What Is at Stake
In finance, the cost of a stale assumption is capital allocated against a thesis that the evidence no longer supports. It is a risk that was flagged in one analysis but buried in another. It is a market shift that no one saw because the models were still operating on last quarter's assumptions.
Financial agents process vast amounts of data: earnings reports, SEC filings, market signals, analyst commentary, macroeconomic indicators. The challenge is maintaining a coherent, current view of what that information means.
What Beliefs Make Visible
Thesis evolution under new evidence
An investment thesis starts as a belief: "This sector will outperform based on margin expansion." As data arrives, quarterly earnings, competitor moves, regulatory changes, the thesis should evolve. Without beliefs, the thesis lives in a static memo. With beliefs, every piece of evidence either strengthens, weakens, or contradicts it.
1┌──────────────────────────────────────────────────────────────┐
2│ INVESTMENT THESIS OVER TIME │
3│ │
4│ Q1: "SaaS margins will expand" 88% conf │ 2 sources │
5│ Q2: Competitor cuts prices → 71% conf │
6│ Q3: Input costs rise unexpectedly → 58% conf │
7│ Q4: New regulation favors incumbents → 67% conf │
8│ │
9│ The ledger shows exactly when and why confidence shifted. │
10│ A portfolio manager can trace any position back to its │
11│ evidence chain, not just the original pitch. │
12└──────────────────────────────────────────────────────────────┘Contradictions across sources
An equity research agent reads a bullish analyst report and a bearish SEC filing on the same company in the same week. Without beliefs, both sit in context with equal weight. With beliefs, the contradiction is flagged. The system knows the sources disagree and can surface the specific claims that conflict.
Risk factors that decay
A risk assessment from six months ago is not the same as one from yesterday. Temporal decay ensures that old risk evaluations lose certainty over time, creating pressure to reassess rather than silently carrying forward stale ratings.
Cross-asset pattern detection
When beliefs are explicit across asset classes, an agent can detect patterns that span equities, credit, and macro: contradictions between what the bond market believes and what the equity market prices in. These cross-domain tensions are invisible to systems that treat each analysis as an isolated document.
What Agents Can See That We Cannot
A human analyst tracks 15-20 positions deeply. A belief-aware agent can maintain structured beliefs across hundreds of securities simultaneously, tracking confidence, evidence chains, contradictions, and decay across a universe too large for any individual to hold.
The agent does not just recall that it read a filing. It knows that filing contradicted the analyst consensus, that the contradiction has not been resolved, and that the claim's evidence has decayed since Q2.
1const beliefs = new Beliefs({
2 apiKey: process.env.BELIEFS_KEY,
3 agent: 'equity-research',
4 namespace: 'tech-sector-review',
5 writeScope: 'space',
6})
7
8await beliefs.add('ACME gross margins contracted 200bps', {
9 confidence: 0.92,
10 evidence: 'ACME Q3 10-Q filing',
11})
12
13await beliefs.add('Sector-wide margin expansion thesis weakening', {
14 confidence: 0.65,
15 evidence: 'ACME Q3 10-Q filing',
16})