Specification

Specification and limits

Method rev. 1.0 · October 2026 · Issued by Warfighter Genetics

This is the whole method behind every readout: the 24 positions we look for in your file, what each is mapped to, how strong the evidence is, what we say, what we won't say, and where each claim comes from. It's open to anyone, with no account and no email. If you find a mistake, write to support@warfightergenetics.com and we'll log the correction in the changelog.

Positions: 24
Grade B: 8
Grade C: 16
Sources: 43, each with a DOI
23andMe v5: 22 of 24 documented
AncestryDNA v2: 21 documented, 2 likely, 1 not on chip

How to read

How to read the table

One row per position. Every row is printed in full, so you can read it, print it, or check it against the papers.

Gene and rsID: The gene the position sits in or near, and its public identifier. Your raw file lists the same rsID.
Variant: The name studies use for it, such as Val158Met.
Chip coverage: Whether each provider's chip is documented to test the position. Yes: it appears in public records of real files from that chip. Likely: inferred from related chips, not confirmed. Unknown: no basis either way. No: not on that chip. Coverage tells you what to expect. Your pre-flight check tells you what your file actually has.
Grade: How strong the evidence is for the claim that puts the position on the dial. Definitions below.
Mapped to: The training stressors the position is tied to. These decide where it sits on the dial for each pipeline.
What we say: The reading for each possible genotype, as your file would show it, with the label studies use. Your readout prints only the one that matches your file.
What we won't say: Printed word for word on your readout, next to the reading.
How we read the letters: Consumer files report the plus strand of the genome; many papers report the other strand, so the same genotype can look different in a study. We store the letters your file uses and state, for every position, how they map to the letters in the papers.
Sources: Each with a DOI link to the paper.

