Skip to content
finalstate.aiThe Decision Premise: How Upstream Assumptions Shape the Choice
108Atlas

108THE SIGNATURE

The Decision Premise: How Upstream Assumptions Shape the Choice

A decision premise is an assumption a choice treats as settled. In the reverse-centaur risk, automated output supplies those assumptions while a person signs.

A campaign memo leads to a signed shutdown order, with a return route to reopen the premises.The memo says the campaign is under target. Its route to the signed order shows influence before approval; the separate return arrow preserves the possibility of challenging the evidence or target.Campaignunder targetShutdownReopen
Illustrative campaign decision. The signature records an action; it does not show who supplied the assumptions or whether they were examined.

A CONCLUSION FROM PREMISES

A claim about the world meets a choice about what matters.

A factual claim and a chosen objective meet at a review junction with reject, revise and request-evidence paths.Two premise records enter a review junction. The outgoing paths represent ways to challenge the proposed action, not a conclusion mechanically forced by the inputs.ClaimObjectiveReviewRejectReviseEvidence
Adapted from Simon’s distinction; the review paths are an editorial application. A changed premise may change a decision.Simon, Administrative Behavior, 4th ed. (1997), ch. I, “Value and Fact in Decision.” https://www.simonandschuster.com/books/Administrative-Behavior-4th-Edition/Herbert-A-Simon/9780684835822

A factual premise is an empirical claim, which may be mistaken. A value premise specifies an objective or preference. In the campaign memo, reported performance and the chosen target supply different reasons for acting. Neither removes the need for judgment. The reverse-centaur risk grows when these premises arrive as instructions. Separating them helps locate prediction and decision: evidence can inform a choice, while its purpose remains open to examination.

SIMON’S DECISION PREMISES

Simon put the decision inside the organization.

A schematic book bearing Herbert Simon's name, Administrative Behavior and its original publication year, 1947.A still bibliographic object marks the original publication year. It is neither a historical portrait nor a reproduction of a particular book cover. The accompanying citation identifies the 1997 edition actually consulted.AdministrativeBehaviorHerbert Simon1947
Schematic source object. Original publication: 1947. Text consulted: 4th ed., Free Press, March 1, 1997, ch. I.https://www.simonandschuster.com/books/Administrative-Behavior-4th-Edition/Herbert-A-Simon/9780684835822

Herbert Simon's Administrative Behavior first appeared in 1947. Its distinction between facts and values helps explain how a recommendation acquires authority. The edition cited here is the fourth, published in 1997; its first chapter connects decision making, organizational influence and accepted authority.

ACCEPTED PREMISES

A target can govern without being argued again.

A target enters a decision record as given; a separate challenge route returns to its source.The forward route shows a target accepted without renewed examination. The return route is an editorial governance proposal, not a claim that Simon's definition guarantees permission to challenge every organizational instruction.TargetGivenChallenge
Simon, 4th ed. (1997), ch. I, “Modes of Organizational Influence—Authority.” Challenge and escalation are the application shown here.https://www.simonandschuster.com/books/Administrative-Behavior-4th-Edition/Herbert-A-Simon/9780684835822

Simon distinguishes accepting authority from being persuaded: a person follows another's decision without independently examining its merits. Authority in this sense can also operate upward or sideways. In the campaign example, treating the target as given lets an earlier choice govern the review. The person signing needs a route to question it, including escalation to whoever can change it.

The pen is only one place to look for influence.

A reviewer can retain discretion while reasoning inside a frame supplied by someone else. The person who fixes the option set has already shaped what reaches the desk.

UNCERTAINTY ABSORPTION

The summary travels farther than its evidence.

