Every claim is registered with its horizon before the outcome exists, judged against realized price when the horizon elapses, and kept append-only — including the misses and the refusals. What you escalate, you can show.
The thing second-line functions cannot get from most market tooling is not another alert — it is a defensible answer to "what did you know, when did you know it, and why did you act on that one?" Fahali is built so that question has an artefact behind it: the claim, its clock, the evidence it rested on, and what happened next, all fixed before the outcome was known.
The claim and the horizon it is made over are written at detection time, not after. A record built this way cannot be authored to fit what happened, and it cannot be backfilled.
What was measured from a real feed, what is a proxy, and what was missing — kept as separate lists. A proxy is never presented as a measurement.
When the horizon elapses the realized outcome is judged and attached, including when the call was wrong. Nothing is deleted or revised; the record compounds, the narrative does not.
Each verdict carries a signed SHA-256 digest over its own content, so alteration is detectable. Detectable, not impossible — Fahali uses the accurate word.
A good deal, deliberately, and the refusals are on the record next to the calls. Where the evidence conflicts, no direction is assigned and the reason is recorded rather than a direction being filled in. Where a stratum of the judged record has too small an effective sample or fails its publication floor, the figure is withheld with a stated reason rather than rounded up or quietly dropped.
The same discipline applies to Fahali's own detectors. Several have been measured over large samples as no better than their own base rate on a forward question; they are gated out of making standalone published claims rather than left in the brochure, and the judged record covers only the lanes that make judged claims. A vendor's willingness to name its own failures is a fact you can check, which is more than can be said for most accuracy figures.
Without asking us. The judged record is a public endpoint requiring no key: app.fahaliai.com/api/track-record/lead-time returns one row per stratum with the effective sample, the signal-conditioned base rate, precision, recall and the misses. The methodology states the judge and the bar it scores against, and the corrections policy states what happens when a published claim turns out to be wrong.
Read-only by design. Fahali does not route orders, manage capital, or provide buy/sell recommendations — there is no path to capital and therefore no execution risk to control for. It surfaces structured observation so you and your systems can decide. Past outcomes are published including the misses; future outcomes are uncertain.
The claim and the horizon it is made over, fixed at registration time; the evidence behind it split into what was measured from a real feed and what is a proxy; what was missing; the confidence and the scale it was drawn on; the condition that would invalidate it; and a provenance hash over the whole thing. When the horizon elapses the realized outcome is judged and attached — including when the call was wrong.
The outcome record is append-only and every claim is registered before its outcome exists, so a claim cannot be authored to fit what happened. Receipts are signed SHA-256 digests, which makes them tamper-evident: any alteration is detectable after the fact. That is a weaker guarantee than the absolute one vendors tend to advertise, and it is the one Fahali can actually stand behind.
No — Fahali holds no security certifications today, and none is claimed anywhere on this site or in any document. Pursuing them is on the roadmap, not in the present tense. If a certification matters to your approval process, treat that as unmet and raise it before any pilot rather than after.
Anything the evidence does not support. Where signals conflict, no direction is assigned and the reason is recorded. Where a stratum of the judged record has too small an effective sample or fails its publication floor, the figure is withheld with a stated reason rather than rounded up or quietly dropped. Refusals are first-class entries, not gaps.
Nothing here is gated, and nothing asks you to take our word for it.