Earnings call lab · 174,989 calls tested

Blood tests
for stocks.

Q: What does "blood tests for stocks" mean?

A blood test takes a small sample, measures markers like cholesterol, and flags anything outside the healthy range. Artul.ai does the same with a company's earnings call: the transcript is the sample, 30 markers such as evasion, hype, stress and hedging are measured, and each is compared with the normal range of 174,989 calls since 2015. Results outside that range get an H or L flag.

Q: What is checked on each earnings call?

Every call gets the same 500 yes-or-no checks (for example, did management decline to answer a question, was guidance raised, were outside forces blamed), plus counted language markers such as hedges, buzzwords, superlatives, AI mentions and figures per thousand words. These roll up into 30 markers and one grade from A to F.

Q: How is the normal range set?

The normal range is the middle half of all 174,989 graded calls: the 25th to 75th percentile for each marker. A call above the 75th percentile on a risk marker (evasion, stress, promotion, hedging) is flagged H; a call below the 25th percentile on a health marker (candor, specificity, numbers) is flagged L.

Q: Is the grade a buy or sell signal?

No. Like a blood test, the grade describes the sample, not the future. It tells you how the conversation went: how candid, specific and calm management was. We tested hundreds of ways to turn grades into price forecasts and publish the ones that failed. Only the Monthly Four carries a buy, and only as a set of four scored in public.

Q: Which companies are covered, and how fast?

About 9,321 US-listed companies with earnings calls since 2015. New calls are tested within about a day of the transcript. 8,126 calls have been tested so far in 2026.

Q: Is it free?

Every company's previous quarter is free to read in full. The newest quarter, all 500 answers with reasoning, the rankings and the Monthly Four picks are part of the membership, which costs $1 for the first 7 days and then $99 a month.

Every earnings call is a sample. We run it through 500 checks, measure 30 markers against the normal range of 174,989 calls, and flag what is out of range: dodged questions, hype, stress and vague numbers.

Try: NVDA TSLA AAPL PLTR AMZN

Last quarter's report is free for all 9,321 companies. No signup.

Artul lab · Earnings call panel

CompanyTesla, Inc.

SpecimenQ4 2025

CollectedJan 28 ’26

Sample size8,798 words

Running checks… 500 / 500211 yes · 289 no

MarkerResultNormal rangeFlag

Honesty panel

CandorStraight answers, bad news included

6/9

normal 6–7

EvasionQuestions sidestepped or deflected

4/9

normal 2–3

Stonewalls"We don't break that out"

normal 0

Hype panel

PromotionSelling instead of reporting

8/9

normal 4–6

AI mentionsAI, LLM, agentic, generative

normal 0

Stress panel

StressTension in management answers

3/9

normal 2–3

Hedging"Kind of", "probably", per 1k words

11.2

normal 5.5–10.5

Outside forcesMacro, tariffs, weather, FX blamed

normal 1–9

Clarity panel

SpecificityConcrete detail vs generalities

7/9

normal 7–8

NumbersFigures per 1k words; low = vague

8.0

normal 21.2–35.3

6 of 10 markers out of range. Evasion, Promotion, AI mentions and more out of range. Bottom 26% of 174,989 calls.

Full report →

174,989earnings calls tested since 2015

9,321companies, each with its own history

500yes-or-no checks on every call

30markers with a normal range

< 1 dayfrom transcript to lab report

The idea in one table

It works like your annual check-up. For a company.

Share prices tell you what happened. The earnings call tells you how management is doing: whether they answer questions, how much they hype, how tense they sound. We measure that the same way on every call, so you can see what is normal and what is not.

Your blood test

Artul's call test

The sampleA tube of blood

A few millilitres tell the lab a lot about the whole body.

The sampleThe earnings call

An hour of unscripted Q&A says a lot about the whole company.

The markersCholesterol, glucose, iron

Specific things a lab can measure the same way every time.

The markersEvasion, hype, stress, hedging

30 markers from 500 yes-or-no checks, measured the same way on every call.

