What it is

Soundings is not a chatbot with a framework pasted into it, and it is not an analytics product. It is an instrument built around a written body of diagnostic doctrine — three volumes, developed over years and still being extended — with the rules of that doctrine enforced in software rather than requested in a prompt.

The distinction matters more than it sounds. A language model asked politely to be rigorous will be rigorous most of the time and confidently wrong the rest, and it will not tell you which run was which. In Soundings, a finding that does not state what would falsify it cannot be built. A finding withheld for want of evidence must carry its reason or the run fails. These are structural constraints, and a run that violated one would not have completed.

The sweep

Before any of it, a pre‑pass: what does the record say about itself? Where does it contradict itself, what is disputed, what is simply absent. Nothing is diagnosed until the ground the diagnosis will stand on has been characterized.

Then nine stages, in this order.

Five readings, plus conditioning, synthesis, forecast and gate — nine stages in total. All seven dimensions run on every engagement. The available record controls how deep each reading goes; it never controls whether a dimension participates.

Stage 8

The staircase

A destination is reachable if there is a series of stable landings between here and there — intermediate forms the system can rest on without falling back. The measurement that governs everything is the longest unsupported span: the stretch it has to cross with no landing at all.

Read against how much strain the system can currently absorb, that span is the risk. Which produces a result most planning gets backwards —a modest‑looking target with one wide gap in the middle is more dangerous than an ambitious target with good footholds all the way up.

Each candidate destination gets sorted by type. Some are returns to a form the system has held before, which carry an evidentiary advantage the others lack, though the environment that sustained that form may have moved. Some are staged climbs needing more than one cycle, which means there is a condition under which the window closes partway up. Some collapse to a single wide span, where the whole question is one gap. Some are single leaps with no landings at all, which at any real scale is rarely survivable. And some are simply not reachable on the current path — reported as a first‑class result with the specific reason named, because knowing a door is closed is as operationally useful as knowing one is open.

It does not invent destinations. Candidates come from the diagnosis or from you; the instrument characterizes each and declines to rank them by desirability. That restraint is structural rather than modest: a forward reading is the point in the whole apparatus most exposed to the temptation to advocate, because to trace a path is nearly to walk it in the mind, and a mind that has walked a path tends to want it taken. Quietly narrowing the option set to the outcomes the analyst prefers would be advocacy wearing the costume of description — and the danger is precisely that you could not see what had been dropped.

It also checks itself against the two ways forward readings characteristically fail. Mistaking a staircase for a leap, where the landings exist but are not visible from where the system stands, so an ordinary staged climb gets judged heroic or impossible. And the inverse — mistaking a landing for a destination, declaring a transition finished when the system has merely paused partway up. That second error stops the analysis exactly where the most consequential part begins, because a system resting on an intermediate form looks, from outside, like a system that has arrived.

Stage 9

Two patients, same drug

Two patients get the same drug at the same dose for the same diagnosis. One recovers, one deteriorates. The drug is not the explanation.

What a system produces from an intervention depends more on its condition when it receives it than on how well the intervention was designed.

Most planning disciplines skip this entirely. So before any recommendation is issued, the gate audits the receiving system’s capacity to process it — dimension by dimension, producing an explicit profile of where that capacity is robust, partial, depleted, or simply not visible.

Where capacity is depleted, the recommendation arrives in two phases: rebuild the capacity first, deliver the intervention second. Where capacity cannot be assessed at all, that is itself reported, because it means anything you do next is going in blind and the person deciding needs to know that before they commit.

The gate applies the same test to whoever commissioned the work. A true finding delivered to someone who cannot metabolize it fails just as completely as a false one— not as a communication problem, but as a capacity problem at the level of meaning.

What the instrument enforces

These are not aspirations printed in a methodology deck. They are conditions a run has to satisfy in order to exist.

• Every finding names what would prove it wrong. A finding with no falsifier cannot be constructed. If you cannot state what evidence would overturn a claim, it is not a diagnosis; it is a position.
• Silence has to give a reason. Where the evidence will not carry a finding, the finding is suppressed — and a suppression without a stated reason will not construct. You always learn what was not concluded, and why.
• What could not be seen is reported as a result. A failure to retrieve becomes an Opaque finding, carried through the sweep and reported at the end. Opacity is a score, not a gap in the deliverable.
• Confidence is graded, and it propagates. A synthesis carries the standing of the weakest input holding it up, and where the reported grade differs from the derived one, both are kept on the record with their provenance.
• Every run is attributable. Each run records the exact build of the instrument that produced it. A finding from March can be traced to the version of the doctrine and the code that produced it.

