Revising Beliefs Under Evidence — Architecture
The dynamic revision discipline — moving belief when evidence warrants and holding it when it does not. The Range-position work itself, made operational.
Foundation
What This Category Holds
The discipline of the Foundation is honest inquiry. The work of this category is the dynamic side of evidence accountability: training the capacity to move belief when the evidence moves, and to hold belief when the evidence does not, regardless of what social pressure or attachment is asking for. It pairs inside the Foundation with Calibrating Confidence to Evidence, which carries the static side of the same accountability: proportioning belief-strength to evidence-strength at a given moment. The static work asks whether your confidence matches the evidence in front of you. The dynamic work asks whether your belief moves when the evidence moves. Training on one of these does not automatically train the other.
The category equips you with one capacity that works on both sides of a single discipline. On one side, the capacity to update — to redraw the map when reality contradicts it, without arguing with reality and without abandoning the prior commitment as if it had cost nothing. On the other side, the capacity to hold — to keep a belief in place when the pressure to revise comes from social cost, identity threat, or surface-level new information rather than from substantive evidence. Both moves are active. The category does not train willingness to update at the expense of willingness to hold, or willingness to hold at the expense of willingness to update. It trains the discipline that does both, and the diagnostic that distinguishes which the moment calls for.
Revising Beliefs Under Evidence holds the Range against two specific pulls. The pull toward Control is the refusal to revise: belief locked in, evidence filtered or dismissed, the cost of being wrong made so high that updating becomes annihilation rather than calibration. The pull toward Decay is the opposite move: belief revised too easily, every gust of new information producing motion without anchor, the mind unable to commit to a stable picture of reality long enough to act on it. Both failures abandon the Range. The discipline is neither — the willingness and capacity to move belief when evidence warrants, paired with the willingness and capacity to hold when it does not.
This is the operational layer of the Range-position work the Foundation introduces. The category is where "calibrated confidence" — a phrase the Foundation chapter uses to name the posture between cowardice and arrogance — becomes a discipline you can train rather than an attitude you can claim. The static side asks you to match confidence to current evidence. The dynamic side asks you to move when the evidence moves. Both moves are accountable to evidence rather than to identity, to social signal, or to the inertia of having held the belief for a long time.
The Tools Inside
The tools inside this category do not form a single ladder. They share a pressure point: a belief, plan, disagreement, or proposed reform is trying to move. The category asks whether that motion is answerable to evidence.
Steelmanning. The discipline of constructing the strongest version of the opposing position before responding to it. Inside Revising Beliefs Under Evidence, Steelmanning is the updating-disciplinary practice: it surfaces the case that should make you update in the form most likely to actually make it possible. Until you can articulate the opposing position clearly enough that its proponent would sign the version you wrote, you have engaged your imagination of the disagreement rather than the disagreement itself, and any update fired against that imagined version is updating against a creature of your own biases. Primary placement here per the cross-discipline tool convention; the cross-reference to Bond → Receiving Disagreement Well sits below. Source: Aristotle, Aquinas, Mill, Dennett, the rationalist community. Disposition: Living.
The Update Protocol. The structured method for actually moving belief when disconfirmation arrives. The mechanism is pre-commitment: before the pressure of being wrong arrives, you articulate what evidence would change your mind, and you write it down with enough specificity that the future encounter with that evidence cannot be quietly dismissed. The protocol does not make updating painless — it makes updating less subject to the motivated reasoning that would otherwise distort the assessment at the exact moment evidence appears. The diagnostic question the chapter carries — "What evidence would change my mind about this?" — is the protocol's entry point. Sources: Popper, Yudkowsky, CFAR, pre-registration, Tetlock. Disposition: Living.
Double Crux. The cooperative disagreement practice for finding the belief both sides would need to update before the surface disagreement can move. Inside Revising Beliefs Under Evidence, Double Crux makes update conditions visible between people: each party names the claims their conclusion depends on, then both search for a shared hinge close enough to evidence that investigation can proceed. The Foundation placement is primary because the distinctive mechanism is belief-revision architecture, even though the practice often happens in Bond territory. Sources: CFAR, Duncan Sabien, LessWrong and rationalist practice. Disposition: Living.
Chesterton's Fence. The discipline of understanding why a structure exists before removing or replacing it. Inside this category, Chesterton's Fence carries the holding side of dynamic revision: the belief "this structure has no function worth preserving" must be tested before action follows. It holds against Control by refusing to preserve every inherited structure merely because it exists, and against Decay by refusing to remove constraints whose function the reformer has not yet understood. Source: G.K. Chesterton's 1929 fence passage, extended through reform, systems, and refactoring practice. Disposition: Living.
Murphyjitsu. The future-facing planning practice that imagines failure in advance, identifies the most plausible break point, patches the plan, and repeats until failure would be surprising. Inside this category, Murphyjitsu treats the plan's assumed success as a belief that must update under simulated evidence before reality supplies harsher evidence. Sources: CFAR's Inner Simulator unit, Murphy's Law, Kahneman and Tversky on mental simulation, prospective hindsight, Gary Klein's premortem method, planning fallacy research. Disposition: Living.
The list is open. Other lineages working on dynamic belief revision can enter through the candidate protocol, admitted on the distinct contribution they make.
Cross-Reference: Steelmanning's Bond Function
Steelmanning operates in two places with the same underlying skill, and both belong on the record rather than collapsed into one.
The Foundation application is what this category carries. Steelmanning here is an updating-disciplinary practice operating on your own position over time: the work of constructing the strongest version of the case against your current view, so that updating fires on a real counter rather than on a weak one. The work is mostly internal. The object is the belief you already hold. The temporal frame is over time, across many encounters with disconfirming evidence.
The Bond application is different. There, the same skill becomes a relational practice: engaging another mind's position seriously before responding, the work that turns collision into productive friction rather than rupture or silencing. The work is mostly external. The object is another's argument, encountered live. The temporal frame is the moment of engagement. That function belongs inside Receiving Disagreement Well when that Bond category is built, where Steelmanning should be named as the receiving-side function: the same skill applied relationally.
This page keeps it as one tool with two applications rather than two separate tools, because the skill being trained is one skill. The cross-reference convention surfaces the cross-discipline function without giving Steelmanning two primary placements.
Chapter Note
The Foundation chapter articulates the dynamic revision discipline cleanly. The Update Protocol section names the Range pulls in both directions: what keeps the Foundation from drifting toward Control through refusal to revise, and what keeps it from drifting toward Decay through revising so easily that nothing holds. The diagnostic question — "What evidence would change my mind about this?" — is on the page. The practice instruction (articulate falsification conditions; write them down) is on the page. Steelmanning is explicitly named as the practice that bridges Foundation and Bond, with the individual-practice / relational-practice distinction articulated directly.
What the chapter does not yet articulate is the structural pairing between this category and Calibrating Confidence to Evidence. The chapter's treatment of confidence calibration is present but implicit, surfaced through the Epistemic Cowardice / Epistemic Arrogance failure-mode section and through the phrase "calibrated confidence" used as a posture-name rather than as a discipline-name. The static/dynamic pair as a structurally articulated set is a Workshop-level finding that the chapter has not yet carried as architecture. If a later pass through the chapter undertakes a structural revision, the static/dynamic pair surfaces as the two halves of evidence accountability inside the Foundation, and Calibrating Confidence to Evidence joins the chapter as a named structural discipline rather than as an attribute of the posture.