Three months into her tenure as a federal agency administrator, Eleanor Voss faced a week that would have broken a less grounded leader. The agency's new AI eligibility tool had produced a cluster of wrong denials. A reporter was asking pointed questions. Her data-science team insisted the model was statistically fine; her field offices were furious; a congressional staffer wanted a briefing; and her own deputy quietly disagreed with the path she was leaning toward. There was no clean answer in any binder. Every option carried real harm to someone. This is where leadership stops being about having answers and becomes about how you carry yourself when there are none.

Leading through complexity is the work of staying clear, steady, and humane when the problem is genuinely ambiguous, the stakes are real, and no framework resolves it for you. AI accelerates this kind of moment in government. The technology is uncertain, the public consequences are large, and the pace outruns the policy. The technical lessons teach you what to do. This one is about who you are while doing it: self-aware, adaptive, resilient, and anchored to values. Under genuine complexity, those are the deciding skills rather than the soft ones.

Why this is harder in government than anywhere else

Leading AI in government is not the same as leading AI in a private company, and the source identifies pretending otherwise as the largest failure mode among executives who cross over from industry. A private AI leader optimizes for product outcomes inside business ethics, market pressure and shareholder accountability. A public one operates inside a web of statutory authorities, separation of powers, due process obligations, civil rights law, appropriations constraints, records and disclosure exposure, inspector general oversight, and a direct accountability to citizens that is qualitatively different from accountability to customers.

The inheritance makes it harder still. You will take over systems whose original designers have moved on, statutes written long before neural networks existed, data agreements nobody can locate, and a political environment that may change completely between the day you launch a pilot and the day you explain the results to an oversight committee. The senior AI role formalized in federal guidance sits at the intersection of nearly all of these streams. Technical excellence is necessary and not sufficient. What is sufficient, in the source's phrasing, is the capacity to lead adaptively, hold your values under pressure, and sustain yourself for the long run.

The policy environment is itself genuinely unsettled, which is the second reason this is not a matter of simply knowing more. The source names the EU AI Act, state-level laws including Colorado's SB 205 and California's SB 942, the NIST AI Risk Management Framework, ISO/IEC 42001 and a growing body of case law as interacting in ways nobody fully understands. Claiming to have that landscape figured out is posturing rather than leadership. Naming the uncertainty, building decision processes that stay legitimate when outcomes disappoint, and inviting the people whose rights are affected into the design is the alternative on offer.

Self-awareness: knowing your own defaults under pressure

Every leader has a stress reflex, and under pressure it fires before thought does. Some leaders, when threatened, grab for certainty and over-control. Others avoid the conflict and go quiet. Some need to be the smartest person in the room and stop listening. The danger is not having a reflex; everyone does. The danger is not knowing yours, so it drives the decision while you believe you are being rational. The first system you have to govern under complexity is yourself.

Eleanor knew her reflex. Under threat, she moved fast toward a decisive fix to relieve the discomfort of not knowing. That instinct had served her in operational crises. In an ambiguous AI failure, it was a liability: a fast fix to the wrong diagnosis would harm more citizens. Recognizing the reflex let her override it. She named it to her deputy: "My instinct is to announce a solution today. I don't trust that instinct here. Slow me down." Naming a default out loud is how you keep it from running the room, and it also gives someone else standing to interrupt you.

Adaptive leadership: matching your approach to the kind of problem

The most common leadership error in complexity is applying a technical solution to an adaptive problem. The distinction comes from the adaptive leadership literature developed at the Harvard Kennedy School by Ronald Heifetz and Marty Linsky, and it is the single most portable idea in this lesson. A technical problem has a known answer that an expert can supply and the work is execution: the model has a bug, fix the bug. An adaptive challenge has no expert answer, because the problem itself is contested and stakeholders must change values, loyalties or habits of mind. Eleanor's crisis was both, and conflating them was the trap.

The source's examples make the distinction concrete. Upgrading a secure facility to support a classified AI workload is technical: hard, expensive, and known. Deciding when an AI system should replace a human adjudicator in a benefits workflow is adaptive, because the real question is how the agency weighs efficiency against due process and no engineering answer settles that. Tax system modernization looked for years like a procurement and engineering problem, which it partly was, while the deeper challenge was reconciling a tax code that Congress continually revises with software release cycles measured in years. Better engineering could not resolve that tension.

