祝融说。

The Sky on Christmas Night

December 25, 1758, nighttime, the village of Prohlis near Dresden. The farmer Johann Georg Palitzsch left the Christmas table and walked out into the yard — he had another identity: amateur astronomer, with a telescope mounted at home. There was one thing in the sky for which the whole astronomical world was waiting. He swept the telescope across the predicted region, found a small dim patch of mist, and checked the star chart: nothing was marked there. What he was looking at was a comet — not just any comet, but the comet whose return had been predicted. For that patch of mist Messier in Paris had already dragged his telescope through more than a year of overcast skies and ill health; on an autumn night a few months earlier, in Taurus, he had mistaken another comet-shaped patch for the quarry he was tracking — that patch was later numbered M1, the first entry of the Messier catalog, and the comet-hunter's error became the starting point of another field. This time, fate did not leave the priority to the professionals.

The prediction had been made more than fifty years before. In 1705 Edmond Halley published his Synopsis of the Astronomy of Comets, and, sifting the historical records, proposed that the comets of 1531, 1607, and 1682 had orbits almost identical: they were not three comets but one and the same star circling the sun on a closed orbit with a period of about seventy-six years; by this reckoning it would return around the end of 1758. Halley knew the weight of this prediction, and knew he would not live to see it redeemed — he died in 1742, sixteen years before the fulfillment. The astronomical world did not forget: Messier searched night after night from 1758 onward; Clairaut, in 1757 and 1758, worked the gravitational perturbations of Jupiter and Saturn into the time of return and corrected Halley's rough estimate — the comet in fact passed perihelion on March 13, 1759, within about a month of the corrected prediction. Posterity named the star Halley's Comet.

The structure of the event is worth recording in the ledger of Chapter One. What Halley did in 1705 was a public, signed prediction with an explicit deadline: he locked a path (one and the same comet, one and the same orbit, one and the same period), wrote out explicitly how he was to be tested, and then died. Fifty-three years later, a German farmer he had never met redeemed it in a yard on Christmas night. The predictor was not present; the verifier had no connection to him and no line of descent from him; between the two there was only one publicly locked path and one patch of night sky — this is the cleanest form the cross-subject test can take: even the interference of favor, authority, and memory had been cleared away in advance by death. The last chapter said that after Newton's synthesis humanity for the first time possessed a machine able to advance determinacy on credit; the patch of mist Palitzch saw was the first instrument that machine printed.

Why Prediction Weighs More Than Explanation

Palitzsch's discovery raises a question old in philosophy yet adjudicated every day in the routine of science: given that both are cases of theory agreeing with observation, why does the success of a prediction carry more weight than the success of an explanation? In explaining what is already known, theory and observation both lie inside the snapshot: the orbital records have long been on file, and any clever mind can go on adjusting hypotheses to fit them — the old story of the Ptolemaic system fitting the planetary observations with epicycle laid upon epicycle gave the warning long ago: goodness of fit has never amounted to truth. The situation of prediction is entirely different: it must lock the path publicly before the observation occurs, surrendering signature and deadline, and then wait for the world to speak. In the interval there is no room whatsoever for after-the-fact adjustment — a comet does not read papers.

The "prediction machine" should also be set back into its own age. Before 1705, predicting the future of nature was almost wholly the territory of astrology and oracle: the latter lived by infallible rhetoric — every sentence left enough room for later reinterpretation, and therefore said nothing. What Halley did amounted to smashing the oracle's signboard and hanging out the plate of a notary's office: date, orbit, period, all set down in writing as evidence. And after the Principia, prediction passed swiftly from spectacle into process — lunar theory, the flattening of the Earth, the rate of precession, the dates of comets' return, one item after another entered the routine production of the astronomical almanac; by the nineteenth century, navigators fixed their position on shoreless seas by the star tables, and engineers computed tide gates by the formulas of gravitation — determinacy on credit had entered the daily life of ordinary people, and the thickness of the snapshot had become the quality of life itself. The more commonplace the machine grew, the less anyone remembered how incredible it had once been: this is the law of Chapter One (the more complete the convergence, the more it looks as if there had never been convergence) taking effect once more, on the level of collective memory.

