Fahali / autonomous portfolio risk analyst

What is your portfolio hiding?

Enter what you own. Fahali checks whether the positions still behave independently, translates a conditional shock into dollars, and shows whether the evidence is active.

Your holdingsUp to 7 symbols

Opening the current public book...

No account · no brokerage login · symbols only · nothing stored · read-only

Sanitized product example · sample position values Portfolio state · WATCH

One shared path · 7 positions

Seven positions. One shared downside path.

The trigger has not arrived. The consequence is still worth seeing before you decide what deserves attention.

Conditional book change −$33,873

sample values if all 7 fall 10% together

7 positions in the sample scenario
shared move−10%
Scenario scope 7 / 7

priced positions included in this transparent sample

  1. PositionsSeven priced positions enter one synchronized scenario
  2. ConsequenceA 10% shared fall produces a -$33,873 sample change
  3. ActivationNo active case in this example

Transparent linear scenario · not a forecast · sample position values

Decision proof · one real book

Five holdings looked diversified. Two began behaving like one.

This is not a throughput counter or a selected winning trade. It is the same public book, read twice, showing what Fahali knew, what it withheld, and what changed.

SPY · GLD · TLT · NVDA · BTC

Fahali measured the book, then refused to overstate the holdings.

Measured: 169 aligned observations. The five holdings provided about 3.3 distinct sources of movement; NVDA and SPY moved together at 0.61.

Withheld: SPY, GLD, TLT, and NVDA had no resolved 72-hour forecast history. Only BTC produced a directional read.

The overlap became more material; the uncertainty did not disappear.

With 472 aligned observations, the five holdings provided only about three distinct sources of movement and NVDA/SPY moved together more strongly at 0.76. All five returned a read, but four were still not yet established by resolved history.

Decision: review the NVDA/SPY cluster. Do not treat the four preliminary directional reads as established evidence.

The current read above refreshes every 10 minutes while this page is open.
Verification notes

The first read is preserved in the 26 July public-surface verification; that source retained the date but not the exact minute. The second read was returned by /api/report/custom at 2026-08-01T12:05:25.093Z. Its SPY shock assumes equal weights and a stable linear relationship. The watcher message above is reconstructed from those two captured reads; it was not sent to a customer.

For AI agents · free key, no card

Your agent has no risk conscience. This is one it can check.

Finance agents can sound certain when the evidence is not. Fahali answers over MCP or REST with typed responses: a missing value stays missing, each published call states its horizon, and the public record keeps the misses beside the hits.

Weight a signal by its measured lead

Per engine: how many hours ahead it tended to warn on its correct calls, with the misses and the sample size. A price feed cannot give an agent this.

Keep the evidence boundary

Responses name the observation time, what was measured, and what was missing, so an agent can preserve uncertainty instead of silently filling a gap.

Refusal is a result

Claims below their sample threshold are withheld rather than rounded up. Quiet markets resolve as unresolved, never as wins. Silence is explicit, never ambiguous.

Check it before you trust it — neither of these needs a key:

curl https://app.fahaliai.com/api/track-record/lead-time
curl https://app.fahaliai.com/api/tape
Get a free key → 50 calls/day · MCP + REST · typed responses · no card

Read-only by design · no order routing, no path to capital · observation, not advice