Positions

Positions · Method rev. 1.0 · 24 rows
Gene	rsID	Variant	Coverage	Grade	Mapped to	What we say	What we won't say	How we read the letters	Sources
ACTN3	rs1815739	R577X	23andMe v5 yes · AncestryDNA v2 yes · MyHeritage likely · FamilyTreeDNA likely	Grade C	Cold water, soft-tissue load	CC · RR Both copies make alpha-actinin-3, a protein in fast-twitch muscle fibers. In a meta-analysis of 90 studies, R was more common among power athletes than among controls; endurance athletes did not differ from controls. CT · RX One working copy, so your muscle makes alpha-actinin-3. RX is the most common genotype in people of European and East Asian ancestry; RR is the most common in African ancestry. TT · XX Neither copy makes alpha-actinin-3. In one cold-water immersion study of 42 young men, XX held core temperature better than RR, by raising muscle tone rather than shivering. XX is less common among power athletes, and a 2026 systematic review of small studies linked it to more muscle and soft tissue injury.	That you are a 'power' or 'endurance' type, that you will tolerate cold water better or get injured more, that this genotype decides any event, or anything about your chance of finishing selection.	C = R (577Arg, makes alpha-actinin-3); T = X (577 stop codon, no alpha-actinin-3). ACTN3 is on the + strand, so file letters match the literature. The reference base differs by genome build (GRCh37 T = X, GRCh38 C = R), so whole-genome VCF calls are converted from REF/ALT to letters; 0/0 is never read as RR.	Chelly et al. 2025, J Strength Cond Res (meta-analysis, 90 studies) · doi:10.1519/JSC.0000000000005192 Wyckelsma et al. 2021, Am J Hum Genet · doi:10.1016/j.ajhg.2021.01.013 Abboud et al. 2026, Genes (systematic review) · doi:10.3390/genes17020212
ACE	rs4343	I/D proxy (tags the Alu insertion/deletion)	23andMe v5 yes · AncestryDNA v2 yes · MyHeritage likely · FamilyTreeDNA likely	Grade C	Running volume, loaded rucking	AA · I/I (by proxy) AA at rs4343 tags the insertion on both copies. The I allele goes with lower ACE activity. In one study of British Army recruits, I/I improved more than D/D on a repeated arm-lift endurance test after 10 weeks of training, and I was more common among high-altitude mountaineers. Many later studies found no association. This proxy was validated in European-ancestry samples only. AG · I/D (by proxy) One insertion and one deletion copy. This is the most common genotype in European-ancestry samples and sits between I/I and D/D in the studies cited. This proxy was validated in European-ancestry samples only. GG · D/D (by proxy) GG at rs4343 tags the deletion on both copies, which goes with higher ACE activity. D was more common among top swimmers in events of 400 m and under (not in longer events), and D/D recruits improved less than I/I on the arm-endurance test in the Army study. Many later studies found no association. This proxy was validated in European-ancestry samples only.	That you suit or don't suit any pipeline, how you will respond to training volume, or anything about blood pressure or heart health. Effects are small and many studies found no association.	Consumer chips cannot read the 287-bp Alu insertion/deletion itself (rs1799752). rs4343 tags it in Europeans (D' = 1, r2 = 0.88, n = 64): A marks I (insertion), G marks D (deletion). The tag has been validated only in European-ancestry samples and is not a reliable I/D proxy in African ancestry: G is 20% in 1000 Genomes Africans against 57% in Europeans, and the validating paper notes that rs4343 and I/D track plasma ACE differently in Africans. ACE is on the + strand.	Montgomery et al. 1998, Nature · doi:10.1038/30374 Woods et al. 2001, Hum Genet · doi:10.1007/s004390100466 Abdollahi et al. 2008, Dis Markers (rs4343 as I/D proxy) · doi:10.1155/2008/813679
PPARGC1A	rs8192678	Gly482Ser	23andMe v5 yes · AncestryDNA v2 yes · MyHeritage likely · FamilyTreeDNA likely	Grade C	Running volume	CC · Gly/Gly Both copies carry Gly482. In one study, Ser482 was less common among 104 world-class Spanish endurance athletes than among 100 older, unfit UK controls (29% vs 40% of alleles), so Gly/Gly sits on the side seen more often in those athletes. CT · Gly/Ser One copy of each. The study cited compared allele counts, so it gives no separate reading for heterozygotes. TT · Ser/Ser Both copies carry Ser482, the allele that was less common among world-class endurance athletes in one study. The samples were small and the finding has not been consistently replicated.	Anything about your VO2max, diabetes or metabolic health, or that this genotype caps your aerobic fitness.	PPARGC1A is on the - strand. The gene-strand G (Gly482) / A (Ser482) appear in consumer files as C (Gly) / T (Ser).	Lucia et al. 2005, J Appl Physiol · doi:10.1152/japplphysiol.00037.2005
COL1A1	rs1800012	Sp1 binding site (S/s)	23andMe v5 yes · AncestryDNA v2 likely · MyHeritage unknown · FamilyTreeDNA unknown	Grade C	Soft-tissue load	AA · s/s (study TT) Both copies carry s, a rare genotype. It was under-represented among South Africans with ACL ruptures in one study, but a 2025 meta-analysis of athletes found no link to tendon or ligament injury. AC · S/s (study GT) One copy of s. The studies cited found no clear difference between S/s and S/S. CC · S/S (study GG) Both copies are the common S form, the reference genotype in the studies cited.	That your ligaments or tendons are strong or weak, anything about bone density or osteoporosis, or your chance of a specific injury.	COL1A1 is on the - strand. Studies report G (S) / T (s) on the gene strand; consumer files show C (= G, S) and A (= T, s). A file 'AA' is the study's 'TT'.	Posthumus et al. 2009, Br J Sports Med · doi:10.1136/bjsm.2008.056150 Fukuyama et al. 2025, Int J Sports Med (meta-analysis, athletes) · doi:10.1055/a-2419-4359 Ryan-Moore et al. 2020, Front Genet (meta-analysis, active people) · doi:10.3389/fgene.2020.00551
COL5A1	rs12722	BstUI 3' UTR variant	23andMe v5 yes · AncestryDNA v2 likely · MyHeritage unknown · FamilyTreeDNA unknown	Grade C	Soft-tissue load, running volume	CC · C/C Both copies C. In Australian and South African case-control samples, CC carried lower odds of chronic Achilles tendinopathy than other genotypes (odds ratio 0.42 in the Australian sample). A 2025 meta-analysis limited to athletes found no association. CT · C/T One copy of each. The studies cited compared CC or TT against the other genotypes, so there is no separate reading for C/T. TT · T/T Both copies T. In pooled studies of mostly European-ancestry cohorts, TT carried about 1.6 times the odds of Achilles, ACL and elbow tendon injuries. A 2025 meta-analysis limited to athletes found no association.	That you will get a tendon or ligament injury, or anything about joint hypermobility conditions.	COL5A1 is on the + strand; file letters match the literature (C/T).	September et al. 2009, Br J Sports Med · doi:10.1136/bjsm.2008.048793 Lv et al. 2018, Oncotarget (meta-analysis) · doi:10.18632/oncotarget.23805 Fukuyama et al. 2025, Int J Sports Med (meta-analysis, athletes) · doi:10.1055/a-2419-4359
