Day 3 — the G317 west

The county sits in the middle of the prefecture, 3,250 metres up on the Xianshui river, and the two trunks meet in the town itself: the G317 coming west from Daofu and the G350 coming north from Kangding. Luhuo calls itself the centre of northern Kham, which is a chamber-of-commerce line with a true premise — five roads, four valleys, one town.

Where the name comes from. The place was 霍尔章谷, Hor Zhangu, the valley of the Hor. In 1897, when the Qing set up a garrison settlement here, they took one character from each half: 炉 from 打箭炉 (Kangding, the old name means “forge”), 霍 from 霍尔. So the county is named after the two places it connects, which is unusual and rather honest. 章谷 itself is usually read as “above the rock” or “head of the rock”; 霍尔 is the old Tibetan word for the northern nomads, and in Kham it means the Mongol-descended houses that ruled here from the seventeenth century.

The earthquake. On 6 February 1973, at 18:37, a magnitude 7.9 event broke directly under the county. The town was flattened; the official toll is around 2,200 dead across five counties, with some sixty thousand rooms down. The surface rupture ran ninety kilometres. This is one of the largest continental earthquakes of the twentieth century, and it is the reason almost nothing in Luhuo is old.

What that means for the monastery. The old site was destroyed. The community rebuilt two kilometres north of town, on a slope between two ridges, and the present complex dates from 1983–88. Keep that date in mind: the shape you are looking at is forty years old, and it is the fourth or fifth building on this site.

Zhangu Gompa. The Tibetan name is brag ’go dgon, the monastery of Zhangu — the same 章谷 that was the valley’s name before the county was invented. In Tibetan histories and on the monastery’s own documents this is the name. The long form, Ganden Rabten Namgyal Ling, translates as something like “the fully liberated, firm, victorious place”, which is the kind of name a monastery receives rather than chooses.

Shouling. The Chinese name, 寿灵, is not a translation of that. It renders a Tibetan title — Tsedé Ling, “the place of long life” — which the monastery says was granted with a seal and a trilingual plaque in the seventeenth century. Two syllables, and the meaning shifts from “victory” to “longevity”.

Drango. The Wylie spelling brag ’go is pronounced something like Drago, and appears in Western scholarship and in old maps as Drango, with the county sometimes given as Drango too. If you see “Drango Gompa” on a map, it is this building.

Not Seling. English-language websites — including some that are otherwise careful — give the Tibetan name as “Seling Gonpa”. No Tibetan source uses that. It is 寿灵 read backwards into Tibetan, and it collides with Songzanlin (Ganden Sumtseling) in Yunnan, which is a different monastery in a different province. If you say “Seling” to a monk here, he will hear the Chinese name badly pronounced.

Why the names diverge. Because the monastery spent the seventeenth to twentieth centuries inside a Chinese administrative system while being a Tibetan religious institution. The county called it by its Chinese name in tax registers, the abbots used the Tibetan name in religious documents, and travellers used whatever the last person told them. Three registers, three names, one building.

The same thing happens to the county. 炉霍 is a Chinese coinage of 1897; 霍尔章谷 is the older name; and the Tibetan brag ’go underlies both. There are parts of the prefecture where a single valley has four names in four languages, and this is one of them.

c. 920 — the site. The monastery’s history traces its origin to Lhalung Palgyi Dorje, the monk who killed the last Buddhist king of Tibet and then fled east. He is said to have practised in this valley and founded a small foundation here. That is a foundation story of the sort every old monastery has, and it cannot be checked; what it tells you is that the site is older than the Gelug order by six hundred years.

1633 — the conversion. The hermitage was Nyingma until Ngawang Puntsok, a disciple of the Fifth Dalai Lama, took it over and converted it to Gelug. This is the date that matters religiously: everything about the place as you see it — the yellow hats, the assembly hall, the monastic university — begins here.

1650 — the monastery. Under the patronage of the Zhangu tusi, the converted house was built up into a proper monastery, and the county dates the institution from this. It is also the date you will find on every Chinese-language page, including the ones that then claim a seventh-century foundation in the same paragraph.