The source's diagnosis of the failure pattern is worth stating in its own terms: most failed federal AI programs, on its account, failed because leaders treated adaptive challenges as technical problems and procured more software when what was needed was a different conversation about the agency's obligations. Take that as the source's argument rather than a measured finding, and notice that the opposite error also exists. Leaders who default only to adaptive work stall, because they never execute the technical fixes that were within reach the whole time.

The wrong denials in Eleanor's agency had a technical layer: a data error the team could correct in days. The deeper problem was adaptive. Field offices no longer trusted the tool. Citizens no longer trusted the agency. Staff did not know when to override the AI and when to defer to it. No patch fixes trust. That required the slower work: convening the people affected, surfacing the real disagreement instead of papering over it, and letting field offices help design the override process rather than have it imposed. She handled the technical fix quickly and the adaptive work patiently.

A diagnostic you can run in the moment

When a hard problem lands, before you act, sort it with these questions:

• Is the answer known to an expert? If yes, it is mostly technical: assign it and move on. If no one actually knows the answer, treat it as adaptive.
• Does solving it require people to change beliefs or behavior? If yes, you cannot solve it by decree. You have to bring people through the change.
• Am I reaching for a quick technical fix to avoid the harder adaptive work? If the honest answer is yes, that is the reflex talking. Slow down.

For a decision large enough to warrant an hour rather than a minute, the source supplies a fuller checklist used in a public service executive program. It asks six questions, counted from the source's own list: what problem are we trying to solve and at what level of the system; who has a stake and what are their commitments; what loyalties or values are at play for the leadership group; what technical work is necessary in any case; what work requires learning, loss or a shift in assumptions by stakeholders; and where will resistance come from if we pursue the adaptive work. Working through it converts a vague anxiety into a structured plan, which is most of what a diagnostic is for.

Sorting the situation before choosing a method

Adaptive leadership has a companion in Dave Snowden's Cynefin framework, which sorts situations into five domains and prescribes a different move in each. In clear situations, cause and effect are known: sense, categorize, respond, and best practice works. In complicated situations, cause and effect are discoverable by experts: sense, analyze, respond, and good practice works. In complex situations, cause and effect are visible only in retrospect: probe, sense, respond, and practice has to emerge. In chaotic situations, no cause and effect is discoverable: act, sense, respond, and novel practice must be invented. Confused situations require breaking the problem apart and sorting each piece into one of the others.

Most AI decisions that reach a senior government leader are complex rather than complicated, and that single sorting error explains a great deal of expensive failure. Whether a new fraud detection approach will reduce fraud without increasing disparities cannot be reliably predicted by any expert in advance; it can only be probed through small experiments that are safe to fail and carefully observed. Executives who treat a complex problem as a complicated one assemble a large expert team, produce a grand design, and execute it in one go. The source names the 2013 federal health insurance exchange launch as the textbook case, and the 2024 tax filing pilot as a counter-example of a well-run rollout in the complex domain.

The leadership implication is uncomfortable and specific: resist pressure to commit to an outcome before you have probed it. Appropriators, political leadership and reporters will ask what the system will do next year. The honest answer is often that you do not fully know, and that you will run these specific probes to find out. That answer is politically harder than a confident prediction, which is exactly why so many leaders default to false confidence. Sustainable leadership here requires the political courage to stay honest in complex domains while still being decisive about the next probe.

Within a probe you still have to choose, and the source names three decision frameworks that senior policy staff use under uncertainty. Expected value weights each outcome by its probability and is useful when the outcomes are commensurable and you can defend the probabilities. Minimax regret asks which choice you would least regret if the worst plausible case arrived, which suits decisions with an irreversible downside. The precautionary principle places the burden on demonstrating safety before acting, and is the posture most often expected where rights are at stake. Knowing which one you are implicitly using is worth more than defending any of them in the abstract.

Being steady without faking certainty

People want their leader to be certain, and that pull is dangerous, because false certainty in an AI failure destroys credibility the moment reality contradicts it. The skill is to be steady without being falsely certain: to give people a stable presence and a clear process even when you cannot yet give them the answer. Eleanor did this in three moves that any leader can copy, and all three are compatible with an oversight body reading the transcript later.

