明,可测:一把结局盲、免效标、跨领域的判断力之尺 Lucidity, Made Measurable: An Outcome-Blind, Criterion-Free Cross-Domain Scale of Judgment

A leader's success or failure is universally attributed to "judgment," and yet judgment still has no clean ruler. Within-domain hard metrics are not portable, cross-domain fame measures the echo of fate rather than the mind, any assessment made after the outcome is known is entangled with hindsight, and judgment itself has no ground truth. Judgment, in short, can be discussed after the fact but not measured before it. The two portraits below are not evidence of anything; they simply put the gap on the table. Talent and judgment are different things, and to speak of judgment one first needs a ruler that measures judgment alone, without looking ahead to how the story ends.

Napoleon. In raw ability almost no one compares. He rebuilt France's administration, law, and finance, and the Code civil still runs across much of Europe two centuries later; on this paper's scale his skill is 89 (peak 90). Yet the same man marched on Russia in 1812 against all counsel, lost six hundred thousand men, repeated the error at Waterloo, abdicated twice, and died in exile on St. Helena with a great deal of time to reconsider (personal survival = 28). The atlas annotation puts it exactly: this was "a failure of lucidity, not a failure of ability." His skill was intact throughout, still 72–82 in the Russia and Waterloo phases; what collapsed was lucidity (peak ≈ 79, down to 16 in the Russia phase). The institutions he built survived (organizational survival = 70); he did not.

Washington. He was far less gifted: his pooled-entry skill is only 58, and the record describes a man who "lost battle after battle yet kept his army intact, wearing the British down through endurance rather than tactical genius." But he handed back power where he could have kept it, stopping at two terms (lucidity ≈ 84). His personal ending is at the ceiling (personal survival = 96), and his enterprise endures too (organizational survival = 96). As the annotation notes, the highest personal-survival reading in the whole field belongs to the man whose statecraft was not top-tier but whose restraint was.

The gap. The two men's talent diverges sharply (89 vs 58), while their lucidity diverges the other way (≈ 64 vs 84); the question this paper asks is whether a ruler can register that second gap without reasoning back from how each life ended. Talent, it turns out, guarantees nothing about how a life ends. The difficulty is that, without a ruler that does not reason backward from the outcome, "Washington was more lucid than Napoleon" is forever just hindsight: impossible to measure on its own terms, compare across domains, or falsify. This paper therefore does not try to predict who will succeed; it first builds a ruler for judgment and shows that the ruler measures something real. The portraits are motivation, not evidence; their relation to outcomes (skill enterprise-tilted, lucidity fate-tilted) is deferred to §5 as a sanity check on the ruler, not as the headline.

The two portraits are not a special case. Widen the lens to a dozen figures across four domains, and the pattern holds (Figure 1): eminence compresses skill toward the high end (the full corpus averages skill 84.5, SD 12.2, and runs 30–99, versus lucidity's wider spread at mean 62.3, SD 18.2), while lucidity fans out across the whole range. Newton at 43 and Darwin at 84 are both first-rank scientists; Alexander at 28 and Genghis at 73 are both undefeated conquerors. Skill is the threshold for entering the atlas at all; lucidity is what actually varies, and it varies across domains on one shared axis, which is the entire point of a cross-domain ruler.

Definition. Lucidity, written 明 in the source rubric, is the product of two capacities: Pattern (理), how accurately one models the world, and Mystery (玄), how well one keeps company with the parts of the world that cannot be modeled. Formally, lucidity is the geometric mean of the two, √(Pattern×Mystery) rescaled to 0–100, and each branch carries three axes scored 0–5:

The product rule means that if either factor goes to zero, lucidity goes to zero, because "one who believes he sees through everything is precisely the one who is not lucid." Lucidity is thus not a sum of six axes but the product of two capacities: seeing accurately, and knowing where one cannot see.

