Step 3 of 8Back to start
Belief graph

Your portfolio can look diversified and still depend on the same belief.

LyraMind maps the assumptions shared across positions, sectors and strategies, so hidden thesis concentration becomes visible. One sentence about capital spending can sit underneath a chip designer, a network vendor, a power-equipment maker, a data-centre REIT and an independent power producer at once, in four different sectors, which is exactly why a sector view can miss it.

Illustrative portfolio benchmark · 240-position simulated portfolio
One belief, seven positions, four sectors A single assumption about AI infrastructure spending connects seven positions. Four sit in Information Technology; three sit in Industrials, Real Estate and Utilities, outside the belief's largest sector. ONE BELIEF AI infrastructure spending remains above 20% annual growth through FY2028 7 positions · 4 sectors Information Technology largest sector 4 positions Industrials outside largest sector Real Estate outside largest sector Utilities outside largest sector NVDA AVGO CRDO ANET VRT DLR VST A sector report groups these four and separates the three below. $126.9M of gross exposure sits outside Information Technology
Illustrative portfolio benchmark. The diagram plots one belief from a 240-position simulated institutional portfolio. Ticker symbols are real so that sector structure is realistic; the positions, weights, beliefs and exposures are invented. This is not a portfolio, a view, or a recommendation.
Hidden belief concentration

Exposure sector views can miss.

240positions
569distinct beliefs
13beliefs shared across positions
11shared beliefs crossing sectors
$707Mof shared-belief exposure sits outside each belief's largest sector
29.5%of gross shared-belief exposure crosses sector boundaries
Illustrative portfolio benchmark · 240-position simulated institutional portfolio. Every figure in this section is generated by proof.synthbook. None of it is observed customer assets, a real fund, or live evidence.
29.5% of gross shared-belief exposure crosses sector boundaries. Sector diversification can still hide belief concentration.
Illustrative portfolio benchmark · 240-position simulated portfolio
“AI infrastructure spending remains above 20% annual growth through FY2028”
7 positions · 4 sectors · $329.8M of belief-linked exposure
NVDAAVGOCRDOANET VSTDLRVRT

$126.9M of that belief-linked exposure sits outside Information Technology.

This is exposure that depends on the belief, not an estimate of what would be lost. Direction is not modelled: a real book holds shorts and hedges against the same belief, so no figure here says which way the money moves. Exposure is reported gross, never net, if the assumption under a short breaks it matters exactly as much as one under a long, so belief exposure nets nothing. There is no blended concentration score: position count, sector spread and exposure share are kept apart, because a 0-100 number becomes the thing a risk committee argues about instead of the thing it is about.

LyraMind shows the belief connecting those positions, the evidence supporting or challenging it, and the decisions that rely on it.

Discovery

You weren't watching this.

Monitoring answers questions about positions you already hold. The harder question is the one you did not think to ask, where a belief you already rely on shows up somewhere you are not looking.

Objects, not documents

Ask in your own words, across everything the record holds.

A management promise, a decision record, an open issue and a filing passage are all addressable in one pass, so “when did they first commit to that number?” is answered from the record with a date, rather than from a paragraph that happens to mention it.

Belief-first search

“Which of our decisions depended on hyperscaler capex?”

The unit of search is the assumption, not the ticker. That returns the set of decisions exposed to a belief, including the ones nobody would have thought to check, because they sit in another sector or another strategy.

Names the gap

An empty result and an unchecked universe must never look alike.

Where coverage does not extend, LyraMind says so rather than returning a confident-looking nothing. “We checked and found nothing” and “we checked nothing” are different sentences and are shown as different sentences.

Discovery is the newest surface described on this page and is being built out with design partners; coverage depends on the sources a firm connects. We describe it here as designed, and we will say when it is generally available rather than before.

For the CIO

What the firm believes, and where it does not agree with itself.

A risk system reports positions across sectors and calls the book diversified. The belief graph asks a different question, and its most useful answer is often about disagreement rather than exposure.

Hidden shared assumptions

Which positions that look diversified are the same bet.

A belief only becomes interesting as a shared object once more than one position rests on it. LyraMind reports the count, the sector spread and the exposure share separately, and shows how much of the book depends on that belief, before it is tested, not after. It does not forecast what a break would cost: direction is not modelled.

Thesis concentration

The firm does not currently agree with itself.

The same assumption recorded under seven positions can carry seven different states, because seven analysts maintained them separately. That is not noise to normalise away, one analyst has already marked a belief weakening and six have not, and nobody could see that from a per-position screen.

Firm-level calibration

Where the house is systematically early, late, or wrong.

Built from the same three records: proposed, decided, resolved, aggregated across desks. Reported only above the sample floor, and reported as counts below it.

Answerable oversight

“Show me every decision that depended on this belief, and who signed it.”

One query against the record rather than an email thread. The evidence, the proposal, the decider and the outcome are all attached, and none of them was overwritten along the way.