ingridleiria Principal Simulator
The expensive failure in a Chief of Staff role is discovering an objection in the room. It costs the meeting, which can usually be rebooked, and it costs a little of the standing that makes the next recommendation land, which cannot. A leader who has twice been handed a document that did not survive their first question starts reading the third one differently, and starts asking someone else for the analysis.
Almost every one of those objections was predictable. Executives are far more consistent than they appear from below: the same first question for years, the same three things they care about disproportionately, the same phrase when they are about to say no. What looks like an unpredictable reaction is usually an unmodelled one. The failure this skill prevents is walking into a room without having spent twenty minutes reading your own document as the person who will read it.
This skill is that reading. It is not a flattery machine and not a devil's advocate for its own sake. It plays one specific person as accurately as the written profile allows, and its most valuable output is often the sentence saying the profile cannot answer the question being asked.
When to use this, and when not to
Use it before a document goes upward: a decision memo, a board narrative, a budget request, an investor update, a plan that asks for headcount. Use it before a conversation you are dreading, where the difficulty is the person rather than the content. Use it when a recommendation was rejected and you do not know why, since replaying the document as the principal usually finds the sentence that lost it. Use it when you are new to a principal and every interaction is still a guess, where the discipline of writing the profile matters more than the simulation.
Do not use it to improve a document that has not been written. Simulation on a half-formed argument returns the obvious objection that the missing half would have answered. Write the thing first, at least in outline, then test it. Do not use it as a substitute for asking. Where the question is factual and the principal is reachable, ask them; a simulation is for when asking is expensive, slow, or would itself signal something you do not want to signal.
decision-memo builds the document this tests. executive-briefing prepares the principal for someone else's meeting rather than testing your own. meeting-to-decisions captures what actually happened afterwards, which is the raw material for updating the profile. ceo-communications writes in the principal's voice for an external audience, which is a different task: this skill predicts how they read, that one produces how they sound.
What you need before starting
A written profile of the principal. This is the whole skill. Without it the simulation produces a generic executive, which is worse than nothing because it sounds plausible and will be believed. Kept as principal.md beside this skill, one per principal where you support more than one. Missing: build a first version from the last five interactions you can remember before running any simulation, and label every prediction as low confidence until the profile has been corrected at least once by reality.
The document or the conversation plan. The actual artefact, at the length it will really be, not a summary of it. A simulation run on a description of a memo tests the description. Missing: run the challenge mode only, which works from the claim rather than the text, and say that the read-through was not possible.
The setting. Read alone at a desk, presented in a leadership meeting, forwarded by someone else with a one-line covering note, opened on a phone between flights. The same document produces different reactions in each. Missing: assume the least favourable realistic setting, which is usually on a phone with three minutes, and say you have assumed it.
What has already been said on the subject. Executives react to a position partly on its merits and partly on its history with them. A recommendation they rejected in March lands differently in September, and lands worse if it is presented as new. Missing: ask the two or three people who were in the earlier conversations, and treat any prediction on a subject with unknown history as unreliable.
The ask. What the document wants from them: a decision, an approval, awareness, air cover, a signature. Missing: this is itself the finding. A document whose ask is unclear to you will be unclear to them, and the predicted reaction is a question about what you want, which costs the meeting.
Who else will be in the room. Principals behave differently in front of their board, their peers, and their direct reports. The same objection raised privately becomes a demand for a rework when raised in front of the chief financial officer. Missing: ask, because this single fact changes more predictions than any other.
The profile
Written once, revised whenever a prediction is wrong. Aim for a page and a half. It is not a personality assessment and should contain nothing you would be embarrassed to have the principal read, which is also a useful test of whether it is accurate rather than resentful.
Capture, with a line or two each:
What they are measured on, and by whom. A chief executive measured on gross margin reads a hiring request differently from one measured on growth. Name the metric and the audience.
The two or three things they care about disproportionately. Not the official priorities. The things that reliably get a longer reply than the subject deserves.
How they read. Whether they open attachments, how much time a document actually gets, whether they read to the end or reply from the first screen. This determines where the argument has to sit.
Their recurring first question. Most executives have one, and it is stable across years. What does this cost. Who else has seen it. What does the customer say. What happens if we do nothing. Write down the exact phrasing, not the gist.
What makes them say no, in their own words. Record actual quotations. A remembered phrase predicts far better than a described trait: "I do not want to be the first person to try this" tells you more than "risk-averse", because it tells you the counter, which is a reference customer.