• Name what is known and unknown plainly. She told staff exactly what the agency knew, that some denials were wrong; what it did not yet know, the full scope and cause; and when it would know more. Uncertainty named is far less corrosive than uncertainty hidden.
• Make the next step clear even when the destination is not. She could not promise the final fix, but she could commit to halting new denials from the tool while the review ran. A clear next step is what steadies a frightened team.
• Decide reversibly where you can. She favored choices she could walk back if wrong, such as a temporary manual review, over irreversible commitments made under pressure. Reversibility buys time to learn.

Values-based leadership when the pressures conflict

Values-based leadership is the practice of staying anchored to a small number of explicit commitments when legal, ethical, political and operational pressures pull in different directions. The source locates the baseline in three places: the federal oath of office, which commits civil servants to the Constitution and to faithful execution of the laws; the Standards of Ethical Conduct for Employees of the Executive Branch at 5 CFR 2635, which commit employees to impartiality, integrity and stewardship of public resources; and the agency mission, whatever specific public purpose your organization exists to serve.

On top of that baseline the source recommends three to five personal non-negotiables specific to AI leadership, and offers five examples: no AI in rights-impacting decisions without a human appeal path; no papering over fairness disparities; naming trade-offs explicitly even when inconvenient; not accepting vendor claims you cannot verify; and investing in successors so the work outlasts you. When political leadership presses you to deploy faster, or a vendor presses you to accept a polished demonstration in place of independent testing, these are the sentences you fall back on.

The source reports that leaders who write their commitments down, share them with their team and reference them in decisions describe fewer regretted decisions and lower decision fatigue. Treat that as reported experience rather than a demonstrated effect. What a written commitment reliably does is narrower and still valuable: it removes the option of quietly discovering, under pressure, that you never held the position at all, and it gives your team a way to hold you to something specific.

Values-based leadership also requires accepting loss, which Heifetz calls the work of the work. When you refuse to deploy a system because its fairness profile is unacceptable, you disappoint the mission team that spent a year building it. When you stop a procurement because the vendor's disclosures are inadequate, you disappoint the leadership that wanted a ribbon cutting. Those losses are real and must be acknowledged rather than dismissed. The task is not to avoid loss but to sequence and communicate it so the organization can absorb it and stay oriented to the mission.

Eleanor's own anchor was simple and stated in advance: when in doubt, protect the citizen over the system's convenience. That single commitment resolved a dozen smaller decisions during the crisis week. It told her to halt the denials even at operational cost, to brief the legislator straight rather than spin, and to credit the field offices rather than blame the vendor. Values decided in calm are the ones you can trust in chaos, because the chaos is precisely when you will be tempted to discover new ones.

Resilience as a professional obligation

Complexity is not a single hard day; it is a sustained load. The source describes research from the Partnership for Public Service and reporting by the Government Accountability Office on senior executive burnout, noting that career executives often work sixty to eighty hour weeks, navigate multiple administrations, and absorb the emotional weight of decisions affecting millions of people. Those figures are the source's. The framing that follows from them is the part worth adopting: resilience is a professional obligation rather than a private indulgence, because a leader whose judgment has eroded is still making decisions.

The source sets out five components. Workload design that reserves protected thinking time in the weekly calendar rather than squeezing it between meetings; the source suggests two to four hours a week, and I would treat that as a recommended allocation rather than a threshold with measured effects. Sleep and physical health, on the source's account that cognitive performance degrades at six hours of sleep or less, which matters when decision quality is your main deliverable. Peer networks, including formal forums such as the Senior Executives Association and the Federal CIO Council alongside informal groups where dilemmas can be discussed candidly. Reflective practice, whether journaling or structured conversation with a coach or mentor. And meaning-making, the deliberate connection of the work to a purpose larger than the current crisis.

Stewardship is resilience's partner. A resilient leader who builds no successors still leaves a fragile program. The source's closing expectation of a senior leader is investment in the profession itself: mentoring the next cohort, contributing to interagency communities of practice, and leaving the program documented and legible enough for a successor to build on. Leaders who do this leave institutions stronger than they found them. Leaders who do not, however talented, leave programs that do not survive a transition, which is a result their own competence made possible.

Dissent is information, not insubordination

Under complexity, the most valuable thing in the room is the objection nobody wants to raise. Amy Edmondson's research on psychological safety, which the source cites, finds that teams do better technical work when leaders invite dissent and treat it as information rather than as a challenge to authority. The federal record supports the negative case sharply: the source points to the Columbia Accident Investigation Board report and to post-incident reviews in the cybersecurity community, where catastrophic failures were repeatedly preceded by dissent that had been raised and suppressed.