An inference travels from a retained evidence file; a return path preserves access to the evidence and caveats.The original records and their marked caveat remain visible while the inference moves onward. A separate return route shows the editorial proposal to retain access for review; it does not imply that uncertainty has vanished or all compression is improper.EvidenceCaveatsInference
March and Simon, Organizations (1958), ch. VI, pp. 165–166. Campaign details and the evidence-access route are illustrative applications.https://web.mit.edu/curhan/www/docs/Articles/15341_Readings/Behavioral_Decision_Theory/March_%26_Simon_Cognitive_Limits_on_Rationality_Ch6_in_Organizations.pdf

March and Simon call this uncertainty absorption: an inference is communicated in place of the evidence from which it was drawn. In the campaign example, “under target” can travel while the attribution choices, reporting window and caveats remain with the analyst. A summary can be useful. Compression also changes what the receiver can inspect. Retaining access to the evidence gives a challenge somewhere to go.

WHAT TRAVELS, WHAT STAYS

Uncertainty absorption takes place when inferences are drawn from a body of evidence and the inferences, instead of the evidence itself, are then communicated.

March and Simon, Organizations (1958), p. 165.

Andrew Pettigrew's 1972 study examines a gatekeeper filtering information during an innovative decision process, within an organization and across its boundaries. It treats information control as a power resource. This supports asking what the gatekeeper made available; it does not establish that the gatekeeper alone determined the outcome. A curated menu poses a related question about what the recipient gets to consider. Source: Pettigrew, “Information Control as a Power Resource,” Sociology 6(2), 187–204; abstract.

THE WELD

A measured return and a chosen target answer different questions.

A campaign record separates into reported ROAS and the chosen target against which it is judged.The split exposes measurement and decision criterion as distinct parts of a campaign rule. No numeric target or empirical campaign result is supplied. Reported attribution is not shown as proof of incremental revenue or profit.Reported ROASChosen target
Illustrative decision rule. ROAS describes return per unit of ad spend; choosing a target is a separate act. Reported attribution alone does not establish incremental revenue or profit.For conversion value/cost and target ROAS: Google Ads Help. https://support.google.com/google-ads/answer/6268637

Return on ad spend, or ROAS, can report attributed revenue divided by advertising spend. Its meaning depends on measurement choices, including attribution and the reporting window. Choosing a minimum acceptable ROAS adds a decision criterion; the measurement does not select that target. A model can help estimate performance or compare proposed objectives under stated assumptions. Accountable people still need to examine both. Even a sound “below target” result leaves choices about testing, waiting or stopping. This is where prediction meets decision.

IN THIS FRAMEWORK

The recommendation and the menu both carry premises.

Both a recommendation memo and an option menu contain a claim and an objective.Two complete examples cross the same empirical and evaluative distinction. The recommendation is not assigned to facts alone, and the menu is not assigned to values alone. No option is preselected.MemoClaimObjectiveMenuClaimObjective
Editorial application across the Atlas. Both examples have factual and value premises; neither column assigns a premise type to machines or humans.

In a recommendation treated as an instruction, a performance claim can arrive attached to an objective. In the construction of a menu, expectations about the options can arrive attached to a preference for one kind of outcome. Each contains empirical and evaluative choices. A curated set may restrict what is considered without removing all discretion. Useful review can question the evidence, the objective and the exclusions.

WHO SUPPLIED THE PREMISE

Trace the claim. Find the target's owner.

The claim's author and target's owner can each be traced from the decision record, which retains a route to reopen review.Separate provenance routes identify two responsibilities; they need not belong to different people. The return path preserves a way to reopen the premises before acting.Claim authorTarget ownerReopen
A review diagnostic, not evidence of misconduct. Separate responsibilities may belong to the same person.

Before accepting a finding as a reason to act, identify its empirical claim, the objective it serves and who supplied each. A descriptive sentence need not contain a hidden value judgment; its use in a decision may still depend on one. Keep a route to the evidence and to whoever can revise the objective. Then inspect which options made the menu.

Read the transcript

01 · THE SIGNATURE

A memo lands on the desk: the campaign is under target. The boss signs an order to shut it down. The signature tells us who authorized the action. It leaves another question open: who set the terms on which stopping looked reasonable? Someone selected a measure, interpreted the evidence and supplied a target. The signer could still reopen those premises, ask for the underlying records, or change the objective. The reverse-centaur risk arises when automated output supplies assumptions that a person accepts as settled.