Their tolerance for uncertainty, and the form they want it in. Some want ranges and confidence levels. Some hear a range as indecision and want a number with a caveat attached. Getting this wrong makes an honest analyst look unprepared.
What they have already decided and do not want reopened. The most reliable source of unexpected friction.
How they behave under time pressure. Most of these conversations happen under it. Some compress and decide, some defer everything, some get sharper and more literal.
Who they consult before deciding. Often one or two people whose objections are the real objections. A document that has already answered the chief financial officer's question is a different document.
The last three predictions and whether they were right. This turns the profile from a description into an instrument.
The method
Load the profile and state its age. A profile more than two quarters old is a hypothesis. People change what they care about when their own measures change, and a new board seat, a new investor, or a missed quarter resets the model. If the profile predates any of those, say so before predicting anything.
Fix the setting and the audience. Written before reading the document, so the prediction is not quietly adjusted to fit the artefact. Setting, time available, who else is present, and what was said last on this subject.
Run the read-through. Read the document at the principal's real reading speed, in the real order, which for many is subject line, first three lines, then a scan for numbers, then the ask. Report where attention goes, where it stops, the sentence that produces a question, the sentence that produces irritation, and the point at which they would stop reading and reply. Report it as behaviour, not as a critique. "They would stop here and reply asking what this costs" is usable; "the cost is unclear" is a note to yourself.
Run the challenge. The three questions they will actually ask, in their words, in the order they would come. Then, for each, whether the document answers it and where, quoting the line that does. An unanswered question at this stage is either a rewrite or a prepared verbal answer, and deciding which is a judgement about whether the answer belongs to everyone reading or only to this reader.
Before adding anything specific to the principal, check the standing set that most upward documents fail on: what does this cost, who is against it, what happens if we do nothing, who owns it, why now rather than next quarter, and what did the person closest to this say. If the document does not answer all six, the principal-specific questions barely matter yet.
Run the decision. What they would decide and why. The rule that matters here: the most common executive answer is neither yes nor no, so predict the conditional. Yes with a condition. Not now, bring it back after the quarter closes. Bring me the version with the finance team's view attached. Predicting a clean yes or no when the real answer is a condition is the most common way a simulation is wrong in a way that costs a week.
Then rank what would move the decision, by effect. Usually two or three things, and usually one of them is a person rather than a fact.
Run the rehearsal, where a conversation rather than a document is the artefact. Play the principal in dialogue. Interrupt where they would interrupt, ask the question they would ask, be as short as they are. Do not concede because the user pushed; concede when the argument earns it, and say what earned it. At the end, step out of the role and give the assessment: what worked, what did not, and the one sentence to lead with next time.
Separate the prediction from the recommendation. Two headings, always. What they would do, and what is actually right. Where these diverge, say both, because the correct move is sometimes to send the document that will not land and accept the friction. A simulator that collapses these two teaches its user to optimise for approval, which is how a Chief of Staff becomes useless within a year.
Name the objection the user is avoiding. It is usually visible in what the document spends its length on, and in what it mentions once and moves past quickly. This is the most valuable single output and the one most often omitted, because it is uncomfortable to write.
State the confidence, and what the profile cannot answer. Three levels is enough: confident, because the profile has direct evidence; likely, because it follows from the pattern; unknown, because the profile has nothing on this. "This depends on how they react to being surprised in front of the board, which is not in the profile" is a good answer and a prompt to add it.
After the real event, record what happened and correct the profile. Whether each predicted question was asked, whether the decision matched, and what phrasing they actually used. This takes five minutes and is the only thing that makes the next simulation better than the last.
Calibration
A simulator nobody checks becomes a confident fiction. Keep a short prediction log: date, document, the three predicted questions, the predicted decision, and afterwards the actual. Five entries is enough to see the pattern. The two error directions are asymmetric: over-predicting objections produces over-long documents that pre-answer questions nobody asked, and its cost never shows up as a visible failure, only as work that takes three days instead of one.
A profile is calibrated when the first question is predicted correctly around three times in four. Below that, the profile is describing a type rather than a person, and the fix is more quotations and fewer adjectives.
Worked example
Situation. A Chief of Staff at a business software company of about 300 people supports a chief executive who came from an engineering background and has been running the company for six years. The company missed its quarterly number by eleven percent for the first time. The head of product wants to move an unreleased product forward by a quarter, which requires pulling four engineers off the platform migration, and has asked the Chief of Staff to get it on the leadership agenda. The memo argues that the earlier launch recovers roughly 900,000 dollars of the shortfall in the following two quarters. All figures in this example are US dollars.