1654–57 — the group. When the Fifth Dalai Lama sent Ngawang Puntsok back to Kham to found thirteen Gelug monasteries, this one was already running, and it became one of the thirteen. That is not a fourth foundation; it is the moment the monastery joined a network.

And the date that isn’t there. The building. The present complex is from 1983–88, after the 1973 earthquake. Every older description you read is describing a building that no longer exists.

The man. Ngawang Puntsok was a Kham-born monk who had studied in Lhasa and become a close disciple of the Fifth Dalai Lama. In 1654 he was sent east, and over the following three years he founded or converted the houses that came to be called the Hor thirteen — 霍尔十三寺 — of which this is one. The list is not fixed: sources differ on two or three entries, and scholars point out that the number is a label rather than an organisation. But the core is real, and it runs along the road you are driving.

Why thirteen. Two stories, both told in Kham. In the first, Gushri Khan — the Mongol ruler who had just conquered this country for the Gelug cause — repented of the killing and the burned monasteries, and paid for thirteen foundations as expiation. In the second, Ngawang Puntsok was simply copying his teacher, who had built thirteen monasteries in central Tibet with the silver the Dalai Lama brought back from Peking. Both are stories about legitimacy, and the second is the one the monasteries prefer.

The politics. Either way, the thirteen were a settlement: they replaced or absorbed local houses, they were tied to the new Gelug state, and they sat on the trade road to Tibet. The monastery here was one of the hinges — Zhangu was where the road crossed the river.

The tusi’s part. The seventh Zhangu tusi, Jigme Tseden Norbu, gave the monastery the sites of two Nyingma houses next to his own residence, then rode to Lhasa to ask the Dalai Lama for three things: a silver image of the Buddha cast by Tibetan smiths, a resident abbot, and permission to hold a Monlam festival on the Lhasa pattern. He was granted all three, and the right to put a gilded roof on the Buddha hall — which is why one roof here is gold and the rest are not.

The gifts, and what happened to them. The Panchen Lama gave the monastery fifteen thangkas illustrating the Buddha’s life. One was lost in the 1960s; the other fourteen were hidden by a monk inside his mother’s coffin and handed back when the monastery reopened, and a fifteenth was painted to replace the missing one. That story is told by the monastery, and it is the sort of detail that survives because somebody took a risk.

The arrangement. You come up from the road, along the covered corridor of prayer wheels, and the buildings are stacked above you in three levels. The lowest tier is service and housing; the middle holds the great assembly hall; the top is the Buddha hall and the chapels. This is the standard Kham solution to a steep site, and it is the reason the monastery photographs as a wall of white rather than as a cluster.

The hall. The assembly hall holds two thousand people at Monlam, on eighty-eight pillars — the pillar count is the traditional way of describing a Tibetan hall, and it is more precise than floor area. The Buddha hall above it contains a silver Shakyamuni about three metres high, said to have been cast in Lhasa at the Fifth Dalai Lama’s instruction; the gilded roof over that hall is the one the tusi was allowed to add.

The paintings. The monastery is associated with Namkha Gyen, the thangka painter born in a village in this county, whose “six ornaments and two great disciples” and Pure Land compositions are in the hall. The county has built an entire identity on him — it calls itself the home of Tibetan thangka — and the claim is at least partly justified: his line is a recognised school, and it is taught here.

The numbers. 14,000 square metres of building, 277 rooms and chapels, 232 monks in residence in 2024, four colleges and, at its height, seventeen branch monasteries. The monastery is a working institution with a management committee, a legal registration dating from 2023, and a primary school curriculum for its novices — not a monument.

1896. When the Qing moved to take direct control of Kham, this monastery put several hundred armed monks into the field on the Tibetan side. The campaign failed; the county was turned into a garrison settlement the following year.

1905. The county’s headmen resisted the new tax register; the official sent to enforce it was killed. The reprisals fell on the villages, not on the monastery, which was deliberately protected. This is a different event from the Batang rising of the same year, with which it is constantly confused.

