Anatomy of a thesis
The clearest way to explain CommonQuant is to show you what it produces. So this page follows a single claim all the way through the product: from loose hunch, to fixed conditions, to evidence, to a verdict that stays open.
The claim is one you have probably heard this month and may hold yourself. What follows is a worked example, constructed for illustration. The live system runs continuously; the shape of the document is exactly what it produces.
01
Thesis
“AI capex is peaking” is a feeling. You could hold it for two years and never find out whether it was true. The system's first job is to give the feeling a way to be wrong: five conditions, each one measurable, fixed before any test runs. After registration the conditions lock, so nobody can quietly move a goalpost later. Including you.
THESIS T-0192 · registered 2026-05-14
“AI capex is peaking”
sharpened — if AI capex is peaking, then within two quarters:
C1 semiconductor lead times shorten
C2 hyperscaler capex guidance decelerates q/q
C3 data-center pre-leasing slows
C4 GPU spot pricing softens
C5 optical component book-to-bill falls below 1.0
conditions locked at registration
This step is the hardest part of the product. Most beliefs resist being pinned down, and it is worth noticing which of yours resist the most.
02
Evidence
Evidence arrives on two fronts. Data: what shipments, prices, and filings are doing. Narrative: what is being said, and by whom. Every item carries a source and a trust tier, and nothing enters the record without both.
EVIDENCE E-4471 · data · 2026-06-02
TSMC advanced-node lead times: 38wk → 31wk over two quarters
source: supply-chain series · trust tier A
bears on C1 · supports
EVIDENCE E-4519 · narrative · 2026-06-11
hyperscaler earnings calls: capex language unchanged. “accelerating investment” in 3 of 4 transcripts
source: earnings transcripts · trust tier A
bears on C2 · contradicts
These two items disagree, and the system keeps them that way. A gap between what is happening and what is being said gets recorded as a finding in its own right.
03
Test
Tests run against the registered conditions and render pass or fail. Not a score. We tried scores in our own research: how strongly a signal fired predicted nothing beyond whether it fired at all, so we took the decimal places out.
TEST RUN · T-0192 · 2026-06-15
walk-forward 2019–2026
PASS
placebo (shuffled dates)
PASS
regime split, pre/post 2023
FAIL
C1 lead-time threshold
PASS
C2 guidance deceleration
FAIL
The placebo run feeds the same machinery data with shuffled dates. If a thesis passes on nonsense, the test was measuring nothing, and it is the test that gets rejected.
04
Verdict
If you came here from the front page, you have already seen this verdict. This document is where it came from.
VERDICT · T-0192 · as of 2026-06-15
CHALLENGED
2 of 5 conditions failing
flips to FALSIFIED if C4 fails
recovers to SUPPORTED if C2 reverses within one quarter
Challenged means the claim is losing on two of its five conditions while the rest hold. A verdict is a state, not advice. It records what the system believes right now and names the evidence that would move it, in either direction.
05
Ledger
Every claim you run lands here and stays. Claims close; they never disappear.
LEDGER · append-only
T-0192 AI capex is peaking
CHALLENGED · open
T-0187 yen carry unwind, round two
SUPPORTED · open
T-0164 small caps lead in easing cycles
FALSIFIED · closed 2026-04
T-0158 freight rates lead CPI by 4 months
INSUFFICIENT EVIDENCE · open
After a year, the ledger reads as something few people ever get to see: which of your beliefs survived contact with the market, and which kinds keep dying on you. The second list is usually the more educational one.
↺
Nine days later
EVIDENCE E-4602 · data · 2026-06-24
GPU spot pricing: −7% month over month
bears on C4 · supports
→ tests queued · verdict will re-render
A new item arrived after the verdict, bearing on condition four. This is why the document has no last page: markets drift, and a conclusion from March can be wrong by June without anyone noticing. A thesis stays open for as long as you hold the belief.
The contract
The document you just read kept to three rules. So does the product.
Judgment stays with you.
CommonQuant tests claims. It never tells you what to trade — the position, if you take one, is yours alone.
Every result shows its work.
A verdict traces back through the tests and evidence that produced it, down to individual sources. No number ships without a test behind it, ours included: the method page keeps a graveyard of the ideas we killed.
The work matches the timescale of the claim.
A thesis resolves over weeks and months, so that is where the engineering went — holding a claim under test the whole way, rather than shaving milliseconds off a quote.
The machinery underneath — the causal modeling, the inference engine, the validation regime — has its own page. Read about the method.