Task. Decide whether the memo is ready to send, and if not, what to change, before Thursday's leadership meeting. Good means the meeting produces a decision rather than a request for more analysis.
Action. The profile had two directly relevant entries. The chief executive's recurring first question is "what does engineering think", asked of every proposal touching the roadmap, for six years. And a quotation from eighteen months earlier: "I have paid for a migration twice and I am not paying for it a third time."
The read-through predicted attention stopping at the second paragraph, where the memo said the four engineers would return to the migration after launch. That sentence is the one that collides with the quotation, and the prediction was that it would produce not a question but a decision, before the revenue argument on page two had been read at all.
The challenge produced three questions. What does the engineering lead think about pausing the migration. What is the new migration completion date. And what happens to the 900,000 dollars if the launch slips four weeks, which it will, because the last two did.
The memo answered none of them. It contained one line saying engineering was supportive, which under examination turned out to mean the head of product had mentioned it to the engineering lead in a corridor.
The wrong turn: the first response was to rewrite the memo to defend the revenue estimate, which was the part the head of product was most attached to and had spent the most time on. Two hours went into a sensitivity table showing the estimate held under slippage. The simulation of the redraft predicted the same stop at the same sentence. The revenue figure was never the objection, and strengthening it made the memo longer without moving it. The table was cut.
The rewrite instead did three things. It moved the migration impact to the opening, with a dated new completion date agreed in writing with the engineering lead. It replaced the corridor conversation with the engineering lead as a named co-author. And it added the slippage case honestly, showing the recovery at 900,000 dollars on time and about 500,000 dollars four weeks late, with the note that both were above the cost of the delay to the migration.
The prediction on the redraft was a conditional yes: proceed, with the migration date committed publicly to the leadership team and reviewed monthly. The recommendation section noted a divergence: the simulation predicted approval, but the Chief of Staff's own view was that the engineering lead's agreement had been obtained under pressure and was worth testing separately before the meeting. That test was run and the agreement held, with one condition added about a contractor for the migration.
Result. Decided in the meeting in about fifteen minutes. The first question asked was "what does engineering think", in those words, which the memo answered in its second sentence with the co-author's name. The approval carried the predicted condition on the migration date, plus the contractor, which had come from the separate conversation and was already in the memo as a footnote.
The launch slipped three weeks. Recovery came in around 560,000 dollars over the two quarters, inside the range the memo had stated, which mattered more than the number itself: the next roadmap memo from the same team was read to the end.
The profile was updated with one new line, that the chief executive accepts a downside case stated plainly and reacts badly to one discovered by asking.
A second scenario, where it goes differently
Same company, a year later, a new chief financial officer four months into the role. The Chief of Staff has to send a memo requesting a change to how the sales commission plan is accrued, and the profile for the new chief financial officer has three entries in it, all from group meetings.
Running the four modes on that profile produces plausible output and it should not be trusted. The right move is to say so explicitly and change the approach: instead of predicting, gather. A fifteen minute conversation to ask directly how they prefer to receive a proposal like this, whether they want the detail up front or the recommendation, and who they would want consulted, produces more than any simulation from three data points. The output of the skill in this case is one paragraph saying the profile is too thin to simulate, and a list of the four questions worth asking to build it.
What changed is the evidence base, not the method. A profile built from observed behaviour over months supports prediction. A profile built from three group meetings supports only a decision about what to find out, and pretending otherwise produces confident output that will be wrong in a way nobody can trace back.
Output
The default is a single block, written to be read in three minutes.
PRINCIPAL SIMULATION
Principal: [name] Profile last updated: [date]
Artefact: [document or conversation]
Setting: [where, how long, who else is present]
History: [what has been said on this subject before]
READ-THROUGH
Stops at: [the exact sentence, quoted]
Reaction: [what they do at that point, as behaviour]
Reads to: [where they stop reading]
Takes away: [what they would say the document said, in one line]
THE THREE QUESTIONS
| # | Question, in their words | Answered? | Where |
| 1 | | yes/no | |
| 2 | | | |
| 3 | | | |
Standing six: cost / opposition / do nothing / owner / why now / closest person
[which of these the document fails, if any]
PREDICTED DECISION
[Yes, no, or the conditional, stated as they would state it.]