Psychological safety is compatible with accountability, and the confusion between the two is why leaders resist it. Inviting dissent is not lowering the standard; it is refusing to let deference decide a technical question. The practical version is narrow and testable. Ask for the objection explicitly, by name, from the person most likely to hold it. Record dissent that you overrule, along with your reason. And notice whether anyone has told you something you did not want to hear this month, because a leadership team with no bad news is not a healthy one, it is a quiet one.

Communicating in four arenas

Senior AI leaders communicate to their teams, to political leadership, to the legislature, and to the public. The norms differ; the core discipline does not. Describe reality accurately, including what you do not yet know, and connect that reality to the next concrete action the audience can take or expect. The source states the underlying claim plainly: false confidence creates larger downstream failures than honest uncertainty does. That is a claim about candour and it deserves to be carried without softening.

With teams, the job is a shared sense of mission, current state, known risks, and the next probe. With political leadership, translate technical reality into consequences for the principal's priorities without losing accuracy: a Secretary does not need the formula behind disaggregated performance monitoring, but does need to know that without it the agency cannot meet its obligations and is exposed to oversight and civil rights inquiry. With the legislature and the public, the highest-leverage move is honesty about what AI can and cannot yet do, because overclaiming sets a program up for a crash when reality asserts itself.

Two specific occasions deserve rehearsal before they arrive, because both are adversarial and both are permanent. The first is testimony, where the written record outlives the hearing and a number given carelessly becomes a correction later. The second is the response to an inspector general or audit finding, where the temptation is to litigate the finding rather than answer it. In both, the source's discipline holds: state what is accurate, name what remains unknown, and attach the next concrete action rather than a reassurance.

The source's conclusion is that the defining discipline of senior AI leadership in government is the willingness to hold complexity publicly: to make the hard choices visible rather than hide them behind technical language. Eleanor's week ended that way. Her agency fixed the data error in four days and rebuilt the override process over six weeks with the field offices in the room. The reporter's story ran, tough but fair, in part because she had briefed straight. The congressional staffer left reassured by the candor rather than the polish. None of that came from a clever answer. It came from a leader who knew her reflex, sorted the technical from the adaptive, stayed steady without faking certainty, protected her own judgment, and let her stated values break the ties.

Anti-Patterns

• The technical fix for the adaptive problem. Procuring more software, retuning a model, or reorganizing a team when the real question is what the agency owes the people its systems decide about. The tell is that the fix is fully within your authority and requires nobody to change their mind.
• The grand design in a complex domain. Assembling a large expert team to produce a comprehensive plan and executing it in one release, for a problem whose cause and effect will only be visible in retrospect. Probes that are safe to fail are slower to announce and faster to learn from.
• False confidence as reassurance. Answering an appropriations or press question about next year with a prediction you cannot support, because "we will run these specific probes to find out" sounds weak in the room. It sounds far worse read back to you a year later.
• The values statement that has never cost anything. A written set of commitments that has not once caused a refusal, a delay or a disappointed stakeholder. Writing them down removes the option of quietly abandoning them; it does not by itself make you the kind of leader who keeps them.
• Resilience as endurance. Treating long hours as evidence of commitment rather than as a slow erosion of the judgment that is your actual deliverable. The people harmed by a degraded decision never see the calendar that produced it.
• Confusing psychological safety with the absence of standards. Concluding that inviting dissent means accepting weak work. Dissent is information about the decision; accountability is about the work. Suppressing the first does not strengthen the second.
• The unnamed reflex. Believing you are reasoning when you are reacting. A leader who cannot state their own stress default in one sentence has not eliminated it, only made it invisible to themselves.
• Adaptive work as an excuse to stall. Convening, listening and reframing indefinitely while the technical fixes that were available in week one remain undone. Both halves of the diagnosis carry obligations.