1935. The monastery’s militia surrounded the county government and drove out the magistrate; a Sichuan army unit then occupied the town, executed the ringleaders and took four monks hostage until a fine was paid.

1956. During the uprising across Kham, the monastery was besieged and taken by the People’s Liberation Army in late March. This is the event the older monks will not discuss and the guidebooks do not mention.

1960s–70s. The monastery was closed and its contents dispersed; the buildings that survived were finished off by the 1973 earthquake. Of the fourteen surviving Panchen thangkas, all were saved by being hidden.

1982 onwards. Reopened, rebuilt, and reopened again to visitors. In 2023 it completed the registration that makes it a legal religious body rather than an informal one. There are 232 monks. The corridor is swept every morning.

The lake sits in Chonggu township, on the north side of the G317, fifty-seven kilometres out of Luhuo on the road to Garze. The villages are above it at about 3,700 metres; the water is at 3,510. That two-hundred-metre difference is the whole reason the place photographs well — you look down into it from the road.

How big. Two figures are in circulation and they are not the same measurement. The provincial government and the 2026 scenic-area filing give 1.92 square kilometres of water; older gazetteers and local reports give 1.09. The reserve around it is a different number again, and has been revised three times: 317 square kilometres at listing in 1999, 331.8 after the 2020 boundary optimisation, 363.96 in the 2024–2035 master plan, of which 4,501 hectares lie in Garze county. Do not add these up; they describe different polygons.

Why there is water here at all. The basin is a pull-apart on the fault, and the fault is still moving — left-laterally, the far side sliding north-west. The depression filled with meltwater and rain, and later a rockfall or a moraine helped close the low end. So it is not a glacial lake and not a dammed lake: it is a structural hole that happens to drain badly. Its outflow is at the north-west corner, into the Daqu, and from there into the Yalong.

What it is legally. A county reserve in 1984, prefecture-level in 1995, provincial in 1999. In November 2023 the national forestry administration listed it as important habitat for the black-necked crane — one of the first batch of such listings. In October 2026 the surrounding Two Rivers scenic area was graded 4A, with 6.57 square kilometres of developed ground, a visitor centre, an eco car park and a boardwalk. None of that changes what the water is; all of it changes who is standing on the shore.

The route. Black-necked cranes breed on the high wetlands north and west of here — Zoigê, and the Shaluli mountains in Sichuan — and winter in the Yunnan–Tibet border country, above the great river gorges. Kasa sits on the way. The birds that use it are passage birds: they arrive, feed for days or weeks, and leave. Nobody has found a nest here.

That distinction matters more than it sounds. A 2014 survey of the species’ Sichuan range lists Luhuo as a transit county, not a breeding one, and the reserve’s own description calls the lake a stopover and a habitat. The tourist boards that call it a breeding ground are wrong, and the reason to say so is that it changes what you should expect in June: not chicks, but nothing at all.

What the counts are. There is no published waterbird census for this lake. The one quantitative bird survey on record is from 2004 and lists forty species in nine orders — with three national first-class and two second-class among them. The 50 black storks you may have read about were counted in the Xianshuihe wetland park, not here; the “tens of thousands of birds” figure is a county-wide line from a newspaper. Neither is a count from this water.

Mammals. A 2006 survey of the reserve, published in 2013, lists 54 species in seven orders, twenty of them protected, five of them first-class: white-lipped deer, forest musk deer, alpine musk deer, snow leopard, leopard. None of these is a lakeshore animal. They are in the hills inside the reserve, and you will not see them from the road.

Why the water suits them. Three reasons, all unglamorous. It is shallow at the west end, so a crane can wade; it has mud and aquatic plants, so there is something to eat; and it freezes late and thaws early relative to the bogs around it, so it is open when the neighbours are not.

The habit that threatens it. Feeding the birds and releasing fish into the lake look like kindness and are neither. Released carp and catfish eat the spawn and the weed, cloud the water, and pull the whole shallow ecosystem towards a pond. Local fisheries officers removed 31.8 tonnes of alien fish from the lake and its channels in 2023; in 2023–24 they moved on more than fifty people feeding them and confiscated 22 bags of fish food. A 2025 prefecture rule now requires any release of aquatic animals to be registered and traceable. This is not a suggestion on a sign.