Confidence: [confident / likely / unknown] because [evidence in the profile]
WHAT WOULD MOVE IT
1. [largest effect first]
2.
3.
RECOMMENDATION, WHICH MAY DIFFER
[What is actually right, and whether to send the document that will not land.]
THE OBJECTION YOU ARE AVOIDING
[One or two sentences, named plainly.]
WHAT THE PROFILE CANNOT ANSWER
[The gaps that affected this prediction, as prompts to add.]
For the rehearsal mode, deliver the dialogue itself, then a short block: what worked, what did not, the one sentence to lead with, and the question you were not ready for.
For the prediction log, one row per run:
| Date | Artefact | Predicted first question | Actual first question | Predicted decision | Actual | Profile updated |
Failure modes
The generic executive. Recognise it because the predictions would fit any senior leader: they will ask about cost, they will want to know the risks. Fix by requiring at least two quotations from the profile in any prediction, and by refusing to run the simulation when the profile has none.
Simulated approval. Recognise it when the output says the document is ready and lists no unanswered question. A Chief of Staff told the memo is fine, who then watches it fail, stops using this and is right to. Where a document genuinely is ready, say what would still go wrong and at what probability.
Optimising for approval rather than accuracy. Recognise it when the changes prompted by the simulation are all removals of inconvenient facts. Fix by keeping the recommendation section separate and writing it before the prediction, not after.
The stale profile. Recognise it when predictions start being wrong in the same direction after the principal's own situation changed. Fix by re-reading the profile after any board change, funding event, missed quarter, or new executive hire.
Using it on someone you resent. Recognise it in the adjectives. A profile that calls someone impatient or political is a complaint, not a model. Rewrite it as observed behaviour with quotations, and if you cannot, ask a colleague to write it instead.
Edge cases
No profile at all and the meeting is tomorrow. Run the standing six questions against the document, predict from the role rather than the person, and label the output as role-based. A chief financial officer will ask about cash timing, a chief technology officer about maintenance load, a head of sales about what the customer said. Weak but honest, and it catches the most common gaps.
Multiple principals with conflicting priorities. Simulate each separately and produce the conflict explicitly rather than an averaged reaction. The useful output is the sentence one will object to and the other requires.
A principal who is genuinely unpredictable. Some people decide differently on the same facts depending on the week. Where the log shows this, predict only the questions, which stay stable even when the answer does not.
Quality bar
- The profile is loaded, dated, and any gap that affected this prediction is named.
- At least two predictions quote the principal's own recorded words.
- The three questions are in their phrasing, not paraphrased into neutral language.
- The predicted decision includes the conditional form where that is the likely answer.
- The prediction and the recommendation are separate sections, and any divergence between them is stated.
- The objection the user is avoiding is named explicitly, in one or two sentences.
- No approval is simulated that the document has not earned; where the document is ready, the residual risk is still stated.
- Confidence is labelled on the decision prediction, with the profile evidence behind it.
Adapting this to your context
Written for a Chief of Staff supporting one or two executives in a company of one hundred to a thousand people, where the principal is reachable and decisions are made by a person rather than a committee.
- One principal, one profile. Where the decider is a committee, an investment board or a public body, profile each voting member, simulate them separately, and treat the conflict between two as the output rather than averaging them into a composite nobody resembles.
- The recurring first question. Stable in an individual over years. Committees have a standing agenda and published criteria instead, so read the last three sets of minutes rather than relying on remembered quotations.
- A profile older than two quarters is a hypothesis. Shorten that window during a new board, a live financing or a restructure, and lengthen it for a long-tenured leader in a stable organisation.
- The standing six questions. These fit commercial decisions. In clinical, academic, legal or regulated settings, add the two that always come: what does the evidence base say, and what does the governing rule require.
- What not to change. The prediction and the recommendation stay in separate sections, and no approval is simulated that the document has not earned.
Related skills
decision-memo produces the document this most often tests, and its reversibility judgement usually predicts the principal's depth of questioning. executive-briefing prepares the principal for a meeting rather than testing your document against them. meeting-to-decisions records what actually happened, which is the input that keeps the profile calibrated. ceo-communications writes in the principal's voice for others to read, where this skill reads in their voice. board-and-investor-management and board-deck extend the same profiling method to a board, at lower confidence. crisis-and-incident-comms uses the rehearsal mode under time pressure, where the profile entry on behaviour under pressure carries most of the weight.