Claim ledger
You Can Win the Wrong Game for Years
You usually cannot tell whether you are playing the wrong game. Write down what would prove it, and by when.
18 claims · 11 verified to primary
What the essay claims
Effort is spent on the visible game while the outcome is set by a game one abstraction level up. That much is old. What makes it operable is that the misidentification is not random: the wrong game is reliably the more visible, more effort-rewarding, more culturally-praised one, so the felt sense of "this is productive" is a poor guide to whether you are on the right board — it is closer to a warning sign. But the frame only earns its keep if it can be wrong, and it can. In simple, rule-bound domains (chess, a standardised test) game-identification is trivial and execution quality dominates. And in some domains the crude, visible game is a sufficient statistic for the real one — next-token prediction was called the wrong game for intelligence, and scaling that crude objective worked — so the confident "wrong game" verdict is itself the targeting error, and the loudest critic is the one who misread the board. The move is therefore not "always look one level up." It is a test: has the deciding game actually moved, and would you know if it had not? Treat a suspected wrong game as a hypothesis to test, not a verdict — which is Law V as amended 2026-09-02.
The claim ladder
| Rung | Claim | Evidence | Scope limit |
|---|---|---|---|
| 1 | The pattern is real and cross-domain: effort goes to the visible game while the outcome is set one level up | Palmer (VERIFIED): the Inquisition was "obsessed with catching Lutherans and Calvinists … only executed one person for doing science" — effort aimed at the legible threat. Dalio (VERIFIED): "they think that they are betting on the technology when they buy the stocks … That's not true." Bolte Taylor (REPORTED, her account): fighting the feeling sustains it; the game is attention direction | Illustrative across domains; NOT a frequency claim. The neuroscience is her subjective account, framed as such |
| 2 | The misidentification is DIRECTIONAL, not random: the wrong game is the more visible, effort-rewarding, culturally-praised one — so "this feels productive" is a warning sign | The three cases share the signature: censoring the named heresy feels like defending the faith; grinding stock-picks feels like conviction; fighting anxiety feels like trying. Visibility bias + effort-as-progress + sunk cost | A tendency with a mechanism, not a universal law |
| 3 | REVERSAL / BOUNDARY: the "wrong game" verdict is itself a common wrong-game error — when the crude visible game is a sufficient statistic, or the domain is simple and rule-bound | Law V facet (b), amended 2026-09-02: next-token prediction was called the wrong path to intelligence; scaling the crude objective worked, and the confident critics committed the targeting error. Plus the Claim's own counter-evidence: chess/standardised tests, where execution dominates | This is the honest boundary; WITHOUT it the frame is unfalsifiable. Pre-committed futility rule requires this case to exist |
| 4 | THEREFORE run the game-check: name the deciding game; test whether it moved and whether you would know if it had not; hold "wrong game" as a hypothesis | Synthesis of 1–3 + amended Law V (a misfire is a search cost when detectable/re-aimable, a harm when not). The reader-runnable artefact; distilled to the free PDF card | A diagnostic the reader applies, not a guarantee; it tells you when to stop trusting the felt productivity, not that you are certainly on the wrong board |
What would make it wrong
The frame fails if: (a) in complex, ambiguous domains, execution quality consistently dominates game-identification — the boundary grows until it swallows the claim; or (b) the game-check cannot in practice separate a real wrong-game situation from a sufficient-statistic one, i.e. "has the deciding game moved?" has no operable answer, making the diagnostic unusable.
The evidence, row by row
Each row is a claim the essay makes, the words in the source that support it, where in the source they are, the day the source was read, and the status the check assigned. Verified means the primary was opened and read; Executed means we ran it ourselves; Reported means carried from a source we could not open in full.
-
History anchor (as used in body): by Palmer's count there were twelve trials of scientists over their science, Galileo's among them, and only one ended in execution (Giordano Bruno)
"There are 12 total trials of scientists about science. Galileo is one. Giordano Bruno is one. Giordano Bruno is the only one executed. Of those 12 trials, only three were convicted."
