Metacognition is the bandwidth connecting every node.
It is the ability to observe my own thinking while I am thinking—to inspect not only the question in front of me, but why I noticed it, what I might want the answer to be, where my experience creates insight, and where that same experience could introduce bias.
The Watchman question is a useful way to show the mechanism.
Not because the clinical question has been settled.
Not because I am attempting to settle it.
It is useful because it shows how vigilance moves: from an unresolved signal to a bounded question, through evidence and incentives, into a model that must survive correction before it earns the right to say anything at all.
This is the wiring:
Lived experience → signal detection → unresolved contradiction → disciplined inquiry → incentive mapping → economic modeling → adversarial review → human judgment
Node 0: The objective function
Before the signal, there is the reason I am listening for it.
Nearly immediately after the rhythm of my own heart was quieted, I began searching for answers.
I wanted to understand psychological distress in people living with arrhythmia. I knew what the physiological experience had done to me—the uncertainty, the disruption, the constant awareness that something inside my body could change without asking permission.
What I did not yet understand was how many people carried a similar psychological burden.
Then I found the literature.
Physiologically, I had experienced an isolating and unusually complex course.
Psychologically, I was one of millions.
That realization changed me.
It changed what the vigilance needed to be applied toward.
I became extraordinarily driven to serve people who had experienced cardiac disease in the ways that I had. Not because lived experience made me a clinician. Not because surviving something granted me authority over everyone else who survived it differently.
It gave me a reason to pay attention.
Lived experience gives me standing to ask the question. It does not give me authority to predetermine the answer.
Node 1: The original signal
By April, I was nearing completion of a project designed to build capacity for peer support among cardiac-arrhythmia patients, caregivers, and professionals.
The need was evident to me.
People living with high-burden arrhythmia often occupy a strange space. Their disease can be episodic, invisible, difficult to explain, and psychologically consuming. The people around them may care deeply and still be unable to understand what it is like to negotiate with an electrical system inside your own body.
Peer support cannot eliminate that burden.
It can make it less solitary.
That project was where I expected my attention to remain.
Then another signal appeared.
The electrophysiology and cardiology communities maintain a thriving public conversation online. During that period, I began following a debate surrounding two randomized trials of left atrial appendage closure: CLOSURE-AF and CHAMPION-AF.
It was not a casual disagreement.
The studies appeared to be moving the conversation in different directions. Clinical interpretations diverged. The populations differed. The comparators differed. The endpoints required careful separation.
In CLOSURE-AF, physician-directed medical therapy performed better on the primary composite outcome than left atrial appendage closure among patients at high risk of both stroke and bleeding. The lead investigator subsequently said the findings were likely to constrain his own use of the procedure in that population. The trial population and results are summarized by the American College of Cardiology, and his comments were reported by TCTMD.
CHAMPION-AF examined a different population: patients who were suitable for long-term anticoagulation. The trial met its prespecified primary and secondary endpoints, and Boston Scientific publicly framed the results as supporting WATCHMAN FLX as a potential first-line option. The trial was published in the New England Journal of Medicine, while the company used first-line language in its Q1 2026 investor materials.
One investigator was discussing narrower use after one trial.
A manufacturer was discussing broader use after another.
I could not reconcile the direction of travel.
Node 2: The discrepancy
The two trials are not interchangeable.
That needs to be stated plainly.
CLOSURE-AF enrolled an older population at high risk of stroke and bleeding and compared closure against physician-directed medical therapy. CHAMPION-AF enrolled patients suitable for anticoagulation and compared WATCHMAN FLX with modern non–vitamin K antagonist oral anticoagulants.
Different populations can produce different answers.
Different comparators can produce different answers.
Different endpoints can produce different answers.
The divergence was therefore not proof that one study invalidated the other. Nor was it proof of wrongdoing, clinical failure, or an unsafe device.
It was a signal that patient selection mattered enormously—at the same moment the commercial conversation was moving toward a substantially larger first-line market.
That was the contradiction.
A cardiac patient does not need uncertainty concealed—or amplified for effect.
A cardiac patient needs uncertainty named accurately.
What is known?
What remains contested?
Which population does the evidence actually describe?