The normal rangeHealthy adults

Your result is compared with thousands of other people.

The normal range174,989 earnings calls

Each marker is compared with the middle half of every call since 2015.

The flagsH or L next to a result

Out of range is not a verdict. It is where to look.

The flagsH or L next to a marker

Flags show where management talked unlike almost everyone else.

The summaryYour doctor's read

One line on what matters most.

The summaryA grade from A to F

One letter, with the reasons, tracked quarter after quarter.

How a call gets tested

Sample in. Lab report out.

Three steps, the same for every company, every quarter. Shown here on Tesla, Inc.'s Q4 2025 call.

“But like, as Elon already mentioned, this year is going to be a huge investment year from a CapEx perspective.”

“On top of it, we will also be spending money for building our AI compute infrastructure and will continue investing in our existing factories to build more capacity.”

“For again showing the confidence in us, let's get ready for a future of amazing abundance.”

hedgeblamehype

STEP 01

Draw the sample

The full transcript lands, usually within a day of the call. Every sentence from management and analysts goes in, prepared remarks and live Q&A.

0 / 500

Did management decline to answer a question?…

STEP 02

Run 500 checks

The same 500 yes-or-no questions on every call, from "was guidance raised" to "did an analyst question go unanswered". This call: 211 yes, 289 no.

Promotion8 / 9

0 · plainnormal 4–69 · sales pitch

▲ Flag H · out of range

STEP 03

Compare to normal

Each marker is placed against the middle half of 174,989 calls. Outside it, the marker gets a flag. The flags roll up into one grade, A to F.

What we measure

Four panels. Thirty markers.

Grouped the way a lab groups its tests. Every marker has a normal range taken from every call we have tested. Here are the ones people ask about most.

Honesty panelDo they answer the question?

Candornormal 6–7 /9
Admits bad news, gives straight answers.
Evasionnormal 2–3 /9
Deflects, restates, answers a different question.
Stonewallsnormal 0
"Not going to comment", "we don't break that out".
Questions left hangingyes / no
An analyst question that never got an answer.

Hype panelAre they selling or reporting?

Promotionnormal 4–6 /9
Pitching the story instead of the numbers.
Superlativesnormal 0.3–1.2
Record, unprecedented, tremendous.
"Excited"normal 0–3
Excited, exciting, thrilled.
AI mentionsnormal 0
AI, LLM, generative, agentic.

Stress panelHow tense is management?

Stressnormal 2–3 /9
Strain and defensiveness in the answers.
Hedgingnormal 5.5–10.5
Sort of, kind of, probably, I think.
Outside forcesnormal 1–9
Macro, tariffs, weather, currency blamed.
Uncertaintynormal 3–5 /9
How unsure the outlook sounds.

Clarity panelCan you check what they said?

Specificitynormal 7–8 /9
Concrete detail instead of generalities.
Numbersnormal 21.2–35.3
Figures per 1,000 words. Low means vague.
Reading levelnormal 7.6–9.6
How hard the answers are to follow.
Script sharenormal 35%–55%
Prepared remarks versus live questions.

Where "normal" comes from

Normal is what 174,989 calls actually look like.

No opinion sets the range. We took every call since 2015 and marked the middle half. 68% of calls score 2–3 on evasion. Anything above that gets flagged, the way a high cholesterol reading does.

Evasion across every earnings call174,989 calls · score 0–9

Tesla, Inc.'s Q4 2025 call scored 4: more evasive than 78% of all calls.

How the grades spreadall tested calls

B+

C+

ACandid, specific, calm. The best 10% of calls.11%
CTypical: some dodging, some hype.19%
FEvasive, promotional or stressed. The bottom 9%.9%

A grade describes the call, not the stock price. It is a check-up, not a forecast. How the grade is built

Fresh from the lab

Latest calls tested.

New results arrive every day in earnings season. The newest quarter is for members; the quarter before is open to everyone.