MMP3	rs679620	Glu45Lys	23andMe v5 yes · AncestryDNA v2 yes · MyHeritage likely · FamilyTreeDNA likely	Grade C	Soft-tissue load, running volume	CC · Glu/Glu (study GG) In one case-control study, this genotype had about 2.5 times the odds of Achilles tendinopathy, and risk rose further together with the COL5A1 rs12722 T allele. A 2025 meta-analysis of athletes found no association. CT · Glu/Lys (study AG) One copy of each. Only the study's GG genotype (CC in your file) was associated with tendinopathy. TT · Lys/Lys (study AA) Not associated with tendinopathy in the studies cited.	That your tendons will fail, or anything about joint disease.	MMP3 is on the - strand. Studies report G (Glu45) / A (Lys45) on the gene strand; consumer files show C (= G) and T (= A). A file 'CC' is the study's 'GG'. SNPedia displays this SNP on the gene strand (A/G).	Raleigh et al. 2009, Br J Sports Med · doi:10.1136/bjsm.2008.053892 Fukuyama et al. 2025, Int J Sports Med (meta-analysis, athletes) · doi:10.1055/a-2419-4359
GDF5	rs143383	+104T/C 5' UTR variant	23andMe v5 no · AncestryDNA v2 yes · MyHeritage unknown · FamilyTreeDNA unknown	Grade C	Soft-tissue load, running volume	AA · T/T (study) In Australian and South African case-control samples combined, TT had about 1.8 times the odds of Achilles tendon pathology; on its own, only the Australian sample showed the effect. A later study found no link to ACL rupture. AG · T/C (study) Not associated with Achilles pathology in the study cited. GG · C/C (study) Not associated with Achilles pathology in the study cited.	Anything about osteoarthritis or joint disease, or that you will injure an Achilles tendon.	GDF5 is on the - strand. Studies report T/C on the gene strand; consumer files show A (= T) and G (= C). A file 'AA' is the study's 'TT'. Frequencies agree: T is about 60% in European-ancestry controls and A is 63% in 1000 Genomes Europeans. The reference base differs by genome build (GRCh37 A, GRCh38 G); convert VCF REF/ALT to letters.	Posthumus et al. 2010, Rheumatology · doi:10.1093/rheumatology/keq072 Raleigh et al. 2013, Int J Sports Med · doi:10.1055/s-0032-1316361
GC	rs2282679	Vitamin D binding protein (lead GWAS variant)	23andMe v5 yes · AncestryDNA v2 yes · MyHeritage likely · FamilyTreeDNA likely	Grade B	Bone load	GG · G/G Both copies carry G. In a 2026 study of 331 Polish soldiers, GG marked the subgroup with the lowest 25(OH)D. This raises the odds of low vitamin D; it does not measure it. GT · G/T One copy of G, the allele associated with lower blood 25(OH)D in a genome-wide study of 33,996 people. TT · T/T Both copies carry the allele associated with higher blood 25(OH)D. Season, sun exposure, skin tone and diet still matter more than this genotype.	Your actual vitamin D level (only a blood test shows that), that you will get a stress fracture (vitamin D variants did not track bone changes in 2,550 Army trainees), or any supplement dose.	GC is on the - strand. SNPedia and some papers use the gene-strand letters A/C; consumer files show T (= A) and G (= C). G is the allele associated with lower blood 25(OH)D.	Wang et al. 2010, Lancet (GWAS, 33,996 people) · doi:10.1016/S0140-6736(10)60588-0 Gazdzinska et al. 2026, Nutrients (331 soldiers) · doi:10.3390/nu18183041 Roberts et al. 2026, J Appl Physiol (2,550 Army trainees) · doi:10.1152/japplphysiol.00181.2026
CYP2R1	rs10741657	Vitamin D 25-hydroxylase region	23andMe v5 yes · AncestryDNA v2 yes · MyHeritage likely · FamilyTreeDNA likely	Grade B	Bone load	AA · A/A Both copies carry the allele associated with higher blood 25(OH)D. The effect per copy is small. AG · A/G One copy of each; a small net effect on vitamin D. GG · G/G Both copies carry the allele associated with lower 25(OH)D in genome-wide studies of up to 79,366 people. On its own the effect is small.	Your vitamin D level, your fracture risk, or any supplement dose.	Reported on the + strand in the GWAS and in consumer files. A is associated with higher 25(OH)D, G with lower.	Wang et al. 2010, Lancet (GWAS, 33,996 people) · doi:10.1016/S0140-6736(10)60588-0 Jiang et al. 2018, Nat Commun (GWAS, 79,366 people) · doi:10.1038/s41467-017-02662-2 Fernandez-Lazaro et al. 2025, Int J Mol Sci (CrossFit athletes) · doi:10.3390/ijms26125602
DHCR7/NADSYN1	rs12785878	7-dehydrocholesterol reductase region	23andMe v5 yes · AncestryDNA v2 yes · MyHeritage likely · FamilyTreeDNA likely	Grade B	Bone load	GG · G/G Both copies carry the allele associated with lower 25(OH)D in genome-wide studies. DHCR7 sits in the pathway the skin uses to make vitamin D. On its own the effect is small. GT · G/T One copy of each; a small net effect on vitamin D. TT · T/T Both copies carry the allele associated with higher 25(OH)D. This is the most common genotype in Europeans.	Your vitamin D level, your fracture risk, or any supplement dose.	Reported on the + strand in the GWAS and in consumer files. T is associated with higher 25(OH)D, G with lower.	Wang et al. 2010, Lancet (GWAS, 33,996 people) · doi:10.1016/S0140-6736(10)60588-0 Jiang et al. 2018, Nat Commun (GWAS, 79,366 people) · doi:10.1038/s41467-017-02662-2
WNT16	rs2707466	Thr/Ile missense variant	23andMe v5 yes · AncestryDNA v2 yes · MyHeritage likely · FamilyTreeDNA likely	Grade B	Bone load	CC · Thr/Thr Both copies carry the allele associated with thinner cortical bone (about 0.1 standard deviations per copy) and lower forearm bone density in a genome-wide study of about 5,800 people. CT · Thr/Ile One copy of each; intermediate. TT · Ile/Ile Both copies carry the allele associated with thicker cortical bone and higher bone density.	That you will get a stress fracture, or anything about osteoporosis. Bone density can only be measured with a scan.	WNT16 is on the + strand; file letters match the GWAS. C (Thr) is associated with thinner cortical bone and lower forearm bone density, T (Ile) with thicker and higher. SNPedia displays this SNP on the opposite strand (A/G); consumer files and the GWAS use C/T.	Zheng et al. 2012, PLoS Genet · doi:10.1371/journal.pgen.1002745