Ada Palmer, spoken, Dwarkesh Patel podcast · Transcript, chapter "The Inquisition accidentally invented peer review" (01:41:21), her reply to Dwarkesh on executions · dwarkesh.com
-
2026-09-13: the "Lutherans and Calvinists … only executed one person for doing science" sentence is the episode EDITOR BLURB, not Palmer's speech (the prior row-1 locator was wrong). The peer-review point IS also in her speech ("they effectively invented peer review", crediting Macuglia). Neither blurb wording is used in the body.
"The focus of the Inquisition is really misunderstood - it was obsessed with catching dangerous new heretics like Lutherans and Calvinists - it only executed one person for doing science."
Dwarkesh × Ada Palmer — EDITOR SUMMARY · Episode summary blurb, above the chapter list · dwarkesh.com
-
Inquisitors cared about the fine print of Protestant theology; a 1545 index of banned books promised to print arch-heretics' names in capitals and gave them to Calvin, Zwingli, Luther and Melanchthon; Machiavelli was not in capitals ("not important enough from their position"). Attributed in body to Palmer ("She points to…").
"What we need to worry about censoring is all of these fine minutiae of Protestantism." … "The 1545 edition of the Index of Banned Books says in its introduction, 'We shall put the names of arch-heretics in all caps.'" … "Machiavelli is not in all caps."
Ada Palmer, spoken · Transcript, same chapter, the censorship passage before the executions exchange · dwarkesh.com
-
Markets anchor: investors think buying the stocks is a bet on the technology; Dalio says it is not
"What a lot of people don't realize in bubbles is that through all technologies, they think that they are betting on the technology when they buy the stocks in the companies. That's not true."
Ray Dalio on the All-In Podcast (reported) · Article body · finance.yahoo.com
-
Dalio's mechanism: technology success and company survival are different behaviours; most early companies do not survive
"There's a giant difference between the behavior of companies and the behavior of the technologies. The norm is … a lot of companies won't survive in the start. Very small percentage."
Ray Dalio on the All-In Podcast (reported) · Article body · finance.yahoo.com
-
Mind anchor: the chemical surge of an emotion clears in about 90 seconds; what sustains it past that is re-triggering it by narrating — so the game is attention direction, not fighting the feeling. HER SUBJECTIVE ACCOUNT, not controlled research
"it takes less than 90 seconds" for the chemical rush of an emotion to surge and flush; "anything you continue to feel is your own choosing" (paraphrase of her framing; not controlled research per multiple summaries)
Jill Bolte Taylor, My Stroke of Insight (via Steven Bartlett podcast; shortform summary) · Book summary; her subjective account · shortform.com
-
Reversal/boundary anchor: next-token prediction was widely called the wrong game for intelligence, and scaling that crude objective produced strong capability — so a confident "wrong game" verdict can itself be the targeting error when the crude game is a sufficient statistic
(interpretive/argument claim; our archive's amended Law V case, not an external figure)
Internal — vault canon · worldview.md § Law V, facet (b)
-
SECOND reversal case (non-AI): a crude statistic the sophisticated baseball market underpriced turned out to track winning; the A's exploited the inefficiency and the market then corrected — the "look at the whole player" instinct was the misread
"These methods support Lewis's argument that certain baseball skills were valued inefficiently in the early part of this period, and that this inefficiency was profitably exploited by managers with the ability to generate and interpret statistical knowledge. Consistent with Lewis's story and economic reasoning, as knowledge of the inefficiency became increasingly dispersed across baseball teams the market corrected the original mispricing."
