I have spent the last few weeks building a tool from the Shiva Svarodaya, a tantric manual on which nostril you are breathing through. Most of that work was the dull kind: indexing verses, computing sunrise, getting the lunar day right.
The decisions I keep thinking about are the other ones. Four things the text teaches are not in the tool. I did not leave them out because they are old, or because they come wrapped in cosmology. I left them out because building each one would have required me to assert something I cannot support — and in two cases, something that could hurt someone.
This is a post about those four. It is probably the most honest thing I can show you about how the tool was made.
The four, before the detail
| What the text teaches | What the tool does | Why |
|---|---|---|
| Forceful retention with abdominal and pelvic pressure | Names it, does not teach it | A web page cannot supervise, and retention measurably moves cerebral blood flow |
| Pressure on the eyeballs during ṣaṇmukhī | Teaches the practice with the eye pressure removed | Nobody has tested it, and the eye is a poor place to find out |
| Reading an element from the shape of a condensation patch | Not implemented at all | The patch is an ambiguous stimulus and the four shapes are imported from iconography |
| Verse 64: refrain from acting when observation contradicts expectation | Displays it, never acts on it | The rule quietly removes the cases that would test the scheme |
The first two are about the body. The second two are about evidence. They are different kinds of refusal and I want to keep them apart.
One: the retention I will not teach
The text is specific about forceful retention, and specific in a way that should give you pause: hold the breath, and apply pressure through the abdomen and the pelvic floor while you do it.
Close the glottis and squeeze against it, and you have performed a Valsalva manoeuvre. Pstras and colleagues describe the physiology of it in Acta Physiologica (2016): intrathoracic pressure rises, venous return to the heart falls, and blood pressure runs through a characteristic sequence of phases, with an overshoot on release. This is ordinary, well-mapped physiology. It is also why people are told not to strain against a closed airway if they have certain heart or eye conditions.
I want to be careful about the step I am taking here, because it is the kind of step this blog exists to catch. Nobody has put a pressure line into someone performing this particular practice with these particular locks and published the phases. That retention with abdominal and pelvic pressure reproduces Valsalva haemodynamics is a mechanical inference, not a measured result. I think it is a sound inference. It is still an inference.
What has been measured is narrower and more interesting. Nivethitha and colleagues, in the Journal of Neurosciences in Rural Practice (2016), put transcranial Doppler on fifteen healthy male practitioners and watched the middle cerebral artery during one minute of bhastrika and one minute of breath-holding. The two did almost opposite things.
Note what that study does not tell you. It did not measure blood pressure. It did not measure carbon dioxide. The authors say so plainly, and they describe their own explanation — that holding the breath raises CO₂ and lowers oxygen — as a hypothesis they could not confirm. Fifteen trained men is a small, selected sample. None of this says retention is dangerous. It says retention is not inert, and that the body goes somewhere while you do it.
On actual harm, the evidence is thinner than people on either side of this argument tend to admit. Cramer, Krucoff and Dobos reviewed every published case report and case series they could find in PLoS ONE (2013): 35 case reports and 2 case series, 76 cases in total. Breathing practices were the single most frequently implicated practice, which sounds alarming until you see the number — four cases, three of them involving forceful techniques. One of those was a pneumothorax in a man practising kapālabhāti, reported by Johnson and Tierney in CHEST (2004).
Here is a claim I had been making, and have stopped making. I had been saying that adverse events are “not rare enough to ignore.” I cannot support that. Cramer and colleagues are explicit that case reports cannot be used to estimate how often anything happens, because there is no denominator — no count of the people who practised and were fine. The honest statement is that nobody knows the rate, and that case reports are the weakest design for finding out. What they establish is that the injury is possible, not that it is common.
So my reason for leaving retention out is not that I have evidence it will hurt you. It is narrower. The practice has a plausible mechanism for harm, a measurable physiological effect, a handful of documented injuries, and no way for me to see your heart condition, your blood pressure or your face through a web page. The tradition’s answer to that was a teacher in the room. I think that was risk management rather than gatekeeping. A static page is not a teacher in the room, so the tool names the retention practices, explains what they are, and does not walk you through one.
Two: the eye pressure I took out
Ṣaṇmukhī is in the tool. The instruction to press on your eyeballs is not.
I will be straightforward: I could not find a study of digital pressure on the eye during this practice. Not a good one, not a bad one. As far as I can tell nobody has measured it.
The nearest relevant evidence is about something else. Jasien, Jonas, de Moraes and Ritch measured intraocular pressure during four common yoga postures in PLoS ONE (2015), with ten people with open-angle glaucoma and ten without. In downward dog, pressure went from about 17 mmHg to about 28. In a standing forward bend, 17 to 27. The rise appeared within a minute of taking the position and came back down within about two minutes of sitting up. The glaucoma and non-glaucoma eyes behaved much the same.
That study is not about pressing on your eyes, and I am not going to pretend it is. What it shows is a general property: intraocular pressure is mobile, it moves a long way on ordinary postural changes, and it moves fast. The authors are careful themselves — they say their results neither warn glaucoma patients off these poses nor rule out harm, and they call for the work that would actually settle it.
Which leaves me with an absence of evidence rather than evidence of harm. Normally that is a reason to stay neutral. I did not stay neutral here, for one reason: the eye is not an organ where I am willing to let an untested instruction ride on the assumption that it is probably fine. The practice works without that step. So the tool teaches it without that step, and says why.