LRP5	rs3736228	Ala1330Val	23andMe v5 yes · AncestryDNA v2 yes · MyHeritage likely · FamilyTreeDNA likely	Grade B	Bone load	CC · Ala/Ala The common genotype and the reference in the studies cited. CT · Ala/Val One copy of Val1330, associated with slightly lower spine and hip bone density (about 14 mg/cm2 per copy at the spine) in 37,534 adults from 18 cohorts. TT · Val/Val Two copies of Val1330, so the small bone density effect applies twice. In physically active people, pooled studies found no link between this variant and fractures.	That you will get a stress fracture, or anything about osteoporosis.	LRP5 is on the + strand; file letters match the literature. C = Ala1330, T = Val1330.	van Meurs et al. 2008, JAMA (37,534 people) · doi:10.1001/jama.299.11.1277 Ryan-Moore et al. 2020, Front Genet (meta-analysis, active people) · doi:10.3389/fgene.2020.00551
P2RX7	rs3751143	Glu496Ala (loss of function)	23andMe v5 yes · AncestryDNA v2 yes · MyHeritage likely · FamilyTreeDNA likely	Grade C	Bone load	AA · Glu/Glu Both copies make a normally working receptor. In one study, female endurance runners with AA had about 4% higher total-body bone density than other genotypes. AC · Glu/Ala One loss-of-function copy. In one report, the Ala496 allele was associated with stress fracture in 210 Israeli military conscripts (43 with stress fractures) and again in 518 athletes. No independent group has repeated it. CC · Ala/Ala Two loss-of-function copies. Same association as above, with few homozygotes studied.	That you will or won't get a stress fracture.	P2RX7 is on the + strand; file letters match the literature. A = Glu496 (normal receptor), C = Ala496 (loss of function). SNPedia displays this SNP on the opposite strand (G/T); consumer files show A/C.	Varley et al. 2016, Purinergic Signal (military conscripts and athletes) · doi:10.1007/s11302-016-9495-6 Herbert et al. 2022, Eur J Appl Physiol (endurance runners) · doi:10.1007/s00421-021-04789-z
P2RX7	rs1718119	Ala348Thr (gain of function)	23andMe v5 yes · AncestryDNA v2 yes · MyHeritage likely · FamilyTreeDNA likely	Grade C	Bone load	AA · Thr/Thr Two gain-of-function copies. The Thr348 allele was associated with fewer stress fractures in Israeli conscripts and fewer repeat stress fractures in athletes. Few homozygotes were studied. AG · Ala/Thr One gain-of-function copy; same association as above, in one study. GG · Ala/Ala The common genotype and the reference in the study cited.	That you are protected from stress fractures.	P2RX7 is on the + strand; file letters match the literature. G = Ala348, A = Thr348 (gain of function). SNPedia displays this SNP on the opposite strand (C/T); consumer files show G/A.	Varley et al. 2016, Purinergic Signal (military conscripts and athletes) · doi:10.1007/s11302-016-9495-6
TMPRSS6	rs855791	Ala736Val	23andMe v5 yes · AncestryDNA v2 yes · MyHeritage likely · FamilyTreeDNA likely	Grade B	Running volume, loaded rucking	AA · Val/Val (study TT) Two copies of Val736. Each copy was associated with lower serum iron (about 0.18 standard deviations), lower transferrin saturation and slightly lower hemoglobin in Australian and Dutch cohorts. AG · Ala/Val (study CT) One copy of Val736, so about half the small effect on iron and hemoglobin described for Val/Val. GG · Ala/Ala (study CC) Both copies carry the allele associated with higher serum iron and transferrin saturation.	That you are anemic or iron deficient (only blood tests show that), anything about iron overload, or any supplement dose.	TMPRSS6 is on the - strand. Studies report T (Val736) / C (Ala736) on the gene strand; consumer files show A (= T, Val) and G (= C, Ala). A file 'AA' is the study's 'TT'. The literature numbers this residue 736 (A736V); Ensembl's canonical transcript numbers it 727.	Benyamin et al. 2009, Nat Genet · doi:10.1038/ng.456
CYP1A2	rs762551	-163C>A (*1F)	23andMe v5 yes · AncestryDNA v2 yes · MyHeritage likely · FamilyTreeDNA likely	Grade C	Caffeine response	AA · A/A The higher-inducibility genotype: in the original study, A/A smokers metabolized caffeine fastest; in non-smokers the genotypes did not differ. In 101 competitive male athletes, caffeine improved 10-km cycling time only in AA. A 2024 meta-analysis of 13 studies found the same direction, but the genotype difference disappeared when studies with reported conflicts of interest were left out. AC · A/C Intermediate. In the 101-athlete study, caffeine had no effect on 10-km cycling time in AC; the 2024 meta-analysis found a small benefit. CC · C/C The lower-inducibility genotype. In the 101-athlete study, the higher caffeine dose made 10-km cycling time worse in a small CC group. Few CC carriers have been studied (34 across the 13 studies in the 2024 meta-analysis), and that result depended on studies with reported conflicts of interest.	That caffeine will or won't help you, how much to take, or anything about heart health.	CYP1A2 is on the + strand; file letters match the literature. A is the higher-inducibility allele often called 'fast'.	Sachse et al. 1999, Br J Clin Pharmacol · doi:10.1046/j.1365-2125.1999.00898.x Guest et al. 2018, Med Sci Sports Exerc (101 athletes) · doi:10.1249/MSS.0000000000001596 Barreto et al. 2024, Med Sci Sports Exerc (meta-analysis, 13 studies) · doi:10.1249/MSS.0000000000003313
ADORA2A	rs5751876	1976T>C (synonymous)	23andMe v5 yes · AncestryDNA v2 no · MyHeritage unknown · FamilyTreeDNA unknown	Grade B	Caffeine response	CC · C/C Not the TT genotype linked to caffeine-induced anxiety. C/C was more common among adults who say caffeine disturbs their sleep, and a twin study of 2,402 adults found a weaker version of the same link: the C allele went with more caffeine-related insomnia. CT · C/T One copy of each. C carriers reported less caffeine-induced anxiety than TT; in a twin study, the odds of caffeine-related insomnia rose with each C copy. TT · T/T In controlled studies (379 people in the largest), TT was more likely to feel anxious after a moderate caffeine dose. Regular users built substantial tolerance. In the sleep studies cited, the T allele went with less caffeine-related sleep disturbance.	Anything about anxiety disorders or mental health, or how much caffeine to take.	ADORA2A is on the + strand; file letters match the literature (T/C). Older papers number the same SNP differently (1976T>C, 1083T>C).	Rogers et al. 2010, Neuropsychopharmacology · doi:10.1038/npp.2010.71 Childs et al. 2008, Neuropsychopharmacology · doi:10.1038/npp.2008.17 Retey et al. 2007, Clin Pharmacol Ther · doi:10.1038/sj.clpt.6100102 Byrne et al. 2012, Sleep (2,402 twins, caffeine-related insomnia) · doi:10.5665/sleep.1962
AHR	rs4410790	Habitual caffeine intake locus (about 54 kb upstream of AHR)	23andMe v5 yes · AncestryDNA v2 yes · MyHeritage likely · FamilyTreeDNA likely	Grade B	Caffeine response	CC · C/C Both copies carry the allele associated with drinking slightly more caffeine: about 44 mg a day more than TT in a meta-analysis of more than 36,000 people. The effect is small. CT · C/T Intermediate; a very small effect on habitual intake. TT · T/T Both copies carry the allele associated with slightly lower habitual caffeine intake.	How much caffeine you should use. This locus explains under 1% of the difference in intake between people.	Reported on the + strand in the GWAS and in consumer files (T = 38% in the GWAS, 39% in 1000 Genomes Europeans). C is associated with higher habitual caffeine intake, T with lower.	Cornelis et al. 2011, PLoS Genet (GWAS meta-analysis) · doi:10.1371/journal.pgen.1002033