Structure: not one blob but two semi-independent sub-dimensions. Effective dimensionality is ≈ 2. Pattern and Mystery are highly correlated (r ≈ 0.80) yet each retains 27–37% mutually independent variance, so they are genuinely two sub-dimensions rather than two names for one, a claim supported by the validation evidence in App. A and §4.

Lucidity is not wholly new; each of its six axes correlates strongly with an existing construct. The wedge is not "a new dimension" but the mode of measurement. Existing constructs are almost all measured by self-report or subordinate rating, and are thereby contaminated by impression management, social desirability, and the rater's knowledge of the outcome. Lucidity, by contrast, is inferred from public biographical behavior and is blind to both identity and outcome, which is the source of its hindsight-resistance, and of its ability to put Newton and Napoleon on the same axis at all.

Overall, lucidity is closest to practical wisdom (Sternberg 1998; Grossmann 2010), from which it is distinguished on three counts: behavioral inference, outcome-blindness, and cross-domain commensurability. The empirical evidence for discriminant validity (the sub-dimensions each retain 27–37% independent variance; surface features explain only 9%) is in §4. One caveat: Irreducibility Reverence has no mature self-report scale to benchmark against, which is both where the measurement novelty is most concentrated and where human-rater convergence (R4) is most needed.

Upper-echelons theory (UET) is the natural home for this instrument, and also the source of the gap it fills. Hambrick & Mason (1984) argued that organizational outcomes are "reflections of" the cognitions, values, and perceptions of top executives, but conceded in the same paper that these psychological constructs are difficult to observe directly, and therefore proposed observable demographic characteristics (age, tenure, functional background, education, socioeconomic roots) as proxies for the underlying cognition. That single methodological move defined the empirical program for four decades (reviewed in Carpenter, Geletkanycz & Sanders 2004), and it drew an equally durable critique: demographics are a black box, correlated with outcomes for reasons the theory cannot see inside (Lawrence 1997), a limitation Hambrick (2007) himself acknowledged in calling for measures "closer to the psychological constructs they are meant to represent." The subsequent managerial-cognition turn (Helfat & Peteraf 2015 on managerial cognitive capabilities; Eggers & Kaplan 2013) reopened the box conceptually but still lacks a direct, cross-domain, hindsight-free measure of the cognition itself.

This is precisely the slot the ruler is built for. Where UET substitutes demographics for cognition, lucidity scores the cognitive construct directly; where its proxies are population-bound (a CEO tenure variable does not travel to generals or scientists), lucidity puts Newton and Napoleon on one axis; and where demographic and reputational proxies absorb the outcome they are meant to predict, lucidity is scored outcome- and identity-blind. Table 4 lays the four measurement traditions side by side.

Read together with the adjacent literatures, the same gap recurs: the hubris literature (Roll 1986; Hayward & Hambrick 1997; Hiller & Hambrick 2005) lacks a blind longitudinal operationalization; practical wisdom lacks an automatically computable measure; and succession and derailment (McCall & Lombardo 1983; Shen & Cannella 2002; Finkelstein 2003) have never been distinguished from organizational survival (Stinchcombe 1965; Marquis & Tilcsik 2013) inside one framework. The undercurrent throughout is outcome bias (Fischhoff 1975; Baron & Hershey 1988): assessment usually happens after the outcome is known, so lucidity is hard to measure cleanly, and the outcome-blind design here is meant to answer exactly that. The paper's position is therefore modest and specific: it does not propose a new theory of leadership but supplies the measurement instrument UET has wanted since 1984, so that cognitive variables the literature "wants to control but cannot measure" finally admit measurement, comparison, and falsification.

Scoring protocol. Scoring is done by a language model from public biography alone. Each life is coded as a timeline of two or three consequential phases; per-phase lucidity, √(Pattern×Mystery) rescaled to 0–100, is cost-weighted, and skill takes the peak across phases. Two hard rules hold throughout:

The properties on which credibility rests. Lucidity has no external criterion, so it is not validated by computing error against an answer key, which would only measure agreement. It is an opinion instrument, not a truth machine, and its credibility comes from the five criterion-free validations and three pre-specified experiments of §4. Only three numbers are needed here: surface features explain just 9% of the score; named-minus-blind drift is ≤ 0.15 points per axis; and inter-rater reliability is high (ICC(2,1) 0.73–0.93 single-rater, up to 0.98 averaged across raters; full ranges in §4), all between model raters, not convergence with humans (see R4). That "lucidity is outcome-blind" is a design claim (the Consequential Foresight axis reasons only from contemporaneous information), and its empirical verification is in §4 and App. E.