The release. For decades people came to the shore and let fish go — carp, grass carp, catfish, bought at the market in town. The merit is obvious; the ecology is not. These are warm-water fish that eat spawn, root up the weed that shelters invertebrates, and turn clear shallow water into green soup. The lake has no native carp. It had a plateau fish fauna: small, slow, cold-water species, of the sort found across the eastern plateau.

The result. A local fisheries officer told a reporter in 2018 that the native fish were “basically gone”, and estimated the water at half its former extent; both are statements rather than surveys, and should be read as such. But the direction is not in doubt. A 2022 wetland survey of the reserve recorded exactly one fish species — and it was an introduced carp.

The correction. In 2023 the county netted 31.8 tonnes of alien fish out of the lake and its channels. In 2023–24 a joint enforcement campaign moved on more than fifty people who were feeding the fish and seized twenty-two bags of feed. In 2025 the prefecture published rules requiring any release of aquatic animals to be declared in advance and traceable to a supplier.

Why it is worth knowing. Because the same scene is playing out at half the lakes on this plateau, and because a guest who feeds the ducks at Kasa is participating in it, however small the bread. The rule at the shore is not etiquette. It is the rule.

And the water itself. The lake has been dropping. Local reporting and provincial filings both note a fall in level and storage, with the usual candidates — a warming plateau, more irrigation upstream, less snow. There is a small hydro station at the north end, built in 1984, 250 kilowatts. Whether it matters to the level is a question nobody has published an answer to.

The walk. Circumambulation is the ordinary practice — clockwise, on foot, on the road and the tracks round the shore. Households come at the new year and after a birth; the lake is asked for the things lakes are asked for. There is no fixed festival date that anyone official will confirm, so none is given here.

The legend. Told as a legend, because it is one: a valley with no water, an old woman praying for a hundred and eight days, a dream in which a figure in white tells her to release the frog in the ravine. She finds a boulder shaped like a frog, rolls it aside, and the spring comes up and fills the basin. The version in the county’s own booklet has an old man; the version on the information board has an old woman. Both end with barley.

The monastery. Kasa monastery sits on the hillside east of the water, a small white-and-red cluster above the village. It is one of the many foundations in this valley that history has moved around: the better-known Gelug monastery of the district, Jori, was destroyed in the 1973 earthquake and rebuilt at Zhurwo, and its old site is on the slopes above this shore. Neither is a museum; both are working.

The cemetery. In 1984, cutting the channel for the hydro station, workers hit a stone-cist grave field. Two hundred and seventy-five graves were excavated that year; the county now gives a cumulative figure of 335. The grave goods — bronze, bone, stone, cowries, coral, a little gold — date the site to the Spring and Autumn period, roughly the fifth century BC, which makes the people buried here contemporaries of Confucius and of the first cities of the Central Plains. The report was published in Kaogu in 1991.

Why it matters. Because it means the shore has been used for three thousand years, by people who had no cranes to protect and no lake to manage — and it is still being used, by cranes and by you. The cemetery is not open; the graves are under the fields and the road.

What is in the count. One hundred and twenty-five mammals in 26 families; 436 birds in 65 families; 31 reptiles; 30 amphibians; 30 fish. Birds are two thirds of the total, which is not a quirk of counting — it is what a mountain province on a continental flyway looks like. A place that is cold, high and full of wetlands gets birds the way a warm coast gets molluscs.

What is protected. 126 of those species are on the national protected list: 34 first class and 92 second class. Those 34 are 14.5 per cent of every first-class protected animal in the country, and they include the giant panda, the golden snub-nosed monkey, the snow leopard and the white-lipped deer. The protected-area network covers 55,400 square kilometres — more ground than the Netherlands, inside one prefecture — and the ecological red line is larger still, at 69,600.