COMT	rs4680	Val158Met	23andMe v5 yes · AncestryDNA v2 yes · MyHeritage likely · FamilyTreeDNA likely	Grade C	Sleep loss	AA · Met/Met Met/Met breaks down prefrontal dopamine most slowly. In one small lab study of 66 adults (32 kept awake for 38 or 62 hours, only 5 of them Met/Met), Met/Met kept its ability to adapt when task rules reversed better than Val carriers. AG · Val/Met Intermediate enzyme activity. Val carriers as a group lost more of their ability to adapt when task rules reversed during total sleep deprivation; rested, there was no difference. GG · Val/Val Val/Val breaks down prefrontal dopamine fastest. In the sleep-loss study, Val carriers lost more of their ability to adapt when task rules reversed; rested, there was no difference.	Anything about intelligence, personality, stress 'type' or mental health. We don't use 'warrior gene' labels, and this does not predict selection.	G = Val, A = Met on the + strand as reported by 23andMe/Ancestry. COMT is on the + strand, so the literature uses the same letters.	Satterfield et al. 2018, Cortex · doi:10.1016/j.cortex.2017.11.012
TNF	rs1800629	-308 G>A promoter variant	23andMe v5 yes · AncestryDNA v2 yes · MyHeritage likely · FamilyTreeDNA likely	Grade C	Sleep loss	AA · A/A Two copies of -308A, a rare genotype. Only one A/A person was in the 88-person study, analysed together with G/A, so there is no separate reading for A/A. AG · G/A One copy of -308A. In 88 adults kept awake in the lab, A carriers lost less reaction-time vigilance during total sleep deprivation, and a follow-up study of 12 people (4 A carriers) repeated the result. Both studies came from the same research group. The genotype explained under 10% of the difference between people. GG · G/G The common genotype. In the same lab studies, GG showed the typical decline in reaction-time vigilance during total sleep deprivation.	That you can safely go without sleep, or anything about inflammatory disease.	TNF is on the + strand; file letters match the literature (G common, A less common).	Satterfield et al. 2015, Brain Behav Immun · doi:10.1016/j.bbi.2014.12.009 Skeiky et al. 2020, Chronobiol Int · doi:10.1080/07420528.2020.1821044
ADA	rs73598374	Asp8Asn (c.22G>A)	23andMe v5 yes · AncestryDNA v2 yes · MyHeritage likely · FamilyTreeDNA likely	Grade C	Sleep loss, recovery	CC · G/G (study) The common genotype, with typical adenosine breakdown. It is the reference group in the sleep studies cited. CT · G/A (study) One copy of the slower Asn8 enzyme. Carriers slept more deeply and, in matched lab groups that included sleep deprivation, reported more sleepiness and fatigue and had lower attention than G/G. Both studies came from one research group. TT · A/A (study) Two copies of the slower enzyme, a rare genotype. The studies cited compared G/A with G/G, so there is no separate reading for A/A.	Anything about sleep disorders or immune conditions.	ADA is on the - strand. Studies report G (Asp8) / A (Asn8) on the gene strand; consumer files show C (= G) and T (= A). A file 'CT' is the study's 'G/A'.	Retey et al. 2005, Proc Natl Acad Sci USA · doi:10.1073/pnas.0505414102 Bachmann et al. 2012, Cereb Cortex · doi:10.1093/cercor/bhr173
BDNF	rs6265	Val66Met	23andMe v5 yes · AncestryDNA v2 yes · MyHeritage likely · FamilyTreeDNA likely	Grade C	Sleep loss, recovery	CC · Val/Val Typical activity-dependent BDNF release. In a small lab study (22 people), Val/Val had more deep slow-wave sleep than Val/Met at baseline and after 40 hours awake. CT · Val/Met Met reduces activity-dependent BDNF release. In one small study, Val/Met showed less training-driven change in motor cortex responses to brain stimulation; in another (22 people), Val/Met had less deep sleep than Val/Val. These are lab measures of brain activity and sleep, not of performance. TT · Met/Met Two Met copies. The studies cited grouped Met/Met with Val/Met or did not include it, so there is no separate reading.	Anything about intelligence, memory, mood or mental health.	BDNF is on the - strand. The gene-strand G (Val66) / A (Met66) appear in consumer files as C (Val) / T (Met).	Kleim et al. 2006, Nat Neurosci · doi:10.1038/nn1699 Bachmann et al. 2012, Sleep · doi:10.5665/sleep.1690
FKBP5	rs1360780	Intronic regulatory variant	23andMe v5 yes · AncestryDNA v2 yes · MyHeritage likely · FamilyTreeDNA likely	Grade C	Acute stress	CC · C/C The common genotype. Cortisol returned to baseline normally after a standard lab stress test. CT · C/T One copy of T; heterozygotes were similar to C/C in the study cited. TT · T/T Two copies of the minor allele. In one lab study of 64 healthy adults, the few T/T showed incomplete recovery of cortisol after a social stress test.	Anything about PTSD, depression or mental health risk, or how you will handle stress. This is a stress-hormone marker only.	FKBP5 is on the - strand, but the literature, dbSNP and SNPedia name this intronic SNP by its + strand letters (C/T), the same letters consumer files report. T is the minor allele studied in the literature (about 31% in 1000 Genomes Europeans).	Ising et al. 2008, Eur J Neurosci · doi:10.1111/j.1460-9568.2008.06332.x
NR3C1	rs41423247	BclI (glucocorticoid receptor)	23andMe v5 no · AncestryDNA v2 yes · MyHeritage unknown · FamilyTreeDNA unknown	Grade C	Acute stress	CC · G/G (study) In men, BclI G/G showed a smaller cortisol response to a standard social stress test in two studies from one research group (112 and 206 people). In women using oral contraceptives the direction reversed. CG · C/G (study) Carriers of one BclI G copy showed small or no differences in cortisol response to the lab stress test. GG · C/C (study) The common genotype and the reference in the studies cited.	Anything about mental health, depression or PTSD, or how you will handle stress.	A C/G SNP, so the letters alone cannot show strand. NR3C1 is on the - strand: the studies' BclI 'G' allele appears as C on the + strand. Frequencies confirm it: C is the minor allele (38% in 1000 Genomes Europeans), as BclI G is in the literature. A file 'CC' is the studies' 'GG'. Papers are not consistent: some name this minor allele 'C' (+ strand letters). The studies cited use the gene-strand 'G' for the minor allele (18 of 112 men were GG in the first), which is file C.	Wust et al. 2004, J Clin Endocrinol Metab · doi:10.1210/jc.2003-031148 Kumsta et al. 2007, Biol Psychiatry · doi:10.1016/j.biopsych.2007.04.013

