written 2026-07-21, risk 1: “At 214× earnings and roughly 72× sales, the valuation prices in flawless multi-year execution, so any deceleration from the current 56.2% growth rate could trigger sharp multiple compression.” We hold no derivation for 72×.
- SNDK, investment committee written 2026-07-14, bear claim 2: “The $247.90B market cap against just $7.36B in revenue implies a price-to-sales multiple above 30x, a demanding valuation for a business that is not generating profits.” We hold no derivation for 30x.
- SNDK, investment committee written 2026-07-14, risk 1: “Deep unprofitability (EPS -$11.32, net margin -22.3%) combined with a demanding valuation (~$247.90B market cap on $7.36B revenue, >30x sales) leaves little margin for error; continued losses would invalidate the bull's turnaround framing.” We hold no derivation for 30x.
- SNDK, investment committee written 2026-07-14, synthesis, sentence 1: “Sandisk shows double-digit revenue growth ($7.36B, +10.4% y/y) but is deeply unprofitable (EPS -$11.32, net margin -22.3%) while carrying a demanding valuation (~$247.90B market cap, >30x sales) [revenue][eps][marketCap].” We hold no derivation for 30x.
- SNDK, investment committee written 2026-07-14, strongest bear: “The company pairs deep losses (net margin -22.3%, EPS -$11.32) with an extreme valuation (>30x sales at $247.90B on $7.36B revenue), leaving almost no margin for error—amplified by the 12.63% single-day drop [eps][marketCap][revenue][price].” We hold no derivation for 30x.
How a sentence is checked
- Every figure in the sentence is read out: an amount, a percent, a multiple, a score such as 6/8, a count, a date. Form names (10-Q, Form 4), windows (52-week, the last 90 days) and bare years are not figures.
- The figure is given to the nearest metric the sentence names that can hold it (revenue, net margin, EPS, market cap, the close, P/E, F-Score, Form 4 filings, congressional disclosures, the filing date) and compared with our figure for that metric. A figure exactly equal to another of our figures is counted as that figure, because naming by the nearest word is not exact.
- A figure no metric claims is compared with every figure of its kind in the read's evidence. A match agrees; no match leaves it untestable, never a disagreement.
- The tolerance is the precision the sentence wrote: $95B agrees with 94.83B, $94.9B would not. “Nearly”, “about” and “roughly” widen it to 5 percent; “more than” and “under” are checked as the inequality they state; a direction word (“fell 2.9%”) is compared with the sign of our figure.
- A sentence agrees when all its figures agree, disagrees when any figure disagrees, and is untestable otherwise. Sentences with no figure are not counted.
What this does not check: whether a figure our code derived is itself right (each figure on a company page carries its own filing receipt), and anything a sentence says without a number. The comparator is the evidence stored with the read, so a read is checked against the figures of its own period, not today's.
Keep going: filed outcomes our AI reads name, settled · every AI prompt, published · the research brief · methodology