Why here, and not next door. Three things happen at once in western Sichuan. The Oriental and Palearctic biogeographic realms meet along this line, so southern animals and northern animals overlap instead of replacing each other. The great rivers run north to south, which means that during the ice ages the valleys were refuges you could retreat down and during the warm periods they were corridors you could move back up. And the relief is extreme: from about 1,000 metres where the rivers leave the plateau to 7,556 at the summit of Gongga, in one prefecture.

How a number like this is made. Not by one survey. Camera traps on fixed grids, walked transects for large mammals, point counts for birds, and a long accumulation of county records going back decades. The 2020s have added a systematic layer: the prefecture now runs 18 county-level biodiversity inventories, a wolf survey that has logged more than 1,600 animal encounters, a primate survey that counted 1,112 animals in 17 troops, and snow leopard work in Gongga, Shiqu and Xinlong.

Where the 34 come from. Not evenly. Most of them are clustered in the two western counties — Shiqu and Baiyü — and along the Yalong and Jinsha gorges. The county you are driving through today, Ganzi, has snow leopard, brown bear, musk deer and blue sheep, but it is not where the exceptional list is. That is a fact about the map, not about the county.

The five you will hear named. Gongga Shan above Kangding, the big one, protecting the vertical stack from subtropical valley to glacier. Yading in Daocheng, three sacred peaks and their forests. Haizi Shan on the Litang–Daocheng line, a high wet plateau of boulders and small lakes. Changsha Gongma in Shiqu, 6,698 square kilometres of high marsh on the Yellow River divide. And Qingqingsongduo in Baiyü, 1,580 square kilometres, the only national reserve in China named after the white-lipped deer.

What a reserve actually is. Not a park with a fence. A protected area has a core zone where nobody goes, a buffer zone where limited activity is allowed, and an experimental zone where the existing villages, monasteries and grazing continue. The villages were there first; the reserve was drawn around them. That is why a herder’s tent can be inside a national reserve, and why that is not a scandal.

The four monitoring stations. The prefecture runs permanent biodiversity observation points at Zimei Pass above Kangding, at Yading, at Zhenda in Shiqu, and at Luhuo — four places, four habitats, the same protocol. Camera traps and bird counts at those four sites are how most of the numbers on these pages were produced.

What this means for you. The road does not run through core zones, and the animals you see from the car are mostly on the edges of the protected ground, where grazing, road and cover meet. That edge is not a lesser place. It is where the sightings are.

How to tell the gazelle from the others. There are three small hoofed animals on this ground and they are not hard to separate once you know what to look at. The gazelle is the size of a dog and runs with its tail down. The blue sheep is the size of a sheep, grey, and goes uphill when disturbed. The white-lipped deer is the size of a red deer and goes into cover. If it stands in the open and stares at the car, it is a gazelle.

The marmot is a weather instrument. It is up and about from roughly April to September, and for the rest of the year it is underground with its heart at a few beats a minute. In late October you will see very few, and the burrows will already look abandoned. The ones you do see are usually young, or late.

The pika argument, which you will hear in the car. Somebody will tell you the pika is destroying the grassland and that the county pays a bounty. That story is backwards. Pikas concentrate where the sward is already short and open — they are an indicator of grazed or degraded ground, not the cause of it. Poisoning them removes the base of the food chain: the fox, the buzzard, the bear and the weasel all eat pika, and the snowfinches nest in their burrows.

And yet. There is a real conflict, and it is not imaginary: on small, overgrazed pastures pikas do reach densities that make recovery harder. The honest version is that they are a symptom that can become a participant. It is a more interesting sentence than the one the bounty poster gives you.

Why the treeline. A line where forest stops is not a single line but a band, two to four hundred metres wide, of scrub, isolated clumps and open ground. Two kinds of habitat meet there and neither one dominates, so species from both sides overlap and a few live only in the seam. In practice this means the treeline holds more kinds of bird per hectare than the dark forest below it, and it is also the only band you can walk through without a machete.

Where to look for the pheasants. On the ground, at the edge of the trees, in the first hour of light. They walk in single file along paths they have made themselves, and those paths look exactly like the paths of a small dog. If you find a dust bowl under a shrub with feathers in it, you have found a lek and you should back off quietly.