Grades

Evidence grades

A grade describes the claim we print, not everything ever published on the gene. It has to hold for the link that puts the position on the dial: the stressor. When the evidence sits between two grades, we grade down.

Grade	Definition	Rev. 1.0
Grade A	The stressor-linked claim is replicated in independent large cohorts, including military or athlete populations, with an effect that isn't trivial.	0 positions
Grade B	Replicated in independent cohorts, but the effect is modest, or the evidence comes only from the general population or one ancestry.	8 positions
Grade C	A single study, a single research group, small samples, a proxy of limited accuracy, or replication that conflicts or comes up empty in athletes or military samples.	16 positions

Positions without at least one peer-reviewed paper with a DOI are left out. A stressor tag needs at least one cited paper that measured something related to that stressor; otherwise the tag goes.

On the dial, tick length is the grade. With no grade A in rev. 1.0, no tick is long.

Limits

It can't tell you whether you'll finish, and nobody can.
It does not predict injury. Readings about tendon, ligament and bone describe associations in groups of people.
Most effects are small. 16 of the 24 positions are grade C: one study, one research group, small samples or mixed results.
Much of the evidence comes from European-ancestry samples. The ACE reading relies on rs4343 as a stand-in for an insertion/deletion that chips can't read, and that stand-in was validated only in European-ancestry samples. Every ACE reading says so.
It reads only what your chip tested, and consumer files can contain errors. If a result matters to a health decision, confirm it with a clinical test your physician orders.
It reads your genotype, not your physiology. A genotype linked to lower vitamin D or lower iron is not a vitamin D or iron level. Only a blood test shows that.
Some stressors have no position on our list. Nothing clears the bar for breath-hold, heat or swimming volume in rev. 1.0, so those weeks carry no tick.
Research changes. Grades and readings can change in a later revision, and each change goes in the changelog. Your readout keeps the revision it was built with, printed on every page.

Not in your readout

We don't report these, at any price, and nothing in the readout hints at them:

"Warrior gene" reads of COMT, or any personality read: aggression, empathy, leadership, team fit
Intelligence, memory or learning ability
Mental health, including PTSD risk, depression and anxiety disorders
Alzheimer's risk, APOE status or any other disease risk
Night vision, navigation aptitude, spatial reasoning or motion sickness
Supplement or medication doses, or advice to start or stop either
A score, a pass/fail, a "fit" rating or any forecast of selection
Your ancestry

Changelog

Rev. 1.0 · October 2026

First published method. What this rebuild removed, and why.

Before October 2026 this site sold a different product under the same name. Rev. 1.0 replaces it. These are the claims and features we removed, and the reason for each.