Put the difference plainly in RC's reading. Explaining the known is a path walked back and forth inside the snapshot — each "agreement" is absorbed by material already in the snapshot, and the account grows only by a little. Prediction is the public locking of a convergence path before the observation occurs, and the handing of it over to the world for verdict: if the path is right, the coming frame will develop in the locked way; if it is wrong, the whole world will see at which step it went astray. This is the most stringent cross-subject test of the "stability of a path" — stringent in its irreversibility: the prediction has been notarized, the observation is arriving, and neither party can change its testimony. RC's General Outline says in Section 4.1, discussing prospective focusing, that "prospective cognitive activity essentially constitutes a process of pre-construction of potential reality", unfolding as a feedback loop of "expectation, reality, re-expectation"; and the General Outline says in Section 2.4, discussing ideas and consensus, that "when a particular interpretive framework demonstrates predictive power and practical value, individuals and groups will form a consensus of ideas through comparison". The scientific institution of prediction is the engineering of these two sentences: making the pre-construction public, and making the deviation of expectation an auditable signal.

Halley's own caution deserves a further entry. He did not write 1758 as an ironclad year; he frankly acknowledged that the comet, pulled by Jupiter and Saturn on its journey, must return with a discrepancy, "to be corrected by those who come after". This avowal is no false modesty; it is part of the institution of prediction: a true predictor marks his own uncertainty in public, so that the verdict may have a precise target. The perturbational correction Clairaut later computed stepped precisely onto the interface Halley had reserved. Today's medical trials speak of "preregistration" — hypotheses, methods, and criteria of judgment must be published and archived before the data arrive — and the logic is still the same; only the notary's office has been moved from the Royal Society to a database. In three centuries the instruments have gone through eight generations; the spirit of the institution has not changed: lock first, judge after; the more public the locking, the greater the weight of the verdict.

Here the account joins that of Chapter Two on induction. Hume's verdict still holds: from any finite number of repetitions, logic cannot derive an infinite necessity — were Halley's comet to return on schedule ten thousand times, that would still not constitute a logical proof that the ten-thousand-and-first return is necessary. Prediction does not cancel this crack, nor pretend to; it does another thing: it turns the crack into an institution. Each fulfillment of a prediction is, in Chapter One's words, a "re-emphasis"; each failure is an explicit entry of falsification on the books. The philosophical meaning of the prediction machine lies not in delivering necessity but in keeping the difference between "so far" and "forever" forever open and forever audited — the scientific ledger is therefore unlike every oracle: the oracle cannot be wrong, and so cannot learn; prediction can be wrong, and so the ledger learns.

The Planet at the Tip of the Pen

In the second half of the eighteenth century the prediction machine ran ever more smoothly, until in 1846 it turned its most astonishing stroke.

In 1781 Herschel discovered Uranus with a telescope — the first planet in human history to be "discovered" rather than "seen": observation in the age of the telescope rewrote the membership list of the solar system for the first time. Trouble followed: over the ensuing decades Uranus's actual position departed ever further from the tables computed under the Newtonian framework. By the 1840s the discrepancy had grown too large to ignore: the planet simply refused to follow the path the theory had computed for it. Two explanations lay on the table: either the Newtonian framework failed at great distances, or something had been left out of the computation. In 1845 and 1846, two men independently chose the latter. Adams at Cambridge computed the result first and delivered the predicted position of the unknown planet to the Astronomer Royal Airy, but no timely arrangement for observation followed (Challis, searching later at Cambridge, in fact recorded it twice without recognizing it); Le Verrier in Paris completed the same calculation, and on September 18, 1846, sent the predicted position to Galle at the Berlin Observatory, with explicit coordinates in the letter: in the stretch of sky within one degree of that point on the ecliptic, you will find a new planet.