What trade-offs are being combined inside the endpoint?
What happens when the treatment moves from selected use toward population-scale adoption?
That final question moved the inquiry into territory where I could contribute.
Node 3: Establish the boundary
This is where metacognition becomes essential.
Before asking what the evidence means, I have to establish what I am qualified to claim.
I am not an electrophysiologist.
I am not making an individual treatment recommendation.
I am not arguing that one therapy is clinically superior for every patient.
Clinical interpretation belongs to clinicians, trialists, methodologists, guideline bodies, regulators, and—at the point of care—to patients working with professionals who know their histories.
My professional experience helps me understand healthcare markets and institutional incentives.
My lived experience helps me understand the human consequence of cardiac uncertainty.
Neither converts me into a clinician.
The boundary of my inquiry is economic.
More specifically:
If left atrial appendage closure is moving toward broader first-line use, do historical economic claims remain valid after the clinical population, evidence base, comparator price, and potential scale of adoption have changed?
That is a bounded question.
It is falsifiable.
And answering it does not require pretending to answer the separate clinical debate.
Node 4: Evidence ingestion
Vigilance requires diligence.
I read the trials.
Then I read the reviews and editorials surrounding them.
I watched the clinical discussion continue.
I examined the component outcomes rather than relying only on composite headlines.
I reviewed Medicare coverage, procedural payments, drug prices, survival assumptions, post-procedure treatment exposure, and older economic models.
Then another variable entered the frame.
Medicare-negotiated prices for Eliquis and Xarelto took effect on January 1, 2026. CMS established negotiated 30-day prices of $231 for Eliquis and $197 for Xarelto, compared with the 2023 list-price figures of $521 and $517 used in the agency’s comparison. CMS estimated that applying the negotiated prices across the first ten selected drugs to 2023 utilization would have reduced aggregate net covered prescription-drug spending by approximately 22%. Those figures come directly from CMS.
That did not prove that left atrial appendage closure lacked economic value.
It proved that a major economic input had changed.
Many historical cost-effectiveness arguments depended, in part, on avoiding years of comparatively expensive anticoagulant therapy. When the recurring cost of the comparator changes materially, the previous break-even calculation cannot simply be carried forward.
The question became unavoidable:
Does the old economic conclusion survive the new price environment?
I did not know.
That is why I built the model.
Node 5: Incentive mapping
Before modeling the economics, I needed to map the system.
Not to identify malignment.
To identify objective functions.
Every participant has a legitimate interest:
- Patients want protection from stroke, relief from bleeding risk, freedom from burdensome treatment, affordability, and confidence in the decision.
- Clinicians want better outcomes for their patients. Their judgment is also shaped by training, specialty, experience, patient selection, and the outcomes they encounter directly.
- Hospitals want to provide advanced care. They also operate procedural programs with facilities, staff, equipment, capacity targets, and financial requirements.
- Device manufacturers have obligations to patients and regulators. They also have products, growth targets, market strategies, employees, investors, and shareholders.
- Drug manufacturers benefit when medication remains the standard treatment.
- Medicare and other payers must consider individual access, population outcomes, and total program spending.
- Researchers and professional societies must interpret incomplete evidence while new evidence continues to arrive.
None of these interests automatically invalidates anyone’s position.
But they do not align perfectly.
A hospital can make a rational decision for its service line.
A manufacturer can make a rational decision for its shareholders.
A procedural specialist can make a sincere recommendation based on patients seen in practice.
A payer can make a rational decision for its budget.
A patient can make a rational decision based on fear, convenience, cost, or prior experience.
Every participant can act rationally inside their own node while the total system drifts away from the outcome it claims to optimize.
That is the danger of local optimization.
The system becomes opaque without anyone necessarily choosing opacity.
This is why I do not begin by asking who is wrong.
I ask:
What is each participant optimizing, and who carries the loss when those objectives diverge?
In this system, the patient carries the irreversible consequence.
That makes the patient outcome the primary variable.
Node 6: Metacognition ignition
This is the moment when gathering information becomes watching the inquiry itself.
What do I believe?
Why do I believe it?
What evidence would change my mind?
