The Atlas
10-node core path · 28 reference notes
Atlas helps operators judge where AI changes work, where it merely shifts cost, and what to inspect before buying, deploying, or trusting it. Start with 001-010 as the core path; use 100-127 as concept deep dives and reference notes.
Core path — 001-010
- 001By the Time Anyone Looked, the Option Had Closed.The Recognition Thesis
For time-sensitive opportunities and exposures, noticing can be the first constraint.
- 002When Uncertainty Pays YouConvexity
Survivable exposure, payoff curvature and a worthwhile price require separate judgments.
- 003The Most Valuable Thing an AI Can Say Is "I Don't Know"The Invisible Frontier
A refusal becomes useful when its error rate, coverage and next owner are defined.
- 004The AI Wins You Can Measure Are Table StakesSelection
A productivity gain can be useful even when rivals can match it. Test costs and who keeps the benefit before scaling.
- 005Cheap to Make, Expensive to CheckThe Verification Economy
An offload saves work only if generation, relevant checking, and likely repair together cost less than producing an equally reliable result by hand.
- 006The Reverse CentaurWho Really Decides
Human approval can conceal a reverse centaur when the people checking a campaign cannot challenge the objective it optimizes.
- 007Who Drew the Menu?Agenda Control
Choosing from a menu is different from controlling which alternatives reach it. The construction record can help a reviewer question the boundary and seek authority to reopen the search.
- 008When a Company Wiki Becomes a Graveyard, Fix the ClockKnowledge Escrow
Preserve incident evidence during work. Prepare it for a later permitted question, then test what another person can do with it.
- 009The Cell Is Not a DecisionRisk, Measured Wrong
A risk matrix can route a hazard for review. Accepting the residual risk requires an accountable decision.
- 010The Quadrant Is a Price, Not a PlaceThe Map Redraws Itself
Cheaper watching and handling can change which work becomes routine. The event's frequency, stakes and need for judgment may remain unchanged.
Concept deep dives — 100-127
Reference notes for terms, models, and source ideas used by the core path. They support the spine; they are not a separate theme.
- 100The Shadow Option: The Move Hidden in What You Already Built
In Bowman and Hurry's framework, a shadow option is a possible strategic move embedded in existing resources and awaiting recognition.
- 101The Recognition Engine: Continuous Cheap Attention on Discontinuous Payoffs
A recognition engine keeps a recurring watch on named sources and brings documented changes to an owner while there is still time to respond.
- 102The Frequency-Impact Matrix: A Triage Grid Dressed as a Ruler
A frequency-impact matrix sorts risks into categories. Its bins can hide different loss profiles; a consequential decision still needs assumptions, tested controls and an accountable owner.
- 103Verifiable Space: Where the Answer Can Be Checked
Verifiable space is work where a valid check can evaluate a proposed answer within the available budget and time.
- 104The Prevention Paradox: The Disaster You Stop Never Arrives to Thank You
Three quiet years do not establish preventive capability or an avoided loss. Rose's population paradox offers a related analogy; a firm's renewal decision still needs evidence.
- 105Latent Exposure: The Loss You've Already Agreed To
A latent exposure is an unrecognised dependency, contract term, or knowledge concentration that can turn today's operating setup into a future loss. Accepting the dependence does not mean choosing the loss.
- 106Knightian Uncertainty: When No Defensible Reference Class Supports the Odds
Knightian uncertainty names a limit to the quantitative basis for a decision.
- 107The Reverse Centaur: The Machine Holds the Reins, You Hold the Pen
A reverse centaur is an arrangement where people execute or approve software-issued work without an effective way to challenge its targets or terms. People and institutions remain responsible for adopting those rules.
- 108The 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.
- 109Range Compression: Five Buckets, a Thousandfold Gap
Range compression discards differences within a category. Here, fictional annual risks with probabilities 1,000 times apart share one score under explicitly chosen bins.
- 110The Real Option: A Right, Never an Obligation
A real option preserves the choice to make a future investment without requiring the firm to make it.
- 111The Option Chain: One Move Opens the Next
Buying a customer book can reveal a supplier opportunity. Here, a paid reservation secures a delivery right; exercise starts a separate operating commitment.
- 112Convexity: When Volatility Starts Paying You
Convexity is a property of payoff curvature over a stated range. A call option is one example; a loss cap alone does not establish it.
- 113The Jagged Frontier: The Edge You Can't See Until You're Past It
A jagged frontier is an uneven boundary of model capability. Similar-looking tasks can produce opposite effects from AI assistance.
- 114Out of Distribution: A Map of Where the Data Lived
Out of distribution describes a case relative to a named reference. Model reliability still needs evidence for the task.
- 115The Base Rate Trap: When the False Alarm Rate Decides
A rare event can leave most alerts false even when a detector rejects 99% of non-events. Choosing a usable threshold also requires recall, investigation capacity and the costs of both errors.
- 116Metacognitive Demand: The Load Shifts from Doing to Judging
Delegating production can leave evaluation demands that exceed the operator's knowledge. A marketer may judge a headline yet need help testing an analytics tag.
- 117Prediction vs Decision: The Machine Forecasts, You Judge
A forecast estimates what may happen. Acting on it also requires an objective, an error tradeoff and authority. Systems can execute a configured rule; the forecast alone does not supply those conditions.
- 118Differentiated Value: The Win Everyone Can Buy Is Not an Edge
A workflow gain can save a firm money without giving it an advantage over rivals. An illustrative support ledger separates realized savings, costs and competitive claims; adoption and productivity statistics do not supply its amounts.
- 119Rare for You, Common for the Corpus
A case can be unfamiliar to its owner and documented elsewhere. The comparison still has to hold.
- 120The Generation-Verification Gap: When the Right Answer Loses
A correct answer can be present among generated candidates while the selector returns a wrong one. Measuring that loss requires the same tasks, candidate sets and correctness standard.
- 121The 70% Problem: Fast to Prototype, Hard to Finish
A fast prototype can leave production work that requires different checks and expertise. Addy Osmani's 70/30 is a field heuristic, not a measured share of a job.
- 122The Three Lines Model: Independence Is a Role, Not a Headcount
The Three Lines Model distinguishes management's action and challenge from internal audit's independent assurance to the governing body. Rechecking your own decision cannot create that independence.
- 123Tacit Knowledge: We Can Know More Than We Can Tell
Tacit knowing integrates cues into recognition and action without an exhaustive account of how they combine. A record can support learning; a stand-in still needs to demonstrate performance.
- 124The Illusion of Choice: A Menu Hides the Decision Better Than a Verdict
Here, illusion of choice names the risk of treating a click as proof of control over how a menu was arranged. It is an editorial diagnostic, not a validated measure of felt authorship.
- 125Agenda Control: The Draw Shapes the Pick
Agenda control is power over which alternatives reach the table. A genuine choice can remain confined to that list unless someone can reopen its boundaries.
- 126Deskilling: The Skill That Catches the Error Is the One Automation Quietly Takes
Deskilling is a loss of unaided capability. Automation can remove opportunities to practise while leaving difficult monitoring and recovery tasks to people. Whether that capability has changed needs evidence.
- 127Affordable Loss: Size the Bet by What You Can Lose
Affordable loss limits a commitment to what the decision maker can survive losing under uncertainty.