Cross-cohort scale (code-audited). The pooled regression reads each cohort's hand-authored raw axis values, with no normalization. Lucidity is consistent across cohorts on a shared fixed rubric, the axis the paper leans on most, and the only fully clean one. Skill and institution-building are directionally consistent but domain-relative in unit. Orientation is the problematic axis. It is pooled across incompatible pole conventions — rulers' "inward self-preservation ↔ outward projection" versus presidents' and CEOs' "self-dealing ↔ stewardship." The conventions disagree so sharply that Washington's single act of stepping down scores 43 under one and 95 under the other, so its organizational-survival coefficient cannot be read as a single construct. It is ≈ 0 on personal survival and non-core, with handling deferred to R0. The net judgment is that the skill/organizational-survival relation on which the core corroboration depends is unaffected by orientation.

Sample. The corpus is a hand-coded and model-scored panel of leaders: 459 carry lucidity, of whom 404 carry both outcomes, spanning twenty cohorts. Whether non-organizational leaders such as scientists and artists enter the main analysis depends on journal placement (agenda S2); the conservative option restricts the main analysis to political, military, and business leaders and moves the rest to a generalization appendix. Two fictional cohorts are excluded from the regressions. Variable coding and the full regression tables are in App. D.

The instrument is a human-authored rubric applied by a large language model; being explicit about which is which is a precondition for trusting the numbers. The construct, the six axes, their anchors, the two hard rules, and the aggregation formula are all authored by hand and fixed in advance. The model's job is narrow: read a biography and return an axis score, following the rubric. Three passes run over the same public source text (Figure 3). The atlas pass scores each phase on the twelve rubric components with the figure's identity visible; the R2 blind pass first redacts identity and outcome, then an independent judge re-scores the six axes; the R3a objective pass ignores the rubric entirely and extracts records-based facts, quote-anchored and blind to the predictor scores.

This is the paper's most transferable contribution. Judgment has no ground truth, so every "score accuracy against a criterion" route fails outright: one cannot validate lucidity against an answer key, because there isn't one. We therefore change the question. Instead of asking "are the scores correct," we ask "does this set of scores have every internal property a credible measurement instrument should have."

Measurement philosophy: no answer key, so none is invented. Lucidity has no external criterion; the hand-authored reference set is merely "the R+1-th rater," never an answer key; and computing MAE or ρ against "gold" is void, since it measures agreement rather than correctness. A distinction must be kept. The measurement layer (the criterion-free internal validation of this section) is one thing; lucidity as an independent variable predicting real outcomes is another, a criterion-bearing empirical claim to be defended separately via R2, R4, and out-of-sample prediction, and not under the criterion-free exemption. This is exactly why §5 keeps the outcome relations to corroboration and nothing more.

The depth battery comprises 20 subjects × 3 raters × {named, blind}, for 120 scorings, spanning the knowledge, outcome, and faith domains (though not yet the regression's workhorse cohorts such as CEOs and central bankers; see R6), alongside 12 synthetic biographies. Each of the five validations answers one property a good ruler should have:

Beyond the criterion-free checks, three pre-specified experiments actively try to catch the ruler measuring something other than what it claims (pre-specified = the design, failure line, and analysis were fixed in writing before the scores were seen; they are not lodged in a public registry, so we do not call them pre-registered):

Connect lucidity to the two outcomes (personal survival and organizational survival) of nearly four hundred leaders, and a sensibly behaving ruler should show structured, interpretable relations rather than noise. It does, and the pattern is the one theory would predict.

