Most research products tell you what they measure. Few tell you what they refuse to measure. The refusal list is the more interesting half of the story. It's where a methodology earns its trust, or fails to.
Below is the full list of inputs Quantora deliberately keeps out of the scoring stack, with the specific reason for each. None of them are accidental gaps. All five were considered, tested, and excluded on purpose.
1. Aggregated sentiment
Sentiment indices conflate noise with signal, and they describe where the crowd already is, not where it's heading. By the time an index registers fear, the move that caused the fear is already priced. Greed reads the same way, one step late.
We score the underlying drivers that cause sentiment to shift instead: real rates, growth surprises, positioning extremes, the volatility regime. The drivers move first. Sentiment moves after. We want the first move, not the echo.
2. Social signal
Social chatter is not a primary proof layer in the product. It can help in narrow cases, such as early reactions to scheduled releases or dispersion in retail commentary on a specific instrument. But it is structurally weak. It is gameable, sparse outside the busiest names, and easily dominated by a handful of loud accounts.
The workflow is built on ranked context and explicit market drivers. If a market is interesting, you should be able to see why without the social feed in the room at all.
3. Retail flow
Retail flow is not promoted as a public decision layer. There is a defensible case for reading it as a contrarian tell in narrow circumstances, but it carries no weight in the headline methodology.
The deeper reason is a rule we apply to every input. If a signal can't be explained clearly inside the workflow, if its inclusion rests on "we found it works" rather than "we know why it works", it doesn't get to be a promise. If we can't explain the edge, we won't stake the product on it.
4. Technical patterns
The model holds no view on chart shapes. Head and shoulders, flags, pennants, divergences: none are scored as standalone inputs.
This isn't a dismissal of price action. Price action shows up implicitly in the factors we do measure: positioning extremes, the volatility regime, breadth, confirmation across assets. Anything a pattern captures that survives a falsification test ends up inside one of those factors. Patterns that don't survive falsification don't get to be scores.
5. Headline commodity proxies
A generic crude or commodity basket is a poor proxy for any single instrument. The aggregate moves for reasons (OPEC, inventories, the dollar) that may have nothing to do with the exposure you're trying to score.
So we use the relevant complex per instrument. Different metals take different inputs. An AUD read leans on iron ore and coal in a way that doesn't transfer to gold. A CAD read leans on energy in a way that doesn't transfer to anything else. Specificity over convenience, every time.
What this list is for
It's a stress test. If you think one of these belongs in the stack, say so. We'll either explain why it stays out, or we'll change the methodology. The doctrine is meant to be argued with, not defended.
The discipline was never "fewer signals is better." It's that every signal has to earn its place by naming what it adds and what would invalidate it. Anything that can't pass that test stays out.