The blue sheep is the reason the snow leopard is here. Everything above the treeline on this slope is organised around one relationship: blue sheep eat the grass, and the snow leopard eats the blue sheep. Where the sheep are, the cat is; where both are absent, the habitat is empty no matter how good it looks. Rangers read a slope by counting sheep, not by looking for cats.

Both photographs. The two pheasants on this page were photographed in Garzê prefecture — the only pictures in this booklet that were taken where you are standing. The blue sheep was not, and the caption in the back says so. It matters: a picture taken elsewhere shows you the animal, not the place.

The lowest gap in the ridge. A pass is the place where a mountain range lets a road through, which is why every old trade route over these ranges goes through one rather than over the top. Zheduo, on the southern road out of Kangding, is the first pass on that route to clear four thousand metres: 4,298 m at the top against about 2,560 m in the town below — some 1,800 m of climbing in the course of one afternoon of bends.

What that buys you. The bands on page 03 are not an abstraction; they are what you drive past. In the hour it takes to reach the top you cross shrub, conifer forest, the last trees, and open meadow, in that order, and you can watch each one end. The treeline stops being something printed on a diagram and becomes a height on the slope in front of you, with grass above it.

One more, further up. Kakarala, west of Litang at 4,718 m, has almost no woody cover at all — grass, gravel, horizon. Above that height on this side of the prefecture the trees simply stop, and what replaces them is a mat of grass you could cover with your hand. It is not a gradual thinning. You reach a certain height and above it the wood is simply gone.

And one that was taken away. Until September 2017 the road to Dege crossed the Chola range on a pass road that took about two hours and lay under snow for much of the year. The tunnel that replaced it runs seven kilometres and takes five to ten minutes, with its portals at 4,378 m. The old road up is still open in summer and almost empty. Whatever lives along it has had a quieter decade than it was used to.

The descent. The world has gone wrong, and the gods decide to send someone down to fix it. He is born in Ling to a herdsman’s household and grows up as Joru — filthy, obstreperous, disliked by the neighbours, useless at everything a boy is supposed to be good at. This is the epic’s first joke at the expense of the powerful, and it never really stops making it.

The race. The throne of Ling is empty. The chiefs agree that the crown, and Dromo, the most beautiful woman in the country, will go to the winner of a horse race. All the princes ride their best horses. Joru arrives on a scrawny, half-starved animal that everyone laughs at, and wins. He is enthroned as Gesar of Ling, and the scrubby mount turns out to be a divine horse.

The court. Two queens — Dromo Bum, and Meza Bumkyi — thirty heroes, and an uncle called Trotung who is the epic’s traitor. Trotung is greedy, treacherous and endlessly resourceful, and he is the reason there are stories at all: without a man inside the camp who can be bought, the campaigns would be over in a chapter. He is also, in the way of these poems, more entertaining than most of the heroes.

The wars. Each campaign is against a castle and the king in it: Hor in the north, Jang in the east, Mon in the south, and the demon lands beyond. The most famous is the Hor war, in which Gesar is away, the Hor king takes Dromo, and the hero comes back to burn a kingdom and fetch his wife. It is the episode every singer knows and the one most often sung on a winter night.

The end. He does not die. In most versions he goes — upward or downward, depending on who is singing — and leaves behind a kingdom, a prophecy, and a story with no last line.

Two registers. Every episode alternates: prose tells you what happened, verse is what the characters sing. The prose carries the plot and moves fast; the verse stops and lets a character be angry, or grieve, or boast for two hundred lines. UNESCO’s nomination describes the form as “beads on a string”, which is the best short description anybody has written.

The castles. The unit of the epic is the dzong — a fortress, and by extension the campaign against it. The classic count is eighteen great castles and eighteen small ones; some traditions list more, some fewer. Each is a complete story: a king who has done something wrong, a campaign, a siege, a conversion or a killing, and a return.

How a singer chooses. By the room. A wedding gets the birth and the race; a long winter night gets the Hor war; a family that has asked for a blessing gets an episode where Gesar prays. The audience is not passive — it knows the frame and waits for the passages it came for.