First, a within-cohort look. Before the pooled regressions, hold the domain fixed. Among US presidents, skill and lucidity come apart sharply within the single office (Figure 6): Lyndon Johnson (skill 95) and Andrew Jackson (skill 88) are among the most capable operators ever to hold the office, yet score low on lucidity (33 and 23); Eisenhower sits high on both; and the high-skill / low-lucidity corner is populated while its mirror is nearly empty. This is the two-portrait contrast reproduced inside one cohort, where domain, era, and role are held roughly constant, so it cannot be an artifact of comparing generals with CEOs.

Corroboration one (robust across named and blind): skill is enterprise-tilted only. Skill's coefficient is β = +0.22 (t = 6.9) on organizational survival and β = −0.03 (n.s.) on personal survival; the difference, Δ = 0.25 (subject-clustered z = 5.46, p < 0.001), is consistent across four SE specifications and, on the subject-clustered test, strengthens after the rescore (z = 5.5 → 6.0). Skill builds the organization but not the self.

Corroboration two (revealed by the blind rescore): lucidity is fate-tilted, within a coding boundary. With named scores lucidity is significant at both ends and the fate-tilt is not significant (Δ = −0.07, p = 0.21), the halo-contaminated, conservative reading; and across 16 specifications the named fate-tilt sits at a median of ≈ 0 and keeps its (negative) sign in only about half of them (App. D), so before blinding it is not a stable effect at all. After the outcome-blind rescore (identity and outcome cues deleted, the whole pool independently re-scored), lucidity's personal β rises to +0.34 (t = 8.3), its organizational β falls to +0.08 (t = 2.4), and the fate-tilt becomes Δ = −0.26 (subject-clustered z = −4.7, cohort-clustered z = −3.8, p < .001); drop any single cohort and the subject-clustered z stays between −4.0 and −5.0. This is not an artifact: had the rescore merely added measurement error, it would have pulled every coefficient uniformly toward zero, whereas here the opposite happened: lucidity's link to organizational survival collapsed and its link to personal survival strengthened, the signature of removing a specific confound (the "famous, large legacy, therefore surely lucid" halo). But the coding-dependence must be stated flatly: when outcomes are replaced with purely objective facts (a manner-of-death ordinal plus institution-survival years, 459 figures blind-extracted), R3a finds that the fate-tilt does not replicate: lucidity on objective fate is β = +0.08 (t = 1.50; t = 0.99 dropping the still-living), and the tilt difference is z = −1.37, p = .17, directional only. Corroboration two is therefore confined to named and blind (expert-style) coding and does not extrapolate to the coarse objective proxy (method and full table in App. F; mechanism in §6).

The overall coordinate: institution-building dominates both ends. Its coefficients are ≈ .5 at each end, the difference is not significant (z = 0.37), and its dominance shares are 56% and 48%, first at both ends. Skill and lucidity are the two wings of a dissociation — robust for skill, coding-dependent for lucidity (§7) — while institution-building carries both, the master beam spanning the structure. R3a shows the master beam is not common-source (institution-building scores t = 3.7 on objective survival and t = 8.5 on objective fate); of the two wings, only the skill-tilt carries through to the objective operationalization, while the fate-tilt does not (see §7 and App. F).

Collinearity is not the culprit. All VIF fall in [1.03, 1.71], so skill's near-zero coefficient on personal survival and blind-lucidity's near-zero coefficient on legacy are real signals, not artifacts.

The main contribution: turning "unmeasurable" into "measurable." Leader cognition (judgment, lucidity, hubris) is invoked constantly in the upper-echelons and hubris literatures, and has always lacked a direct, cross-domain, non-hindsight measure. This paper supplies one: a scale of judgment that is blind to outcome and identity, computable across domains, and defensible without a criterion. Its significance is not that one more predictor has been found, but that it opens a body of research that biography alone could never support.

