Introverted Sensing (Si): The Stored Baseline

Introverted Sensing keeps a detailed impression of how things have been, and runs the present against it continuously. That is the entire function. It is not a memory for facts and it is not a fondness for the old ways — it is a comparator, always on, reporting one thing: this does not match. Everything Si users are known for, including the parts that annoy people, comes out of that single loop.

The mechanism, stated precisely

Si stores a rich impression of a prior experience and compares current input against it. The stored thing is not a fact you could recite. It is closer to a texture: how the room was laid out, what order the steps went in, what the engine sounded like, how your stomach felt an hour afterwards. And it is not filed away to be retrieved on demand — it sits in the background as a reference signal, so the output of the function is almost never a recollection. It is a discrepancy.

That ordering explains the thing outsiders find strange about Si users: the discomfort arrives before the words do. The comparator fires on a mismatch first; locating what the mismatch is against comes second, and sometimes takes days. If you have ever watched someone insist something is wrong while being unable to say what, and then be correct three weeks later, you have watched this run.

Watch it run: the thing nobody else noticed

Three items in that exchange are the whole function. The first is that the alarm came in as a feeling with no content attached. The second is the search for the referent — not “I remember a sound”, but a hunt for a specific past occasion clean enough to compare against, which is why the radio detail matters: it is the condition that made the earlier impression usable. The third is that the finding is datable. Hold onto that; it is what separates Si from Ni later on.

It works on people too, and this is where dominant Si earns its keep. A team lead with heavy Si does not read a colleague’s mood from the content of their messages. She reads it from the shape: Dan used to reply in two lines within ten minutes and put an exclamation mark on the confirmation. For eleven days he has answered in one word, some hours later, no punctuation. The messages say nothing is wrong. The delta says otherwise, and the delta is what she trusts, because the delta is the only thing she is actually measuring.

It is not “good memory”

This is the most common mistaught version and it produces mistypes in both directions. Plenty of strong Si users are poor at names, hopeless at trivia, and cannot tell you what they read last month. Semantic recall is a different faculty. What Si holds is experiential and comparative, and the tell is simple: ask an Si user to describe last year’s version of the menu and you get a vague half-list. Put the new menu in their hands and they will tell you within seconds that three dishes are gone, the portions look smaller and the prices moved. An archive answers questions. A comparator only speaks when something is put next to it.

So people whose recall is genuinely excellent sometimes type themselves Si-dominant on the strength of a party trick, while people running heavy Si rule it out because they lose their keys. Both readings come from testing the wrong faculty. The question is not what do you remember. It is how quickly do you notice that something has changed, and how small can the change be.

And it is not “likes tradition”

Tradition is a downstream consequence, and treating it as the function gets the causation exactly backwards. Si does not prefer old things; it prefers tested things, because a tested thing has a stored result attached and an untested one does not. The stored result is the asset. Age is irrelevant.

You can watch this break the stereotype in a week. An ISTJ who switched note-taking tools three years ago will now defend the new tool more fiercely than the person who talked them into it, because they have run it several hundred times and the result is stored. The same person will refuse a different tool that is objectively better on paper, because paper is not a result. What reads as conservatism is a switching-cost calculation: known outcome versus unknown outcome, with the cost of relearning charged to the unknown side. When the payoff of a change is concrete and the scope is bounded, Si users adopt fast and then become the change’s loudest defender.

Si in all four slots

Position in the stack is authority, not skill. First-slot Si gets the first word on every experience before anything else evaluates it. Fourth-slot Si is not absent — it is unmonitored, which is a different and more expensive problem.

Dominant — ISTJ (Si-Te-Fi-Ne) and ISFJ (Si-Fe-Ti-Ne)

The baseline is the first thing consulted, so a new situation is experienced as a comparison rather than as an event. Both types share Ne in the inferior slot, which is why the anxiety in both runs the same shape: an unbounded field of what-ifs about a change that has not happened yet.