Practice Prompts

• Write your stress default in one sentence, in the form "under threat I tend to ___". Then tell it to the person best placed to interrupt you, and agree the words they will use.
• Take a live problem and run the six-question diagnostic on it: the problem and its level, the stakeholders and their commitments, the loyalties and values in play, the technical work needed regardless, the work requiring learning or loss, and where resistance will come from.
• Sort three current decisions into Cynefin domains. For any you have classified as complicated, ask what evidence would show it is actually complex, and design one probe that is safe to fail.
• Draft your three to five non-negotiable commitments for AI decisions. For each one, name a specific occasion in the past year when it would have changed what you did.
• Identify the most consequential thing you do not know about a system you are accountable for, and write the sentence in which you would say so to an oversight committee. Read it aloud until it stops feeling like an admission of failure.
• Audit your last month for dissent: what did someone tell you that you did not want to hear, and what happened to them afterward. If nothing comes to mind, ask the quietest person on your leadership team directly.
• Look at your calendar for the last four weeks and find the protected thinking time. If there was none, block it for the next four and treat the first cancellation as data about your priorities rather than about your workload.
• Write the loss ledger for a decision you are avoiding: who is disappointed by each option, and how you would sequence and communicate that loss so the organization can absorb it.

Reflection

Recall a decision you made under real pressure, where the information was incomplete and the clock was running. Reconstruct it honestly: what were you actually optimizing for in the moment, and how much of it was the discomfort of not knowing rather than the substance of the problem? Most leaders find, on inspection, that at least one major decision in their career was made to end an unbearable ambiguity rather than because the evidence had arrived. That is not a character flaw; it is the reflex this lesson is about. The question worth sitting with is what would have to be true, structurally, for you to have caught it at the time.

Then consider the horizon you are actually working on. The systems you approve will run under leaders who did not choose them, in an administration that may hold different priorities, judged by an oversight body reading your documents years from now. What in your current program is legible only to you? What relationships exist only through you? A leader who is genuinely irreplaceable has built something that will not survive them, which is a strange thing to be proud of. Institutional stewardship is the least dramatic form of personal leadership and, over any horizon longer than a tenure, the one that decides whether the work mattered.

Glossary

• Technical problem. A challenge with a known solution that an expert can supply, where the work is execution rather than learning.
• Adaptive challenge. A challenge with no expert answer, because the problem itself is contested and stakeholders must change values, loyalties or assumptions for it to be resolved.
• Cynefin. Dave Snowden's sense-making framework sorting situations into clear, complicated, complex, chaotic and confused domains, each calling for a different sequence of action.
• Safe-to-fail probe. A small experiment run in a complex domain to generate information, designed so that its failure is survivable and observable.
• Reversible decision. A choice that can be walked back if it turns out wrong, preferred under uncertainty because it buys time to learn without compounding the error.
• Psychological safety. A team condition in which people can raise dissent, error and bad news without fear of humiliation or reprisal; compatible with, not opposed to, accountability.
• The work of the work. Heifetz's term for the loss that adaptive change imposes on people, which leaders must acknowledge and sequence rather than deny.
• Non-negotiable commitment. A short, explicit, personal standard decided in advance so that it is available as a tie-breaker when pressures conflict.
• Meaning-making. The deliberate practice of connecting current work to a purpose beyond the immediate crisis, named by the source as a component of executive resilience.
• Stewardship. Leaving a program documented, staffed and legible enough that a successor can build on it, treated here as the partner discipline to personal resilience.

Related Lessons

• Defining the Profession of Government AI Leadership sets out the professional obligations that these personal disciplines serve.
• Mentoring Next-Generation Leaders is the stewardship half of this lesson, worked out as a practice.
• Speaking and Presenting on Government AI develops the communication arenas covered here in operational detail.
• Crisis Management for AI Failures handles the incident mechanics that a week like Eleanor's demands.
• Building Institutional Knowledge covers making your program legible to the people who come next.
• Oversight Mechanisms: IG, GAO, Congress explains the accountability structures whose expectations shape how candour is received.
• Building and Maintaining Public Trust extends the candour argument to the public relationship over time.
• Human-in-the-Loop: Design and Implementation supplies the override design that Eleanor's field offices helped rebuild.
• Rights-Impacting and Safety-Impacting AI Safeguards details the safeguards behind the appeal-path commitment.

Closing

Nothing in this lesson resolves a hard decision for you, and that is the point. Frameworks sort problems; they do not choose. What the diagnostic disciplines buy is a few seconds of distance between the situation and your reflex, which is often the entire difference between a decision you can defend and one you spend a year explaining. The leaders who do this well are not calmer people by temperament. They have simply arranged, in advance, for something other than adrenaline to be in charge.