Removed

The "warrior gene". The old site put the "warrior gene" label on COMT Val/Val and said it improves performance under extreme stress. The evidence we reviewed doesn't support that, and the label has no place in a method. COMT stays on the list for one finding only, about adapting to rule changes during total sleep deprivation in a small lab study, at grade C.
PTSD risk. The old hero promised to show your "PTSD risk", and the report had a "PTSD Resilience" domain built on FKBP5. A consumer readout shouldn't forecast a mental health condition, and the evidence doesn't allow it. FKBP5 stays only as a stress-hormone marker from one lab study, at grade C, with "anything about PTSD" on its won't-claim line.
Personality reads. Oxytocin receptor "empathy", "aggression modulation", "team-oriented" dopamine, "social pain", "emotional stability" and leadership reads from OXTR, DRD2, OPRM1, TPH2 and DBH. None has evidence tying it to selection, and personality is out of scope.
Supplement dosing. "SUPPLEMENT NOW", "Must supplement" and "non-negotiable" lines with doses of methylfolate, B vitamins, omega-3 and vitamin D, hung on MTHFR, IL6 and vitamin D genotypes. We aren't clinicians, and doses don't belong in a genetic readout. Where a reading touches vitamin D or iron, it now says to consider a blood test and to discuss it with your physician.
Borrowed endorsements. Tiles that read "Research Partner USARIEM", "Validated by SOCOM", "Community NSW" and "Published in Physiology & Behavior", and badges that read HIPAA compliant and lab validated. We had no agreement with any of those organizations, HIPAA doesn't apply to this service, and there is no lab in it. The initials-only testimonials went for the same reason: we couldn't stand behind them.
Disease-risk reports. A disease-risk report type, including APOE (Alzheimer's) and "longevity" reads. Disease risk needs clinical genetic counseling, and none of it bears on selection.
Forecasts. The "Overall SOF Genetic Fit 88/100" score, a GO badge on every phase, and an "SOF Minimum Threshold" line. The old site said it didn't forecast selection while printing all three.
Pipeline "genetic factors". Night vision adaptation, spatial reasoning, motion sickness, navigation aptitude, hypoxia tolerance and "cold tolerance markers", listed against pipelines without sources.
IL6 rs1800795. On the draft list until the evidence review. Neither paper behind it measured recovery or tissue load, so its dial ticks had no support.
A misquoted statistic. The old report said "82% of sleep-resilient soldiers" in one place and "82% of sleep-vulnerable" in another, citing "ScienceDirect".
Raw-file upload and the AI report writer. The old service accepted your whole file on our server and had a language model write report text. Your file now stays on your device, and readouts are built by fixed rules from this table.
"Burn After Reading." "No storage. No database. No trail" and "100% data privacy" weren't true: reports were stored. We now say what we keep and for how long, and deleting it is a button that returns a receipt.

Corrected

Marker count: "35+ SNPs" and "50–100" became 24 positions, each listed here with its sources.
BUD/S: 24 weeks with three 8-week phases became 28 weeks (NSWO 7, then three 7-week phases), a 9-week land warfare phase became 7 weeks, and SQT went from 26 weeks to 19 (NETC 2023). Attrition, given as 75%, 70–85% and 70–80% on different pages, became 68% attrition and rolls since 1998 (NETC 2023).
SFAS: a 14-day assessment plus 10-day selection split, which no current official source supports, became three event-based weeks. Attrition 50–70% became 65% (NETC 2023), with its scope caveat.
RASP 1: attrition 50–60% became 47% (NETC 2023).
MARSOC A&S: 19 days became about 4.5 weeks of events (MARSOC letter to the candidate). Attrition 50–60% became 73% enlisted, 59% officer (NETC 2023).
PJ / CCT: "Special Warfare Prep (8 weeks)" became the 7-week Special Warfare Candidate Course, which replaced it in 2022. The 80–90% PJ figure is gone: we found no official source for it.
160th SOAR Green Platoon: attrition 40–50% became 34% enlisted, 7% officer (NETC 2023).
The case study is now AAR-001, labelled an illustrative case study. It no longer says every injury was preventable.

Added

This table: 24 positions, each with a grade, its sources, what we say and what we won't.
The pre-flight check, which reads your file on your device and reports the bytes it sends.
A report-date window: no sale within 6 weeks of your date, or more than 24 months before it.
Deletion with a receipt.
Pipeline facts from official sources, with the disagreements between sources stated.

Known gaps in rev. 1.0

No position for breath-hold, heat or swimming volume.
Coverage for MyHeritage and FamilyTreeDNA files is inferred, and Living DNA isn't charted in the table yet. None of the three has been checked against real files. The pre-flight check reads all three and shows your real count.
24 positions, below the 25 to 35 we aimed for. We didn't pad the list to reach it.

Marker list approved by Warfighter Genetics on 7 October 2026.

Sources

Every paper behind a reading, alphabetical, with the positions it supports. DOIs link to the paper.

Background, not evidence for any reading

Petrovick et al. 2026, Physiology & Behavior · doi:10.1016/j.physbeh.2025.115150. A study of genetic markers in 800 male SFAS candidates. Its abstract names genes, not the specific variants we read, so it supports no reading in rev. 1.0. We cite it as a source. Its authors and their institutions don't work with us or endorse us.

Pipeline sources are listed under each pipeline on the pipelines page.