What differs is the second function, and it decides what the baseline is of. The ISTJ runs Te second, so the stored reference is procedural — the sequence, the standard, the version that worked, the way the numbers normally sit at this point in the month. The ISFJ runs Fe second, so the reference is relational: who normally sits where, who has gone quiet, what this person is like when they are fine. Same comparator, two different libraries. Give them a strained family dinner and the ISTJ notices the schedule slipped; the ISFJ noticed twenty minutes earlier that one person had stopped joining in.

Auxiliary — ESTJ (Te-Si-Ne-Fi) and ESFJ (Fe-Si-Ne-Ti)

Second-slot Si does not set the agenda; it supplies precedent to a dominant that is already pushing outward. The dominant decides what to do and Si supplies the reference class: how this went last time, what we already know breaks, what the normal version of this looks like.

In an ESTJ that reads as institutional memory with executive force behind it — a fast “no, we changed that in the spring, use the current form” delivered before anyone else has finished reading the document. In an ESFJ it reads as social precedent: remembering that these two do not sit together, that this one always brings a plate, that last year’s version of this event ran forty minutes over and everybody got cold. Auxiliary Si is consulted. Dominant Si is running. The test is whether the comparison still happens when nobody is asking you to do anything.

Tertiary — INTP (Ti-Ne-Si-Fe) and INFP (Fi-Ne-Si-Te)

Third-slot Si is real, private, and mostly emotional in flavour rather than procedural. It shows up as an unusually vivid attachment to specific past states — a particular flat, a particular six months, a particular album played to death at nineteen — and as comfort loops: the same four meals, the same route, the same show rewatched for the eleventh time because a known experience costs nothing to process.

The loop has a real function. Ne generates possibility at a rate that becomes exhausting, and tertiary Si is where an Ne-heavy inner life goes to be finite for an evening. It becomes a problem when it turns retrospective: the stored good period gets compared against the present and the present loses, every time, because the stored version has been smoothed. In our typing study — 1,179 conversations after de-duplication — 42.3% came back INTP and 23.3% INFP, so two thirds of that sample runs Si in exactly this slot. If you have ever wondered why so much of that population writes about a specific year like it was a country they were deported from, this is the machinery.

Inferior — ENTP (Ne-Ti-Fe-Si) and ENFP (Ne-Fi-Te-Si)

Fourth-slot Si means the comparator is not running in the background, and the consequences are almost entirely about maintenance. Anything whose trigger is a schedule rather than an event simply does not surface: the dentist, the oil change, the prescription refill, the filter, the annual renewal. It is not forgetfulness — nothing generated the reminder, because generating the reminder is the job of a function tracking the normal interval between things.

The second consequence is worse and gets covered in the next section. With no running baseline, accumulation is invisible: five nights of six hours’ sleep, three coffees and no lunch do not register as a trend. They register as an event, all at once, when a threshold is crossed.

Si and the body

Si-led types tend to have a far more accurate read on their own physical state than anyone else, and it is the same mechanism with the body as the input stream. Because there is a stored normal for what rested feels like, what hungry feels like, and what the day before getting ill feels like, deviations get reported with resolution. Not “I feel bad” but “I am not ill, I slept badly and this is the second day of it, and it will lift after a proper meal.” They will call a cold twelve hours before the first symptom — not by prediction, but because the current sensation matched a stored one and the match came with a filed outcome.

The same accuracy is why Si-led health anxiety, when it appears, is so specific. A stored bad episode makes an ordinary sensation feel like the return of that episode, and the comparator reports a match with total confidence. Vague dread is not the Si version. “This is the same pain as the week before the hospital” is.

Si-inferior types run the opposite failure. The signal is not missing — the comparator is. Nothing is keeping a running total, so an ENTP or ENFP can work through the sixth consecutive short night with genuinely no sense of a slope, right up until the body becomes the only thing they can perceive. And then the second half of the problem lands: with no baseline, the sensation is unprecedented by definition, so every symptom presents as a first-time emergency. This is the 2am pattern — a heartbeat suddenly audible, a mole that was probably always there, an hour of searching, a resolve to see somebody, and by lunchtime it is gone and Ne has moved on.