On September 23 the letter reached Berlin. The night was clear; Galle set to work with the student d'Arrest: check against the star chart, and find near the predicted position a star that was on no chart. Within the hour they had found it — about fifty-two arcminutes from the position Le Verrier had computed, less than one degree. Neptune, the eighth planet of the solar system, was born on paper first and claimed by the telescope after. Arago said that Le Verrier had seen it at the point of his pen. England and France quarreled over priority for years afterward — Adams's calculation came first, Le Verrier's publication and redemption came through; the historical accounts of today mostly list the two as co-predictors, and leave that night to Berlin. The next day the planet entered the registers; later retrospection found that as early as 1795, in Lalande's observation records, it had appeared twice, transcribed as a star, and the imperceptible displacement between the two entries was never questioned — observation not guided by a question sees without recognizing. This detail deserves an entry of its own: the value of a theory lies not only in answering questions but in making them — without the perturbation equations, those two records would have remained two lines of harmless transcription forever.

In RC's reading, the structure of this scene deserves to be taken apart layer by layer. Uranus running off its course was an observational divergence: the position the framework prescribed and the position measured would not reconcile. But the resolution did not overturn the framework; it expanded the account — one member was missing from the ledger, and once he was entered, the divergence vanished at once; moreover, the manner of the entering was itself a prediction carried to the summit. Note the answering of the last chapter here: controlled observation is humanity putting questions to nature; what happened in 1846 was the reverse miracle — theory pointing the way to humanity: "go look there, and you will see what I have said". Prediction turned from defense into offense: the world must not only answer the questions we ask, but also hand over evidence in places where we have not yet asked. The Newtonian snapshot in that moment reached the very peak of its thickness.

The Outstanding Debt of 43 Arcseconds

But beside the peak, the same machine was also printing another instrument — an IOU.

Mercury, the planet nearest the sun, has a perihelion that slowly travels around its ellipse. Celestial mechanics had long known how to explain such a precession: the gravitational perturbations of the other planets. Le Verrier — he again — after refining the account in 1859, reported that even with all the known perturbations allowed for, the perihelion of Mercury still carried a residual precession of about 38 arcseconds per century (later refined to 43 arcseconds) that could not be booked. The figure is astonishingly small — 43 arcseconds per century, only one-fortieth of the angular diameter of the full moon — but for the precision celestial mechanics of the nineteenth century this was a bad debt hanging on the books, plain as day and not to be explained away.

There was only one respectable-looking way to pay, and it had just won a great victory at Neptune: find one more invisible member. Le Verrier himself played the card again, proposing an undiscovered planet inside the orbit of Mercury, which even had a name already — Vulcan. For a time the proposal looked set to re-enact the glory of 1846: in 1859 a country doctor, Lescarbault, reported having seen a dark spot on the face of the sun; Le Verrier personally computed its orbit and labored on its behalf for over a decade; but that "planet" was never reliably seen again, later transit observations and photographic searches found nothing, and Vulcan never arrived on the account. The other repairs were tried one by one: a dust belt around Mercury, a correction to the sun's mass, a small alteration of the law of gravitation; Newcomb in the 1890s pushed observation and theory to the limit of the precision of the day, and the 43 arcseconds stood unmoved. From 1859 to 1915, the debt hung for fifty-six years.

RC's reading further requires the two maneuvers by which a framework absorbs divergence to be booked separately. Expanding the account — charging the divergence to a member not yet on the register — is hard currency: it makes a new prediction (go look in such a place, and you will see such a thing), which can be redeemed or can fail; Neptune is the redeemed specimen. Adjusting the parameters — adding correction terms to the laws, adding decimal places — is paper money: each issue dilutes by one degree the persuasive power of the framework as a whole, because it explains the old account while staking nothing new. For half a century the repairs on Mercury's account were all paper: dust belt, mass correction, fine-tuning of the law — each could swallow a few arcseconds, but no one dared sign another large certificate in the manner of Neptune — because to find one more planet is to have to say when and where it can be seen, and Vulcan never once appeared. How much life a framework has left is seen not in the positive assets on its books but in whether it still dares to open new bets. The position of the Newtonian snapshot at the end of the nineteenth century could not be described more exactly than this: the family fortune still frighteningly thick, and not one new bet left that could be opened.