The rest is preparation done when nothing is on fire. Know your default. Decide your commitments and write them down. Build the peer relationships and the protected time before you need them. Ask for dissent while the stakes are low, so the habit exists when they are not. And document your reasoning so that the person who inherits your program understands not just what you decided but why the boundary sits where it does. Complexity will not be resolved by any of that. It will be survivable, which under these conditions is what leadership actually offers.

Key Takeaways

• Govern yourself first. Your stress reflex fires before thought, so know your default under pressure and name it aloud to someone who can interrupt you.
• Government complexity is structural. Statutory authority, due process, civil rights law, appropriations, disclosure exposure and oversight make this a different job from private-sector AI leadership, not a harder version of it.
• Sort technical from adaptive. Technical problems have expert answers and can be assigned; adaptive challenges require people to change values or assumptions and cannot be solved by decree, or by procurement.
• Most AI decisions are complex, not complicated. Cause and effect appear only in retrospect, so probe with small safe-to-fail experiments instead of committing to a grand design.
• Be steady, not falsely certain. Name what is known and unknown, give a clear next step without the final answer, and prefer reversible choices under pressure.
• Anchor on three to five commitments. The oath, the standards of conduct at 5 CFR 2635 and the agency mission form the baseline; personal non-negotiables decided in calm break the ties in chaos.
• Accept and sequence loss. Refusing a deployment or stopping a procurement disappoints real people; the task is to communicate that loss, not to pretend it away.
• Resilience is professional, not personal. Protected thinking time, sleep, peer networks, reflective practice and meaning-making protect the judgment that is your actual deliverable.
• Dissent is information. Catastrophic failures are repeatedly preceded by suppressed objections; ask for the objection by name and record the dissent you overrule.
• False confidence costs more than honest uncertainty. In every arena, describing reality accurately, including what you do not know, is the discipline that survives the transcript.

Frequently Asked Questions

How do I tell an adaptive challenge from a technical problem when they are tangled together? Most real situations are both, so the question is not which one it is but which parts are which. Ask whether an expert somewhere already knows the answer. If yes, that part is technical and should be assigned immediately. Then ask what would still be unresolved once that fix shipped. If people would still distrust the system, still disagree about when to override it, or still be arguing about what the agency owes those it decides about, that residue is the adaptive challenge and it will not be closed by anything you can procure.

Is admitting uncertainty to Congress or the press actually safe? It is safer than the alternative, which is the more useful way to put it. Government AI operates under audit, and a confident prediction that reality contradicts becomes both a correction and a credibility problem, while an acknowledged unknown paired with a specific next step is a position you can hold consistently for years. What makes candour survivable is pairing it with decisiveness about the immediate action: not knowing the year-out answer is defensible, having no next step is not.

Does writing down my values actually change my decisions? The source reports that leaders who write and share their commitments describe fewer regretted decisions and less decision fatigue, and that is worth knowing as reported experience rather than a proven effect. The mechanism you can rely on is smaller: a written, shared commitment removes the option of quietly discovering under pressure that you never held it, and it gives your team something specific to hold you to. A commitment that has never cost you anything has not yet been tested.

My agency will not fund a coach or protected thinking time. What can I do? The two components with no budget line are the peer network and the reflective practice. A standing conversation with two or three peers facing similar decisions, with an explicit agreement that dilemmas can be discussed candidly, does most of what a coach does at the level of judgment. Protected time is a calendar decision rather than a funding one; the resistance is usually cultural, and the honest test is whether you would defend the block to your own leadership if asked what it was for.

How do I invite dissent without losing authority? Ask for it specifically rather than generally. "Does anyone have concerns" produces nothing; "Priya, you have run more of these evaluations than anyone here, what is the strongest case against this" produces the objection. Then do the second half, which most leaders skip: when you overrule the objection, say that you are overruling it and why, and record it. That is what makes dissent safe to offer next time, and it also produces the record you will want if the decision turns out badly.

What if my political leadership demands certainty I do not have? Distinguish the two things being asked for. Certainty about the outcome you cannot honestly supply. Confidence about the process you can: what you will test, by when, what would cause you to stop, and who will be told what you find. Most demands for certainty are really demands for control, and a specific probe plan with dates answers that need without requiring you to assert something the evidence does not support and the record will later contradict.