Si vs Ni, from the Si side

This is the discriminator worth getting right, because both functions produce quiet certainty from a source the owner did not consciously assemble, and both get described as “a gut feeling.” The difference is whether the source can be retrieved.

Si’s reference is traceable. Ni’s is not. Ask an Si user why they are uneasy and, given a minute, you get an episode: when it was, roughly, what was different about it, what happened afterwards. Ask an Ni user the same question and you get a direction — this is heading somewhere, it ends badly, that person is going to leave — with no retrievable source underneath it, and often a reconstruction assembled afterwards because the question demanded one. That reconstruction is not dishonest. It is a genuine attempt to source something that never arrived with a receipt.

Two mistypes come straight out of this. ISTJ and INTJ (Si-Te-Fi-Ne against Ni-Te-Fi-Se) share the middle of the stack and both read as careful, certain and blunt about it; ask each why they are sure and the ISTJ produces the precedent while the INTJ produces the trajectory and gets impatient at being asked to cite anything. ISFJ and INFJ (Si-Fe-Ti-Ne against Ni-Fe-Ti-Se) split the same way over people: the ISFJ’s read is built from this specific person’s history — she does not do this — while the INFJ’s is a read of where the person is going that does not come from a comparison at all. The full derivation of why those pairs sit so close is in the cognitive functions guide.

The failure mode: a baseline that outlived its conditions

Si stores results. It does not store expiry dates, and that is the entire failure mode.

A concrete one. A team writes a release checklist in 2021 after a bad outage, and step four is a manual database snapshot before every deploy. Twenty minutes, every time. Eighteen months later the conditions that made step four necessary — no automated backups, a single primary, no way to roll back — have all been fixed. The step survives anyway, for four years, because it is individually cheap and nobody wants to be the person who deleted the backup step the week before an outage. The error was never the step. It is that the step was stored as a rule while the reasons behind it were stored nowhere, so there is no way to notice that the rule is now solving a problem that no longer exists.

The personal version is the same shape and costs more: I do not do well in big groups, formed at nineteen in one genuinely bad year, still running at thirty-four for a person with a different job, different friends and fifteen more years of practice. The experiment was real. It has simply never been re-run.

A re-baselining practice

1. Date the rule and name three conditions. Write the year it formed and the three things that were true then that made it true. If you cannot name the conditions, what you have is a habit wearing a finding’s clothes.
2. Apply the two-of-three test. If two of those three conditions have changed, this is not the same experiment any more — it is a new one, and you have not run it. That is not an argument that the rule is wrong. It is an argument that you currently have no evidence either way, which is a different and much more honest position than “we tried that.”
3. Re-test one rule a quarter, bounded. Scope and end date up front: four weeks, staging only, reverting on the first real problem. Si takes bounded experiments far better than open-ended change, because a bounded experiment terminates in a fresh stored result — which is the only currency the function actually accepts.
4. Say the date out loud. Make the full sentence “we tried that in 2021, when we had two engineers and no backups.” Said in full, it survives when it is still valid and collapses on its own when it is not, which saves everyone the argument.

What Si leaves behind in writing

Functions are legible in text because they are habits of construction rather than opinions. Si leaves comparative framing everywhere: usually, normally, it used to, same as last time, that is not like him. It leaves specificity that serves no argumentative purpose — the day of the week, the weather, where everyone was sitting — because those details came bundled with the impression and are not experienced as decoration. And it leaves sequence: instructions written in the order the steps actually occur, including the step everyone else omits because it is obvious.

That is the case for scoring how someone writes rather than asking them. “Do you value tradition?” is a question about self-image and collects nothing useful, while a pass over your real messages can count how often you answered a question by comparing the present to a dated past state. If you would rather start somewhere lighter, take the free MBTI test first and then check the result against your own outbox.