Trade Intelligence
What APEX understands about a specific trade: the thesis, the setup, the risk, the levels, the context of the moment and the decision — stored together, for good.
The reasoning envelope
APEX used to generate evidence, counter-evidence, scenarios, a confirmation condition and invalidation in every analysis — and throw them away. None of it reached the history, the memory or the learning.
Now every analysis stores its complete reasoning in an immutable column. Not the verdict: the why. And with a separation that matters more than it seems.
- Your thesis
- What you wrote, word for word. Its own field.
- APEX's interpretation
- The model's restatement. A different field. The two used to share a column, and memory ended up quoting APEX's version as if you had written it.
- Evidence
- What supports the thesis, each piece with its strength.
- Counter-evidence
- What questions it, also with its strength. In a fundamental analysis, the risks come in here with their severity.
- Scenarios, confirmation and invalidation
- What would happen, what would reinforce it and what voids it. When a mode does not produce them —the fundamental one does not ask for scenarios— they are counted as missing, not faked.
- Observation and market context
- The environment measured at that moment, with its time and source, stored alongside the reasoning.
“The thesis was reasonable and the market changed” or “the context already contradicted it”
They are two very different stories about the same losing trade, and they can only be told apart if the context of the time survived the analysis.
None used to survive: no market data was stored with the analysis, so reviewing an old decision meant judging it with today's environment. That is literal hindsight, and it makes the same reasoning look brilliant or negligent depending on what happened afterwards.
The same correction was applied to your risk profile: when APEX evaluates whether a trade from three months ago was oversized, it uses the usual risk you had back then — the median of the trades before it —, not today's. Measured: the same trade is flagged in one position and not in another, and where there is no earlier sample, the factor is declared missing instead of inventing a neutral value.
Whoever writes the analysis is whoever saves its reasoning. That is why there cannot be a saved analysis without an envelope, and a test pins that invariant.
A key that is present is not content
APEX rejects a response when a critical reasoning stage arrives with the key in place and empty. An object with blank text and a filled-in strength was a formally valid and semantically empty response, and it used to pass.
The check is calibrated conservatively and measured against real responses: a long production analysis passes, a short but real one passes, and fields that can legitimately be empty still pass. And it is only required on the stages that that analysis's contract asked for: validating a fallback model's response against a contract nobody asked it for would throw it out over a field it had no reason to bring.