Set Neptune and Mercury side by side, and the most important symmetry of this chapter comes into view. One and the same strategy — absorbing an observational divergence by means of an unobserved member — succeeded in a night in 1846, and failed for fifty-six years on Mercury. The difference between the two cases lies not in the cleverness of the astronomers but in the fact that the world's verdict does not accommodate the seniority of a framework: behind Uranus's divergence there really was a planet; behind Mercury's divergence there was none. The ledger does not know this; the ledger knows only that the accounts will not reconcile; and whether a divergence is finally absorbed by the old framework (Neptune) or accumulates into the old framework's death warrant (the next chapter) is decided not by the bookkeeper but by the world. This is exactly the expansion of Chapter Three's sentence: science's interlocutor is too good at talking back — and its way of talking back is to let one and the same move succeed at one time and fail at another, with no one able to tell in advance which it will be.

The Cognitive Relay Station

The debt hung, and the machine ran on — the ephemerides, nautical astronomy, and theory of gravitation of the late nineteenth century ran with their usual precision while the 43 arcseconds hung unresolved. This apparently contradictory fact pushes this chapter's core transcription to the front of the stage: what is a theory, that it can go on serving in office with a known defect?

RC's General Outline gives the answer in Section 2.3, discussing theoretical dimensional reduction: every theory is a system of interpretation under a finite horizon, a dimension-reducing projection of the real world, whose validity is constrained by the triple boundary of the level of observation, the imprint of the subject, and the decay of time; but "the dimensional reduction of a theory does not constitute a negation of its value; on the contrary, it establishes its standing as a cognitive relay station — a theory achieves the continuous renewal of cognition through being continuously re-observed". The cognitive relay station: a theory's value lies not in being forever true, but in converging vast numbers of observations into workable determinacy at the least cost. The Newtonian framework is precisely the most brilliant instance of this concept: a few laws, one volume of the Principia, advanced the dates of comets' return, the positions of planets, the times of tides, all on credit; as for the debt hanging on Mercury — a relay station's accounts are depreciated from the first by the triple boundary; the IOU is kept on open record while the positive assets remain in ordinary use, until one day a transaction forces the whole ledger into liquidation.

The four words "the prediction machine" can now have their account written. The machine's fuel is controlled observation (the last chapter); the machine's product is determinacy advanced on credit; and the machine's ethic is a single line: the accounts are public. A successful prediction is a publicly locked path receiving the world's stamp — Neptune and Halley's Comet are stamped stubs; a failed prediction is not the theory's instant scrapping, but the formal entry of observational divergence into the snapshot — the dynamics RC's General Outline gives in Section 1.4, discussing the unity of subject and object: observational consensus, questioned or negated by different modes of observation, produces observational divergence; the subjects of the divergence re-intervene in the Ground of Possibility to achieve re-convergence, completing an innovation of observational consensus, which finally brings an innovation of the rule framework. The relay station has one further quality, often overlooked: when it changes hands it does not burn the instruments. The changing of stations that Chapter Three foreshadowed and that the next chapter will play out in full — relativity taking over the Newtonian framework — renewed every stub the old station had issued (ephemerides, ballistics, tide tables) at the exchange rate of "limiting cases": what changed was the station, not the account. A theory's validity has never been a choice between forever-true and written-off; it is the continuous depreciation and renewal of a relay station within the triple boundary. At the close of Part Three it was said that the fourth entry of Wigner's question is "the suturing of re-convergence"; this chapter completes it on the side of science: mathematics supplies the language, prediction supplies the notarization, observation supplies the verdict — only the three lines together make a complete machine.

The other face of the machine must also be honestly recorded. The successes of prediction are magnified by narrative, like the survivorship ledger of the introduction: Halley and Le Verrier have been celebrated for two hundred years, while the celestial computations made by the same method in those same years that came to nothing have entered no textbook. Prediction weighs more than explanation — but that judgment holds only for predictions that are "public in advance"; a fit anointed as "prediction" after the fact is still only an explanation. The machine's credit is staked, all of it, on the notarization.

In the next chapter the machine will meet the heaviest IOU of its life: not a corner debt of 43 arcseconds per century, but an interferometer set turning in Cleveland — an instrument that in 1887 reported not a deviation but a zero. The 43 arcseconds were at least a debt one could bargain over; the null result removed even the party to bargain with. The world's way of talking back this time was not to answer differently, but to cancel outright a question everyone had supposed existed.