Enabled questions. With computable lucidity, questions that could only be discussed qualitatively can now be tested longitudinally, cross-domain, and blind. Do boards betting on talent versus on lucidity get systematically different succession outcomes? Can leader hubris (the negative pole of Ego Undistortion and Boundedness Awareness) be read before a crisis rather than diagnosed after? Is lucidity trainable, and along what trajectory? Each of these shares one prerequisite: that judgment be measurable at all.

Methodological spillover: removing hindsight sharpens the picture. A finding here holds for the whole tradition of assessing leaders. Measuring judgment is trapped by outcome bias, because assessment usually happens after the outcome is known. The outcome-blind rescore does not weaken the signal but sharpens it: lucidity's link to organizational survival collapses while its link to personal survival strengthens, the pattern expected if a halo, not the construct, drove the former. That yields a reusable test: if removing hindsight weakens the signal, the signal was hindsight; if it strengthens, the instrument is measuring the mind. The logic does not depend on this particular construct and transfers to any leader-cognition measure that must be cut loose from hindsight.

A reminder on dependent variables. The ruler also exposes a construction problem easy to miss in upper-echelons work: folding "did the person end well" and "did the organization survive" into a single "performance" erases, in one stroke, the distinct payoff profiles of talent (enterprise-tilted) and lucidity (fate-tilted). Napoleon is the limiting case: top talent, an enduring legal code, and death in island exile. But this is corroboration, not a claim; its strong causal form rests on R4.

Judgment is universally treated as decisive to a leader's fate, and yet has long been discussable only through the halo of hindsight, because there was no ruler that refused to reason backward from the outcome. This paper builds and validates such a ruler: lucidity, decomposed as Pattern ⊗ Mystery into six operational axes, outcome-blind, identity-blind, and computable across domains. It then shows how to defend the ruler when there is no ground truth: through coverage, structure, reliability, invariance, and information, plus three pre-specified adversarial experiments. The ruler takes a thing everyone calls central, yet historically unmeasurable, and makes it computable, comparable, and falsifiable.

Connected to nearly four hundred leaders, the ruler behaves sensibly: the structural spine is institution-building, and a tilt of lucidity toward personal fate appears once hindsight is removed. But this is a sanity check, not a claim, and it carries one boundary worth stating plainly: replace the outcomes with purely objective facts and the fate-tilt yields to institution-building. The claim throughout is a measurement one: a ruler for judgment, not a model that predicts fate. Its value is that a whole class of "we wanted to measure it but couldn't" questions in upper-echelons and leader-cognition research now has an instrument to reach for.

A.1 The six scored axes (canonical names as displayed in the LucidiAtlas; machine keys in italics). Pattern (理): Consequential Foresight (secondOrderSight — foreseeing the knock-on consequences of one's own decisions, reasoning only from contemporaneous information) · Ego Undistortion (egoUndistortion — accuracy of assessing one's own ability and limits, not modesty) · Signal Discrimination (signalDiscrimination — telling real regularity from noise). Mystery (玄): Confidence Calibration (calibratedUncertainty — confidence tracking evidence) · Boundedness Awareness (boundednessAwareness — lucidity about a framework's boundary) · Irreducibility Reverence (irreducibilityReverence — letting the irreducible core remain suspended rather than forcing a false closure). Six scaffold lenses (Doctrinal Non-Capture / Dissent Permeability / Affect Undistortion / Other-Mind Accuracy / Frontier Openness / Contingency Acknowledgment) only aid reasoning and are not scored; the reduction from 12 to 6 axes is justified by effective dimensionality ≈ 2.

A.2 The two hard rules. The amoral firewall (credit only cognitive accuracy; a pathological breakdown is still scored by the stakes at the time) · the anti-halo clause (a uniform exchange test strips out identity information).

A.3 Four auxiliary outputs. Coverage κ · record type · evidentialBasis · cost-weighting.

A.4 Measurement philosophy. Lucidity has no external criterion; the hand-authored reference set is only "the R+1-th rater"; computing MAE or ρ against "gold" is void. The distinction to keep: the measurement layer is criterion-free internal validation; lucidity as an independent variable predicting real outcomes is a criterion-bearing empirical claim, defended separately via R2/R4 and out-of-sample prediction, not under the criterion-free exemption.