Latest calls in the labnewest first

Company	Call	Previous grade	New grade
FROFrontline's	Aug 28	B+	Members
BJBJ's Wholesale Club Holdings, Inc.	Aug 28	B	Members
JFINJiayin Group Inc.	Aug 28	D	Members
MNSOMINISO Group Holding Limited	Aug 28	C	Members
HOVHovnanian Enterprises, Inc.	Aug 27	C+	Members
HRLHormel Foods Corporation	Aug 27	C+	Members
ATHMAutohome Inc.	Aug 27	F	Members
AAPAdvance Auto Parts, Inc.	Aug 27	B	Members

Popular lab reportsfree, previous quarter

AAPLApple Inc.NVDANVIDIA Corporation TSLATesla, Inc.MSFTMicrosoft Corporation AMZNAmazon.com, Inc.METAMeta Platforms, Inc.GOOGLAlphabet NFLXNetflix, Inc.AMDAdvanced Micro Devices, Inc.PLTRPalantir Technologies Inc.JPMJPMorgan Chase & Co.COSTCostco Wholesale Corporation

Browse all 9,321 companies →

List no. 15 · September 2026

From lab results to a shortlist: the Monthly Four.

Once a month a fixed rule looks for beaten-down stocks whose latest call came back healthy, and picks four. Every pick is scored against the S&P 500 in public, losers included.

+33% / yrabove the S&P 500, 2019–2025 backtest · worst year −12%

8 of 14live monthly lists ahead of the S&P since Jun 2025

0%median live pick vs the S&P (the average alone would flatter us)

13 / 27live picks ahead of the S&P 500

The Monthly FourSeptember 2026

Fallen stock, healthy call · priced at the Aug 27 close · 1 twelve-month hold + 3 six-month holds

Names lockedPicked 2026-09-01

Ticker locked12-month holdCall Aug 12 · Grade D

Guidance lowered on a specific, unevasive call, demand mixed and margins contracting.

Ticker locked6-month holdCall Aug 19 · Grade C

Guidance held on a measured call, demand accelerating and margins expanding.

Ticker locked6-month holdCall Aug 5 · Grade C+

Guidance held on a measured call, demand accelerating and margins expanding.

Ticker locked6-month holdCall Aug 3 · Grade C

Guidance lowered on a specific, unevasive call, demand accelerating and margins expanding.

Tickers, entry prices and full reports for all four.

Full scorecard and backtest

Backtest: return above the S&P 500long only · no leverage

Model era, 2019–2025: average +33% a year, worst −12%2017–18, before its era: the rule only treads water there

+40%average live pick vs S&P

0%median live pick vs S&P

13 / 27picks ahead

Live scorecard · prices through 2026-08-27bought at the month-end close

Month	Picks	vs S&P
Jun 2025	MU +634%	+634%
Jul 2025	CNC +129%	+129%
Aug 2025	CSTM +82%	+82%
Sep 2025	AEO −18%	−18%
Oct 2025	HRI −4%	−4%
Nov 2025	INSP −64%	−64%
Dec 2025	TXG +285%	+285%
Jan 2026	PII −11%	−11%
Feb 2026	SRPT +16%	+16%
Mar 2026	RIOT +50%DK +37%AMRC −29% ×2	+8%
Apr 2026	PBF +51%CNC +14%UNH −1%ALGN −17%	+12%
May 2026	SRPT +19%STAA −27%AMRC −38%RUN −48%	−24%
Jun 2026	NEOG +27%FLR 0% ×2NWL −7%	+5%
Jul 2026	SPSC +12%CNC +2%PII −8%PBF −8%	−1%
Sep 2026	4 picks locked · this month	entered

Read the shape before you buy. Most picks lose to the index. A few multiply, and they carry the whole list. In a slow 2018-style bear market the rule lost 17% to the index. Months before Mar 2026 had only the twelve-month pick. Lists before this one were generated after the fact by the fixed rule; from October 2026, each month's four are published before their buy price is known.

Questions