The Guanyin oracle is a set of one hundred numbered poems. Temples across Hong Kong, Taiwan, Singapore and Penang print them, and a visitor who draws number 47 in one place and number 47 in another will reasonably assume they have drawn the same thing.
We took two complete editions — one published by Hong Kong's statutory temple body, one from a Taiwanese temple catalogue — and compared all hundred poems character by character.
Not one of the hundred is identical
After stripping punctuation to compare the verse text directly, the differences are clear.
| Difference between the two editions | Lots |
|---|---|
| Identical | 0 |
| 1 to 3 characters differ | 7 |
| 4 to 8 characters differ | 24 |
| 9 or more differ | 69 |
Lot 1 is among the closest, differing in two characters. Lot 100 is, for practical purposes, a different poem. Lot 9 shares a single line verbatim and diverges around it.
The form does not drift: both editions are four lines of seven characters throughout. It is the wording inside that form that moves.
What does hold still
Each lot carries an allusion — a named figure or episode from history or legend attached to that number. Those match across the two editions on 95 of the 100.
The poems are fluid, but the index of stories attached to them is not.
The Hong Kong body states the position itself. It compared several current versions and revised the ranks, the allusion figures, the verses and the interpretations — describing the goal as suiting the times and supplying what individual versions lacked, and expressly not as seeking a single unified text.
The grades disagree, and not randomly
Most editions sort the hundred into three bands — upper, middle, lower. Counting complete listings:
| Edition | Upper | Middle | Lower |
|---|---|---|---|
| Three widely-circulated sites | 22–24 | 58–60 | 18 |
| Hong Kong statutory temple body | 34 | 46 | 20 |
The three popular editions barely differ from one another — two lots apart, and an identical set of eighteen in the lowest band. The Hong Kong edition is the outlier, differing from each of them on sixteen lots, nearly all of them shifted upward out of the middle band. Across three editions, 83 of the 100 agree on which band they belong to.
The Hong Kong body had already said it revised the ranks, so the outlier is the edition that told you it would be one.
Some editions use six bands rather than three, but this is not simply the same judgement at a finer resolution. Collapsing the six bands back into three produces a distribution that still disagrees with the three-band editions on eighteen of the hundred lots.
The draw, and the arithmetic inside it
A temple following the full procedure does this: The enquirer states name, birth date, address and the matter at hand, then casts a pair of crescent-shaped blocks to ask whether a lot may be drawn at all. One flat face and one curved gives assent. Two flat faces is read as the deity smiling, and the question is restated. Two curved faces is a refusal.
Given assent, the cylinder of numbered sticks is shaken until exactly one falls. If more than one leaves the cylinder the draw is void. The number is noted, the stick is returned, and the blocks are cast again to confirm that this is the intended lot — conventionally requiring three consecutive assents. Otherwise the draw is repeated.
The casting blocks introduce a peculiar statistical property. They are not fair. A school science project threw a single block a thousand times and recorded the curved face 529 times, and a second project testing material, size, throwing angle and bounce concluded that every pair has its own probabilities.
But assent requires one of each face, so with a per-block probability q of landing curved, the chance of assent is 2q(1−q). That expression peaks at q = 0.5 and falls away gently on either side. At the measured 0.529 it gives 0.498.
The pairing is self-correcting: a biased block, thrown with its twin, still produces very nearly a one-in-two chance of assent. This puts the probability of three consecutive assents close to one in eight, for an expected eight attempts before a lot is confirmed.
What the confirmation step does not do is reshape which lot you end up with. This is inference rather than a sourced result, so take it as reasoning: provided the blocks are indifferent to which stick fell — and there is no mechanism by which they could be otherwise — rejecting a lot and drawing again changes how many attempts are needed, not the distribution over the hundred. The confirmation inherits whatever the shaking produces.
Whether the shaking itself is uniform, nobody appears to have measured. Searching in three languages turned up no study of stick-draw uniformity at all. Treat any claim that the hundred are equally likely as untested.