Three: the shape I will not read
Breathe onto a mirror and you get two patches of condensation, one under each nostril. Comparing them for size is a crude but real measurement — it is roughly a test of which side is moving more air, and the tool includes it, graded as weak.
The text goes further. It says you can read which element is active from the shape of the patch: square, crescent, triangle, circle. That is not in the tool, and will not be.
Two problems, and the second is the serious one. The first is that those four shapes are yantra forms. They come from iconography, from the diagrams used to represent the elements in ritual, and there is no reason a patch of water vapour on glass should organise itself into them. The categories arrived from somewhere other than the mirror.
The second is that a condensation patch is almost a laboratory-grade ambiguous stimulus. It is irregular, it has soft edges, it changes as you look at it, and it does not strongly resemble anything. Ask someone to sort that into four shapes they have just been told to expect and you are not measuring the patch. You are measuring the expectation.
That is the claim I would like to support with a clean experiment, and here the evidence is weaker than I want it to be. Williams and Blagrove tested a version of it in PLoS ONE (2022): sixty-one people listened to short, degraded audio clips and said whether they heard a voice, after being told either that the study was about ghost voices or about voices in noisy rooms. The priming did not raise overall detection rates — no main effect of group, and no increase in false alarms in white noise. What it did was shift the pattern: primed listeners reported more voices in the purportedly paranormal clips than in the merely degraded ones, while unprimed listeners treated the two the same. The authors describe this as only a partial replication of the original effect, and their sample was small and local.
So I cannot tell you that experiment proves expectation writes the answer. It does not. It is a partial, qualified result, and the general literature here is messier than the confident version you will read elsewhere.
My reason for leaving the shape-reading out does not actually rest on that study. It rests on something simpler. A measurement is only worth having if two careful people would get the same answer from the same patch, and I do not believe they would. One incidental detail in the Williams and Blagrove data makes the point better than their main analysis: when listeners wrote down what they thought the voices said, under 1% of their guesses matched what the original investigators had heard in the same clips. Everyone heard something. Almost nobody heard the same thing.
Four: the rule that cannot be wrong
This is the one I find most interesting, and it is not a safety question at all.
Verse 64 says that when what you observe contradicts what the scheme predicted, you should refrain from acting. Read charitably, that is humility — do not proceed on a confused reading. Read as a method, it is the most effective way I know of to make a claim immune to evidence.
Follow what happens to the cases.
When observation and prediction agree, you act, something happens, and the case is available to be remembered. When they disagree, you do not act, nothing happens, and there is no outcome to remember. The disagreements are precisely the cases that would have told you whether the prediction works, and the rule disposes of them before they can.
There is a second mechanism layered on top. A contrary observation does not get recorded as “the scheme was wrong today.” It gets reinterpreted as an inauspicious condition — the conditions were not right, so the rule did not apply. A claim that absorbs its own counterexamples that way is not a weak claim. It is not a claim at all, in the sense that it no longer forbids any observation.
I want to separate two things that are easy to run together. This is not evidence that the scheme is false. It is the reason I cannot find out either way from the tradition’s own record, however long that record is. A thousand years of satisfied practitioners is not a thousand years of evidence if the dissatisfied cases were never allowed to count.
So the tool shows verse 64, labels it, and never acts on it. If what you observe disagrees with what the calculation expected, the tool tells you they disagree and lets you carry on. That is the only version of this I can defend: record the disagreement rather than explain it away.
What this costs
Leaving things out has a price, and I would rather state it than pretend otherwise.
The tool is less impressive than it could be. A version that read elements off a mirror and told you which hours to avoid would be far more compelling to use, and I expect some people will find what I have built timid — a manual stripped of exactly the parts that made it a manual.
I also cannot claim my refusals are themselves well-evidenced. Two of them rest on absence of evidence rather than evidence of harm, and I have tried to say so each time rather than dressing a judgement call up as a finding. Someone could reasonably draw the line in a different place.
What I am not willing to do is build an instrument whose readings cannot be wrong, or teach a practice I cannot supervise, and then let the tool’s general air of carefulness lend credibility to either. The text itself warns about that move — mixing verifiable and unverifiable items into one composite, so the good items carry the bad. I think it is right about that, and I have tried to apply it to the text more consistently than the text applies it to itself.
The parts I kept are the parts you could check. That is a smaller tool. I think it is the only honest one.
Sources
- Pstras and colleagues. The Valsalva manoeuvre: physiology and clinical examples. Acta Physiologica (2016). PubMed 26662857
- Nivethitha L, Mooventhan A, Manjunath NK, Bathala L, Sharma VK. Cerebrovascular hemodynamics during pranayama techniques. Journal of Neurosciences in Rural Practice (2016), 8(1):60–63. PubMed 28149083
- Cramer H, Krucoff C, Dobos G. Adverse events associated with yoga: a systematic review of published case reports and case series. PLoS ONE (2013), 8(10):e75515. PubMed 24146758
- Johnson and Tierney. Kapalabhati pranayama: breath of fire or cause of pneumothorax? A case report. CHEST (2004). PubMed 15136413
- Jasien JV, Jonas JB, de Moraes CG, Ritch R. Intraocular pressure rise in subjects with and without glaucoma during four common yoga positions. PLoS ONE (2015), 10(12):e0144505. journals.plos.org
- Williams JM, Blagrove M. Paranormal experiences, sensory-processing sensitivity, and the priming of pareidolia. PLoS ONE (2022), 17(9):e0274595. journals.plos.org