A.5 Worked exemplars from the top cohorts. A profile reads more plainly than a definition. Table 6 lists the six axis scores (0–5) and the aggregate lucidity (0–100) for eight figures drawn from the five highest-scoring cohorts; every value is an open-dataset score (axes_long.csv). Two regularities stand out. First, at near-equal skill lucidity still varies widely: Feynman and Heisenberg are both first-rank physicists (skill 97 and 95), yet score lucidity 88 and 57. Second, which axis gives way is diagnostic of how the lucidity was lost. Figure 9 sets two physicists of near-equal skill side by side: Galileo's Signal Discrimination is near the ceiling (he read the science correctly), yet his Consequential Foresight and Ego Undistortion sit low (the confrontation he neither foresaw nor stepped back from), whereas Feynman is high on all six. Lucidity is not a scalar a figure "has"; it is the shape of a profile, and the aggregate only summarizes it.

These four views render the open dataset the way the interactive atlas does, so a reader can see the distributions the regressions summarize. Colour is used the same way throughout: the gold iso-lucidity bands in Figure 10 are lines of constant lucidity; the red→amber→teal scale in Figures 11–12 is the atlas's ending language (red = collapse, teal = a durable end).

B.1 The two coordinates of lucidity: Pattern × Mystery. Lucidity is a pair, not a scalar. Figure 10 plots the two factors it is built from — Pattern (how accurately the world is read) against Mystery (how accurately the limits of the knowable are held) — for the 317 figures scored on the full construct. The faint gold contours are iso-lucidity bands: because lucidity = √(Pattern·Mystery), a line of constant lucidity is the hyperbola Pattern·Mystery = c², and the shading deepens toward the top-right where both factors are high. The dashed diagonal is the balance line Pattern = Mystery; the region below-right of it (Pattern outruns Mystery) is the brilliant-but-closed reader — sharp sight, thin reverence.

B.2 The two capability axes: Skill × Institution-building, coloured by ending. Figure 11 places the two hand-coded capability variables against each other and colours every point by its ending (the atlas's red→teal scale). The upper-right — high skill and high institution-building — is where the durable (teal) endings concentrate; raw skill without an institution behind it (lower-right) collapses to red as often as not. This is the descriptive shadow of the regression result that institution-building, not skill, is the master beam.

B.3 Lucidity × ending, with the outcome bands drawn in. Figure 12 is the single relation §5 leans on for nomological corroboration: lucidity against personal-and-organizational ending. The horizontal bands are the atlas's outcome zones (collapse / mixed / durable). The relation is real but loose (r ≈ 0.50): high lucidity rarely ends in the red band, but the spread is wide — the ruler measures judgment, not fate, exactly as the measurement-first framing insists.

B.4 Mixed cohorts on one axis: Skill × Lucidity. Figure 13 overlays four very different cohorts on the shared skill–lucidity plane. They stack almost vertically at the high-skill edge, yet occupy different lucidity bands: physicists sit high (mean lucidity 80), commanders in the middle (66), revolutionaries low (48), with financiers spread across. Skill barely separates them; lucidity does — the cross-domain commensurability claim, shown rather than asserted.

B.5 The lucid elite of sixteen fields: one summit, scattered fates. Figure 14 takes the most lucid figure from each of sixteen cohorts — seventeen in all, since the arts, uniquely, contribute two ceiling-level figures (Beethoven and Bach) — and places them on the same skill–lucidity plane, encoding two further variables at once: marker shape is the broad domain family (thought, arts, power & state, enterprise & finance), and marker colour is the figure's ending on the atlas's red→amber→teal scale. Two things read at a glance. First, the field-toppers pack into a narrow ceiling — skill 72–99, lucidity 81–91 — regardless of domain, which is the cross-domain commensurability claim in its strongest visual form: a founder of a religion, a mathematician, a central banker, and a liberator occupy one small corner of the plane. Second, and against any temptation to read lucidity as a fate-predictor, their endings fan across the entire range within that same corner: Darwin and Buddha end in the deep teal (95, 90), while Socrates — tied for the highest lucidity in the whole atlas — ends at 5, the hemlock. Shape shows judgment travels across domains; colour shows it does not buy a soft landing.