What it will not do

• It will not originate an intervention. Characterizing the option space that is reachable is description and stays inside the remit. Selecting one, or arguing for one, is a different act and belongs to a person.
• It will not reason backward from a desired conclusion. No stage is permitted to start from the answer. This is the single most common failure of commissioned organizational analysis, and it is structurally excluded here.
• It will not present an estimate as a measurement. The framework’s highest confidence grade requires a numerical layer this build deliberately does not have, and it is unreachable in code rather than discouraged in prose. The instrument cannot accidentally claim a precision it did not earn.
• It will not fill a hole with something plausible. An absence in the record stays an absence, and is told to you as one.

What comes back

A written diagnostic read. Not a slide deck, not a score, not a maturity model.

Seven readings of the system, each naming the evidence it rests on and the specific finding that would overturn it. Where the record will not carry a reading, a suppression with its reason stated — so you learn what was not concluded and why, rather than simply receiving a shorter document. Where the record went dark, an opacity score carried to the end rather than quietly dropped.

Then the synthesis: an account of how the machinery actually runs, and the step that is rate‑limiting it — the place where, if anything is going to move, movement has to start.

Then the forward read: which configurations are reachable from where the system now stands, the landings between here and each of them, and the longest stretch with no landing at all. Including the destinations that are not reachable on the current path, named with the reason.

And last the capacity profile: whether this system, in its present condition, can absorb what the diagnosis implies — and where it cannot, what has to be rebuilt before anything else is attempted.

Every finding carries its confidence grade, and the grade propagates: a synthesis is held to the standing of the weakest input beneath it. Every run records the build that produced it. What is not in it is a chosen course of action — the instrument characterizes what is reachable and declines to rank by desirability, because selecting is yours.

And what it reads

Whatever you have. Strategy documents, an org chart, board minutes, the post‑mortem nobody acted on, the engagement survey you already ran, a reorganization plan, the last three years of decisions and what happened after each one. Under denied access, the public record and whatever the system displays about itself.

The record controls depth and never participation. All seven dimensions run every time, and where the evidence thins the reading says so instead of stretching to cover the gap.

Two access conditions

Which volume of the framework operates is determined by what can actually be reached, not by analyst preference.

What you can see about a system is shaped by what that system displays. An instrument that searches the open record about an adversarial organization and treats what comes back as neutral evidence has failed before it began — it is reading the cover network and calling it the operating network.

Under denied access the reading is ordered so that perception and bias are examined first, and everything downstream inherits that examination. It is the difference between an assessment of a rival and an assessment of the story a rival is telling about itself.

Judge it yourself

What it produces, in public

Everything above is a claim about an instrument, made by the person who built it. You should not have to take any of it on trust — particularly not from a page that argues findings ought to state what would prove them wrong.

So a number of Soundings diagnoses have been written up and published, free to read. To be exact about what these are: they are essays built from a diagnosis, not the raw run files. The reasoning is visible — what was read, what was concluded, where the record went dark, what would overturn it — while the machinery that produced it stays behind the page. You can judge the reading on its merits, which is the point.

Most of them are denied‑access work on public record: an adversary’s peace plan, a negotiation under scrutiny, an electricity grid, a technology transition read forward from present signals. Subjects that will not open their books to an analyst, which is the harder half of the problem and the one worth watching an instrument attempt.

More of them

No risk, and no catch worth the name

The first one is free

Not a sample. Not an excerpt. Not a scoped‑down version of the real thing.

Bring me the problem and whatever record you have — and whatever you have is enough to start, including a stack of strategy documents and nothing else. I run the diagnosis. You get the read: findings, falsifiers, suppressions, opacity, the mechanism account and its rate‑limiting step, the forward read, the capacity profile.

It is yours. To keep, to circulate, to disagree with, to hand to somebody else entirely. You owe nothing for it — not a fee, not the cost of running it, not a follow‑up conversation you would rather not have.

The argument on this page cannot be settled by more of this page.

Everything above is a claim about an instrument, made by the person who built it. A testimonial would not repair that. A quote from a client I selected is only a second person making my claim for me, and you would still be taking it on trust — on a page whose whole argument is that you should not have to.

A diagnosis of your own system is the one test that settles anything, because you are the only person alive who can tell whether it read you correctly. If it did, we will have something to talk about. If it did not, you will know that for certain, at no cost, which is worth something on its own.

Where it stands today

Soundings runs. It is under active development, and it is operated as an instrument inside an engagement by a practitioner — not sold as a self‑service product, and not left to produce a report unattended.

The framework behind it has been applied for years and across a wide range of systems: startups trying to find footing in a crowded field, large enterprises, private equity and venture firms, and consulting problems for think tanks — from critical infrastructure at the state level to the shape of American electricity a decade out. Much of the business work was done with an earlier, machine‑assisted version of the method; the instrument described on this page is newer, and the published diagnoses above are its own.

Clients are not named here and will not be. What is offered in place of a reference list is the reasoning, in public, and the standing offer above.

The development discipline is the same as the diagnostic one. Every claimed defect is established against the source before it is recorded, on the working assumption that an instrument built to find defects will report its own as the subject’s. Results that fail their own pre‑registered criterion are written down as failures and kept.