Hakes & Sauer, "An Economic Evaluation of the Moneyball Hypothesis", Journal of Economic Perspectives 20(3):173–186, 2006 · Abstract · aeaweb.org
-
The specific undervalued skill was on-base percentage / drawing walks (the paper's body finds OBP ~2× as important as slugging for runs and wins); traditional scouts derided the statistical approach in favour of gut, five-tool, whole-player judgement
(OBP-vs-SLG magnitude reported from the paper's body/AEA summary, not read in the abstract; the scouts-vs-stats clash is the documented Moneyball account)
Hakes & Sauer (body) + Michael Lewis, Moneyball (2003) · Body findings + Lewis · aeaweb.org
-
The LAG point (load-bearing for §4's honesty): next-token prediction looked unimpressive as a route to "understanding" for years and capability rose with scale over time (the GPT-2 → GPT-3/4 arc), so for a long stretch the sceptics' evidence looked like it supported them. Frame as the historical arc, NOT a sharp emergence-threshold mechanism
(interpretive/historical; the strong "emergent abilities at a threshold" framing is CONTESTED — Schaeffer et al. 2023 call it a metric artefact — so assert only the slow-then-capable arc, no hard threshold)
ML history — scaling-laws arc (Kaplan et al. 2020; GPT-2 2019 → GPT-4) · historical arc, not a cited number
-
Palmer's general conclusion: censors are always wrong, from our perspective, about what they should be worried about
"whatever they're looking at, they're always wrong, from our perspective, about what they should be worried about censoring"
Ada Palmer, spoken · Transcript, same chapter ("One fun game when I study the history of censorship…") · dwarkesh.com
-
Palmer's next NON-FICTION book is a history of censorship (body: "whose next non-fiction book is a history of censorship"; she also writes fiction)
"my next non-fiction book is gonna be a book on the history of censorship"
Ada Palmer, spoken · Transcript, same passage · dwarkesh.com
-
The 2002 Oakland A's won as many games as the 2002 Yankees (103 each)
"a record of 103–59" (A's); Yankees 103–58 in the standings table
Wikipedia, 2002 Oakland Athletics season · Infobox + standings table · en.wikipedia.org
-
The Yankees' 2002 payroll was about three times the A's (A's ~$39.7M vs Yankees ~$125.9M Opening Day)
Opening Day: A's $39,679,746 · Yankees $125,928,583 (ratio 0.315); Aug 31 ESPN/AP: $41.9M vs $133.4M (0.314)
Doug Pappas / SABR 2002 payroll table; ESPN/AP · 2002 payroll table · roadsidephotos.sabr.org
-
Prior art credited in body: Annie Duke's kill criteria pair a state and a date
"Kill criteria, generally, include both states and dates"
Annie Duke, Quit (2022), via Behavioral Scientist excerpt · Article body, kill-criteria section · behavioralscientist.org
-
The next-token sceptics argued the approach lacked understanding/meaning (dated examples)
Marcus on GPT-2 (The Gradient, 2020-01-25); Bender & Koller, "Climbing towards NLU" (ACL 2020)
Marcus; Bender & Koller · Paper abstract · aclanthology.org
-
Many companies that went public on the internet wave in 1999 did not survive the crash (body: "a great many", no figure)
Webvan (IPO Nov 1999, bankrupt Jul 2001) and eToys (IPO May 1999, Chapter 11 Feb 2001) among CNBC's failed dot-com IPOs; ~4,800 of 7–10k late-90s online ventures sold or gone by mid-2003
CNBC, "Failed IPOs of the Dot-Com Bubble" · Slideshow entries · cnbc.com
-
Palmer quotes inquisitors' letters: no need to censor Lucretius (only learned people can read it); what needs censoring is "all of these fine minutiae of Protestantism"
"We don't need to bother censoring Lucretius. Only learned people can read it" … "What we need to worry about censoring is all of these fine minutiae of Protestantism."
Ada Palmer, spoken (paraphrasing inquisitors' letters) · Transcript, censorship passage (chapter at 01:41:21) · dwarkesh.com
Cite this
- A single claim
"[claim text]" (Floyd, Harry, 2026, https://durabilitycurve.com/claims/the-wrong-game/)Replace the bracket with the row's claim text. The page URL carries the source; the row's own source link is in the row.- The essay
Floyd, Harry (2026). You Can Win the Wrong Game for Years. The Durability Curve. https://durabilitycurve.com/blog/the-wrong-game/- This ledger
Floyd, Harry (2026). Claim Ledger: You Can Win the Wrong Game for Years [structured claims with sources]. The Durability Curve. https://durabilitycurve.com/claims/the-wrong-game/
Quote with attribution and a link to this page or the essay. Say if you changed the wording. Not licensed for model training. Plain-text copy for machines: /md/claims/the-wrong-game.md.