B.6 The two capacities that carry the structure: Institution-building × Lucidity. Figure 15 plots the master beam (institution-building) against the fate-tilted wing (lucidity) for all 404 leaders, coloured by ending. The two are positively but only loosely related (r ≈ 0.46, about a fifth of the variance shared): high institution-building neither requires nor implies high lucidity. Both off-diagonal quadrants are populated — 79 leaders build lasting institutions without commensurate lucidity (lower-right, e.g. Wagner), and 70 are highly lucid yet leave little institution behind (upper-left, e.g. Socrates and Feynman). This is the discriminant-validity case for treating the two as separate capacities, and the visual root of the dissociation-and-master-beam structure of §5: precisely because they vary semi-independently, they can load differently on the two outcomes.

B.7 The twenty cohorts at a glance. Table 7 collapses the whole atlas to one row per cohort, sorted by mean lucidity. Two columns are heat-shaded on the atlas's own scales — Luc on the white→gold lucidity ramp, End on the red→amber→teal ending ramp — so the cross-cohort structure reads at a glance. Lucidity ranges from physicists (80) down to revolutionaries (48); skill is high and nearly flat across almost all cohorts, which is precisely why skill cannot be doing the discriminating work.

B.8 The identity signal, measured then removed: named vs blind lucidity. Figure 16 plots each figure's named lucidity (scored with the name visible) against its outcome-blind lucidity (scored from a redacted biography) for the 397 fully re-scored figures. Two facts sit together. The two agree on average — both means are 61.1, so naming adds no aggregate inflation — yet they disagree per figure by about 11 points on average (r = 0.68), and the fit line (slope 0.49, flatter than the dashed 45°) shows the blind scores shrink toward the mean: naming stretches the tails. This is the halo made visible and quantified — the reason §5 and Table 2 run the fate analysis on the blind scores rather than the named ones.

B.9 Reputation is not a soft landing: legacy vs ending. Figure 17 plots posthumous legacy (how the world came to rate the work) against personal-and-organizational ending. They are positively related (r ≈ 0.58) but far from identical: the vertical scatter at every legacy level is wide, and the upper-left — towering legacy, hard ending — is well populated. Being remembered well and ending well are correlated, separable things, which is why the atlas scores them as two variables rather than one.

B.10 The shape of lucidity: a six-axis profile. Figure 18 draws the mean of each of the six axes on a radar — the three Pattern (理) axes on the upper-right, the three Mystery (玄) axes on the lower-left. The pooled profile is not round: Pattern (mean 3.2–3.3 of 5) consistently outreaches Mystery (2.5–3.0), and the single lowest axis across the whole corpus is Irreducibility Reverence (2.54) — these figures, on average, read the world better than they hold its limits. Two exemplars make the asymmetry concrete: Socrates (teal) balloons on the Mystery side; Napoleon (brick, dashed) is the mirror image — strong on Pattern, thin on Mystery, the brilliant-but-closed reader.

Table 8 is the complete atlas in one compact index — every one of the 459 figures with its full metric line, grouped by cohort (cohorts ordered by mean lucidity, figures within each cohort by lucidity). Two columns carry the atlas's colour language: Luc is shaded on the white→gold lucidity ramp (deeper gold = more lucid), and End on the red→amber→teal ending ramp (red = collapse, teal = a durable end). Reading down the two heat columns together is the paper's thesis in one view: the gold of lucidity and the teal of a good ending are correlated but visibly independent — Socrates and the other deep-gold names are scattered across every ending colour, red included. Column keys: Coh cohort, C century (negative = BCE), Sk skill, In institution-building, Or orientation, Luc lucidity, Df difficulty, End ending, Leg legacy.