The composition is a decision, not a fact about the world
The mix of good and bad lots is a deliberate choice. At least one temple has published its breakdown and the reasoning behind it.
Sensō-ji in Tokyo gives the breakdown of its hundred: 17 great blessing, 35 blessing, 5 half, 4 small, 3 lesser small, 6 late, and 30 misfortune. The temple's own explanation is that many shrines once used a ratio around seven to three and gradually raised the proportion of favourable lots because unfavourable ones discouraged people from drawing. Sensō-ji kept its thirty.
Because these proportions are adjusted deliberately, and because editions disagree on sixteen or seventeen of the hundred grades, the same lot number can be a middling draw in one temple and an upper-tier one across the strait.
How old the hundred actually is
The set appears in the textual record by the thirteenth century. Two Buddhist histories, one with a preface dated 1237, refer to a hundred-lot set of the bodhisattva and to a separate set of one hundred and thirty from another monastery.
The oldest surviving physical edition is older than any other printed Chinese temple oracle. It belongs to the Upper Tianzhu monastery at Hangzhou, a major Guanyin pilgrimage site, and was acquired by a collector in Beijing in the 1930s; it is now held by the National Library of China. Its verses are illustrated, which later editions are not.
How much of it survives is usually given as 85 of an original hundred, numbered from 5 to 92 with two more missing from within that range. Note that 5 to 92 inclusive is 88, and removing two leaves 86 — so the figure of 85 implies one further absence that the scholarship does not itemise, and some sources do say 86. It is safer to write "85 or 86".
The hundred is not a natural size, incidentally. Lot texts in the Daoist canon run to 49, 53, 64, 99, 100, 120 and 360.
Four things this set is not
Four claims circulate widely enough to be worth naming.
It is not descended, textually, from that Song edition. No scholarship claims it is. The comparison that has been published is formal — the same convention of verse plus interpretation, persisting. The prosody argues against descent outright: the Song verses are five characters to the line, and the living Chinese and Taiwanese Guanyin hundred is seven. Shared institutional ancestry at the same monastery is defensible. A shared text is not.
There is no sixty- or sixty-four-lot Guanyin set. The sixties belong to other deities. A sixty-four appears to be contamination from either the Book of Changes or a different, unrelated lot text that does have sixty-four. Sets of 28 and 130 do exist.
The Japanese hundred is a separate recension, not a translation. Its first lot is five characters to the line and has no counterpart in the Chinese set. The two corpora should not be merged. A version transmitted in Okinawa does open with the Chinese seven-character first lot, and is explicitly distinguished from the Japanese one.
The Japanese set did not begin in the Edo period. The popular account credits a seventeenth-century monk with its discovery. Two Muromachi attestations predate that: an inscription of 1395, and a box dated 1409 at a temple in Iwate that still holds 88 of its original bamboo slips, inscribed to the effect that they were copied from a Chinese printed original. The seventeenth-century figure, then, popularised the practice but did not originate it.
Where these numbers came from
The verse comparison and grade counts were computed for this article from scraped listings of the editions named. The Hong Kong body’s statement is from its own site, as are the Tokyo temple’s lot composition and explanation. The block probability is our arithmetic applied to a published experimental figure. Textual history is drawn from peer-reviewed scholarship on the Song edition and a Japanese academic article on its transmission.
Read on 20 September 2026. One caution from this work is worth passing on. During this work an automated summary of one source returned a grade table that was entirely fabricated — plausible numbers, correct format, no basis. Every count above was taken from the underlying pages directly for that reason.
Two things could not be settled. The specific editions used by temples in Singapore and Penang are not documented online, so the comparison here covers Hong Kong and Taiwan only. And no dedicated study collating the variant recensions of this set appears to exist; if one does, it is most likely inside one of two large comparative monographs that could not be read for this article.