02 · A CONCLUSION FROM PREMISES

The campaign memo contains two kinds of premise. A factual premise makes a claim about the world: this is the campaign's measured return. It remains open to error, uncertainty and challenge. A value premise supplies a preference: this level of return is what the organization wants. At review, the signer can reject the proposal, revise an assumption, or request evidence. These premises need not settle whether to stop, run another test, or wait for more information.

03 · SIMON’S DECISION PREMISES

Herbert Simon published Administrative Behavior in 1947. He examined how organizations influence decisions, including their purposes and the acceptance of authority. The text cited here is the fourth edition, published in 1997; that edition is separate from the original publication date.

04 · ACCEPTED PREMISES

A target can be accepted as given. Simon calls it acceptance of authority when someone follows another person's decision without independently examining its merits. Persuasion is different: reasons are considered. This kind of authority can operate upward or sideways, as well as down a hierarchy. In the campaign review, a useful safeguard is a challenge route back to the target's owner. The signer may need to escalate a proposed change. The inherited assumption can then be examined.

05 · TO SHAPE THE PREMISE IS TO SHAPE THE DECISION

Look upstream of the pen. A reviewer may retain discretion while reasoning inside someone else's frame. The target defines a standard. The selected options define what can be chosen. Finding those earlier decisions gives the final reviewer something concrete to question.

06 · UNCERTAINTY ABSORPTION

An analyst sends the inference that the campaign is under target. The memo can travel without the attribution choices, reporting window or caveats that produced it. March and Simon called the substitution of inference for evidence uncertainty absorption. The recipient has less material with which to assess the summary. Retaining access to the evidence gives the reviewer a route back to those omitted details. Useful compression still leaves a question about what was omitted.

07 · WHAT TRAVELS, WHAT STAYS

March and Simon write: Uncertainty absorption takes place when inferences are drawn from a body of evidence and the inferences, instead of the evidence itself, are then communicated. Their account concerns what the recipient can evaluate. In 1972, Andrew Pettigrew studied information filtering by a gatekeeper during an innovative decision process. Information control can be a power resource. Filtering can influence what reaches a decision maker. That alone does not show that the gatekeeper determined the outcome.

08 · Advertisement · Stability AI Brand Studio

For creative teams, speed only matters if the result still belongs to the brand. Stability AI brings generation and editing tools into production: image, video, audio, 3D, product photography, and campaign assets. Use Brand Studio or deploy through API, cloud, or self-hosted models, with more control over how the work is made.

09 · THE WELD

Return on ad spend, often shortened to ROAS, can report attributed revenue divided by advertising spend. The measurement depends on how revenue is attributed and which reporting window is used. The threshold for an acceptable return is a separate choice. A campaign rule combines that measured result with the chosen target. It can conceal who supplied each part. A model may help estimate performance or compare objectives under stated assumptions. Accountable people need to examine both the measurement and the target. A result below target still leaves a decision about whether to test, wait, or stop.

10 · IN THIS FRAMEWORK

Consider two documents on the same desk: a recommendation and a menu of options. Both carry claims about consequences. Both can carry objectives that make some outcomes preferable. The recommendation may treat campaign performance as a reason to stop. The menu may offer only alternatives expected to meet the existing target. Review can challenge a forecast, question that target, or ask why an option was excluded. Neither document divides neatly into facts for machines and values for people.

11 · WHO SUPPLIED THE PREMISE

Trace the claim to its author and the target to whoever owns it. Those responsibilities may belong to the same person. Ask what evidence supports the finding and why this objective governs the choice. A descriptive sentence need not hide a value judgment, but using it as a reason to act may depend on one. Keep review open: retain access to the evidence and a route to whoever can revise the objective. Then move to the menu and ask which options were left out.

01 / 11 · THE SIGNATURE0:00 / 5:43