Am I weighting the ischemic-stroke signal differently because of my own cardiac history?
Am I undervaluing bleeding because it has not been the defining threat in my life?
Am I assuming institutional incentives prove institutional misconduct?
Am I treating a changed drug price as sufficient to decide an entire cost-effectiveness question?
Am I giving the device enough time for its up-front cost to be recovered?
Am I selecting assumptions because they test the question—or because they support the concern?
This is metacognition as bandwidth.
It allows information to move between lived experience, professional knowledge, quantitative analysis, and self-critique without allowing any one of those sources to seize control of the system.
The objective is not to remove perspective.
That is impossible.
The objective is to make perspective visible enough that it can be challenged.
Node 7: The model
I built a Medicare-perspective economic analysis.
The model asks what happens economically when contemporary clinical outcomes, current procedural payments, negotiated anticoagulant prices, treatment exposure, utilities, survival, and time horizons are placed in the same frame.
I gave the device the long horizon its economic case requires.
I allowed competing utility architectures to produce different answers.
I tested lower anticoagulant exposure.
I tested shorter survival.
I separated short-horizon findings from long-horizon speculation.
And when internal review found that one result had been labeled incorrectly, I corrected it—even though the correction weakened the original headline.
Vigilance is not loyalty to the first conclusion.
It is loyalty to the integrity of the process.
The analysis is being rebuilt after adversarial and consensus review and will be sent to clinical and health-economic experts. Its conclusions remain open to correction.
The work does not claim to have settled whether WATCHMAN should be used more broadly.
The narrower economic claim is already clear:
Evaluations built on the previous anticoagulant-price environment are no longer decision-current merely because they were valid when published.
The revised answer remains open.
The need for revision does not.
Node 8: Adversarial review
A model becomes dangerous when its complexity is mistaken for authority.
Code can reproduce an error perfectly.
A polished table can make an assumption disappear.
A precise number can create confidence the evidence does not deserve.
That is why the next node is adversarial review.
Another model attacks the architecture.
Another checks the calculations.
The assumptions are stressed.
The labels are reconciled.
The output is compared with the manuscript.
The work is rebuilt when necessary.
Corrections are retained in the record.
The purpose is not to produce an analysis nobody can criticize.
That analysis does not exist.
The purpose is to make criticism productive—to give reviewers enough provenance to identify exactly where they disagree and what changes when their preferred assumption is substituted.
This is the difference between defending a conclusion and exposing a model.
Node 9: The human gate
Eventually, the machine reaches the boundary of what it can decide.
It can calculate expected costs.
It can estimate quality-adjusted survival.
It can test thresholds.
It can compare scenarios.
It cannot determine how an individual patient should value the risk of stroke against the risk of bleeding, the burden of medication against the irreversibility of an implant, or present certainty against future uncertainty.
Those are human trade-offs.
They require clinical judgment.
They require patient judgment.
They require transparency because the consequences are carried by humans.
This is why I reject two opposite errors.
The first is assuming that because a question is clinically contested, nobody outside the clinical profession may examine its economics.
The second is assuming that because an economic model produces an answer, the clinical question has been resolved.
Both collapse necessary boundaries.
My work lives between them.
The Watchman question did not create the vigilance.
It exposed its wiring.
The heart supplied the sensitivity to the signal.
The healthcare system supplied the incentive map.
Metacognition supplied the bandwidth.
Artificial intelligence supplied additional capacity.
Economic modeling supplied a testable frame.
Adversarial review supplied resistance.
Governance supplied the circuit breakers.
Human outcomes supplied the objective function.
This is also the origin of Phial.
Phial is an attempt to make this process observable and repeatable: to preserve the evidence, assumptions, transformations, approvals, disagreements, corrections, and human decisions behind a consequential economic conclusion.
It is not designed to remove human judgment.
It is designed to show exactly where human judgment enters.
Because vigilance alone can detect the discrepancy.
Metacognition can examine the mind pursuing it.
Diligence can assemble the evidence.
A model can test the economics.
But only a human can answer the question that governs the entire system:
What are we optimizing for—and who lives with the loss if we are wrong?
That is the wiring.
And the patient remains at the center of the circuit.