Marker evidence (43)

Abboud et al. 2026, Genes (systematic review). doi:10.3390/genes17020212 ACTN3 rs1815739
Abdollahi et al. 2008, Dis Markers (rs4343 as I/D proxy). doi:10.1155/2008/813679 ACE rs4343
Bachmann et al. 2012, Cereb Cortex. doi:10.1093/cercor/bhr173 ADA rs73598374
Bachmann et al. 2012, Sleep. doi:10.5665/sleep.1690 BDNF rs6265
Barreto et al. 2024, Med Sci Sports Exerc (meta-analysis, 13 studies). doi:10.1249/MSS.0000000000003313 CYP1A2 rs762551
Benyamin et al. 2009, Nat Genet. doi:10.1038/ng.456 TMPRSS6 rs855791
Byrne et al. 2012, Sleep (2,402 twins, caffeine-related insomnia). doi:10.5665/sleep.1962 ADORA2A rs5751876
Chelly et al. 2025, J Strength Cond Res (meta-analysis, 90 studies). doi:10.1519/JSC.0000000000005192 ACTN3 rs1815739
Childs et al. 2008, Neuropsychopharmacology. doi:10.1038/npp.2008.17 ADORA2A rs5751876
Cornelis et al. 2011, PLoS Genet (GWAS meta-analysis). doi:10.1371/journal.pgen.1002033 AHR rs4410790
Fernandez-Lazaro et al. 2025, Int J Mol Sci (CrossFit athletes). doi:10.3390/ijms26125602 CYP2R1 rs10741657
Fukuyama et al. 2025, Int J Sports Med (meta-analysis, athletes). doi:10.1055/a-2419-4359 COL1A1 rs1800012, COL5A1 rs12722, MMP3 rs679620
Gazdzinska et al. 2026, Nutrients (331 soldiers). doi:10.3390/nu18183041 GC rs2282679
Guest et al. 2018, Med Sci Sports Exerc (101 athletes). doi:10.1249/MSS.0000000000001596 CYP1A2 rs762551
Herbert et al. 2022, Eur J Appl Physiol (endurance runners). doi:10.1007/s00421-021-04789-z P2RX7 rs3751143
Ising et al. 2008, Eur J Neurosci. doi:10.1111/j.1460-9568.2008.06332.x FKBP5 rs1360780
Jiang et al. 2018, Nat Commun (GWAS, 79,366 people). doi:10.1038/s41467-017-02662-2 CYP2R1 rs10741657, DHCR7/NADSYN1 rs12785878
Kleim et al. 2006, Nat Neurosci. doi:10.1038/nn1699 BDNF rs6265
Kumsta et al. 2007, Biol Psychiatry. doi:10.1016/j.biopsych.2007.04.013 NR3C1 rs41423247
Lucia et al. 2005, J Appl Physiol. doi:10.1152/japplphysiol.00037.2005 PPARGC1A rs8192678
Lv et al. 2018, Oncotarget (meta-analysis). doi:10.18632/oncotarget.23805 COL5A1 rs12722
Montgomery et al. 1998, Nature. doi:10.1038/30374 ACE rs4343
Posthumus et al. 2009, Br J Sports Med. doi:10.1136/bjsm.2008.056150 COL1A1 rs1800012
Posthumus et al. 2010, Rheumatology. doi:10.1093/rheumatology/keq072 GDF5 rs143383
Raleigh et al. 2009, Br J Sports Med. doi:10.1136/bjsm.2008.053892 MMP3 rs679620
Raleigh et al. 2013, Int J Sports Med. doi:10.1055/s-0032-1316361 GDF5 rs143383
Retey et al. 2005, Proc Natl Acad Sci USA. doi:10.1073/pnas.0505414102 ADA rs73598374
Retey et al. 2007, Clin Pharmacol Ther. doi:10.1038/sj.clpt.6100102 ADORA2A rs5751876
Roberts et al. 2026, J Appl Physiol (2,550 Army trainees). doi:10.1152/japplphysiol.00181.2026 GC rs2282679
Rogers et al. 2010, Neuropsychopharmacology. doi:10.1038/npp.2010.71 ADORA2A rs5751876
Ryan-Moore et al. 2020, Front Genet (meta-analysis, active people). doi:10.3389/fgene.2020.00551 COL1A1 rs1800012, LRP5 rs3736228
Sachse et al. 1999, Br J Clin Pharmacol. doi:10.1046/j.1365-2125.1999.00898.x CYP1A2 rs762551
Satterfield et al. 2015, Brain Behav Immun. doi:10.1016/j.bbi.2014.12.009 TNF rs1800629
Satterfield et al. 2018, Cortex. doi:10.1016/j.cortex.2017.11.012 COMT rs4680
September et al. 2009, Br J Sports Med. doi:10.1136/bjsm.2008.048793 COL5A1 rs12722
Skeiky et al. 2020, Chronobiol Int. doi:10.1080/07420528.2020.1821044 TNF rs1800629
van Meurs et al. 2008, JAMA (37,534 people). doi:10.1001/jama.299.11.1277 LRP5 rs3736228
Varley et al. 2016, Purinergic Signal (military conscripts and athletes). doi:10.1007/s11302-016-9495-6 P2RX7 rs3751143, P2RX7 rs1718119
Wang et al. 2010, Lancet (GWAS, 33,996 people). doi:10.1016/S0140-6736(10)60588-0 GC rs2282679, CYP2R1 rs10741657, DHCR7/NADSYN1 rs12785878
Woods et al. 2001, Hum Genet. doi:10.1007/s004390100466 ACE rs4343
Wust et al. 2004, J Clin Endocrinol Metab. doi:10.1210/jc.2003-031148 NR3C1 rs41423247
Wyckelsma et al. 2021, Am J Hum Genet. doi:10.1016/j.ajhg.2021.01.013 ACTN3 rs1815739
Zheng et al. 2012, PLoS Genet. doi:10.1371/journal.pgen.1002745 WNT16 rs2707466