Generic LLM
waitingOne case, two systems
Stop Regenerating Answers.
Interrogate The Model
Generic LLM
generated0 tokens
total spend$0.0000
Shared elapsed time0:00Projection accelerated 20×
Primordia
Copilot usage0 LLM tokens
model spend$0.000000
Primordia
liveOne model. Every answer stays connected.
Search company…↵
Select a company to open its model
Primordia run complete0:20 wall clock · model coherent and ready
36,000× cheaper350× faster
Fully traceable. Always coherent and up-to-date.The most efficient and trustworthy analysis.
Not a research chatbot.
A probabilistic investment system
Every name in the market, modelled as an interrogable program — calibrated, traceable, and forward-tested.
From a question to a vettable case
Six stages. Each one structured, traceable, and adversarially challenged.
Every claim in the memo unpacks like this
Source to Belief
Lattice Q1 2026 Press Release · BusinessWire
ir.latticesemi.com
Lattice Semiconductor Reports 42% YoY Q1 2026 Revenue Growth
“Compute & Communications business achieved record revenue, growing 86% year-over-year, driven by AI data center demand.”
“AI server FPGA attach rates are rising from approximately 1.5 per server in 2024 to over 3 per server in 2026, with densities reaching 50+ chips per server rack.”
Evidence Object
ir.latticesemi.com
GROUNDING
“AI server FPGA attach rate driving accelerating demand for Lattice products.”
Belief Prior
Beta (α=3.5, β=1)
UPDATES DRIVER
AI DEMAND ++ STRONGLY POSITIVE
What ‘-9% expected return’ actually means
Case to Conviction
Case
Lattice Semiconductor
LLSCC
ai demand
→
share price
LIKELY TRUE
AI Capex Acceleration
POSSIBLE
Valuation Mean Reversion
Direction Probabilities
LONG
42%
HOLD
33%
SHORT
25%
CURRENT
$125.43
TARGET (GEO MEAN)
$114.12
EXPECTED RETURN
-9.0%
CONVICTION
Medium
Scenario Distribution — 2027
Edits don’t rewrite prose — they propagate through the model
A conversational interface sits on top of every case. Tune assumptions, attach evidence, find the boundary that flips the call — and watch the simulation rerun.
CASE
Lattice Semiconductor LSCC
3 edits applied · re-simulated in 4.2s
Analyst Console
Live
Analyst
What if AI capex grows at only 5% YoY instead of 15%?
Updating Evidence_AiCapexGrowth prior.
Beta(2.2, 4.1) → Beta(2.2, 6.3) · re-simulating 10,000 trajectories
Analyst
And attach this: Altera is re-entering the low-power FPGA market in H2 2026.
Linked new evidence to Assumption_MarketShare (weight 1.2).
Beta(2.2, 4.1) → Beta(2.2, 6.3) · re-simulating 10,000 trajectories
Analyst
What would I need to believe for this to still be a LONG?
Boundary found: AI Compute revenue CAGR ≥ 38% AND gross margin ≥ 68%.
Beta(2.2, 4.1) → Beta(2.2, 6.3) · re-simulating 10,000 trajectories
Tune an assumption, attach evidence, or ask for a counterfactual…
Before Edits
HOLD
EXPECTED RETURN
$125.43
CONVICTION
MEDIUM
LONG
42%
HOLD
33%
SHORT
25%
RE-SIMULATE ↓
Before Edits
HOLD
EXPECTED RETURN
$125.43
CONVICTION
MEDIUM
LONG
42%
HOLD
33%
SHORT
25%
What the analyst can do
Counterfactual — “what flips this?”
Tune any assumption
Add custom evidence
Re-simulate in seconds
Version & diff edits
Drill to source
Flag for QA
Converse in analyst language
Every edit triggers a fresh Monte Carlo run. The recommendation stays grounded in the math, not in the prose.
Why Trust It
Forward-tested. Calibrated
Forward-tested on 263 predictions. No backtest, no lookahead.
01 / KNOWS WHAT IT KNOWS
Confidence predicts accuracy
5.8× separation between high- and low-conviction calls.
02 / PROBABILISTIC CALIBRATION
Observations match theory
When the model says ‘5% chance below X’, that’s what happens 5% of the time.
Every prediction in this dataset was made before observing the outcome. No curve-fit. No data snooping. Credibility comes from how the model performs on tomorrow’s prices, not yesterday’s. Bottom line: the model’s confidence is actionable. Trade the high-conviction calls.