Nobody agrees on the order. Printed editions arrange the castles differently, and a singer will not thank you for asking which arrangement is correct. The Hor war is nearly always placed late, because it is the emotional centre; the birth and the horse race are always first, because a new audience needs them.

Why it can be that long. Because nothing forces the string to end. A new castle can be added at either end without disturbing the rest, and singers have been adding them for centuries. This is also why no two manuscripts agree, and why the printed editions are transcripts of particular singers rather than of “the epic”.

What a performance is. Not a recitation. The singer works from a remembered spine and fills it in, and a good one will hold an audience for a whole night on a single episode. The text you can buy is the fossil of one of those nights.

Where the metre lives. In the verse. Tibetan epic verse runs on syllable count and alliteration rather than rhyme, and a singer marks the line with a rocking cadence that is unmistakable once you have heard ten minutes of it. You do not need the words to hear where the lines break.

The counts do not match. Chinese state sources give “more than 120 volumes and more than a million lines of verse”. A 2002 report from the same national project gives roughly a hundred volumes at about five thousand lines each — half a million. A third frequently quoted figure is thirty-six volumes and fifteen million words. All three are in circulation, often on the same website.

Why they diverge. First, the unit: a “volume” in a Tibetan pecha edition is not a fixed size. Second, the corpus: do you count the Mongolian Geser, the Buryat Abai Geser, and the Tu and Yugur versions, which are separate poems on the same hero? Third, prose: the metre only counts verse, and a third of the epic is prose. Fourth, and most honestly, nobody has finished counting.

Why it still matters. The number is not vanity. It is the argument for money: a corpus that needs a hundred singers, forty years of recording and a publishing programme that has not finished is expensive, and “a million lines” is what makes the case in a budget meeting. That is true of every heritage listing anywhere.

The trade. A Gesar singer is a specialist, not a hobbyist. He or she will hold an audience for a night, move between prose narration and sung verse without a text in front of them, and do it in a house, a courtyard or a tent. Chinese sources count roughly a hundred working singers among Tibetans, Mongols and Tu, and treat each one as a national asset with a name on a list.

The famous three. Zaba, from Qamdo, who was reckoned to know more than thirty volumes and dictated them for recording in the 1980s; Sangzhu, whose dictated version is still being published volume by volume; and Yumei, one of the few women in the trade. Between them they account for a large share of everything that has been written down.

The economics. A good singer is booked for weddings, births and New Year, and the fee is real. This is one of the reasons the tradition has not died: it was never only an art, it was a service people paid for at the moments that matter in a household.

How they get it. This is the part that resists explanation. Singers describe being given the story, whole, in a dream or in a single moment; or finding it in the heart; or reading it off a bronze mirror held up for them. Scholars record the accounts without adopting them. What can be said is that the transmission is not ordinary memorisation — the volumes are too many, and the singers are usually not literate.

Why the mirror. The bronze mirror is not a prop for the audience. It is where the singer looks, and what he sees is described as a page or a scene rather than a reflection. Several of the recorded accounts are specific enough about this to be unsettling.

Why it matters to the road. When a version is written down, it stops moving. When the last singer in a district stops, a branch of the poem ends. The UNESCO listing in 2009 was not about preserving a book; it was about keeping a performance alive.

Turning nights into books. The recording began in the 1950s, was reorganised in the 1980s under a national leading group, and has never stopped. The method is simple and brutal: find a singer, sit him down, record everything, transcribe it, publish it as a numbered volume under his name. It is the only honest way to do it, and it produces a library rather than a book.

What that library looks like. Thousands of hours of tape; dozens of printed volumes in Tibetan and Mongolian; a handful of Chinese translations; and one dictated version — Sangzhu’s — planned at forty-five volumes, of which a fraction has appeared. Nobody involved expects to finish.

Where scholars put Ling. The consensus, such as it is, places the kingdom in the upper Yangtze country — the belt running through Dengke, Sershul, Golok and on towards Chamdo. That is where the place names in the epic cluster, and it is where the oldest local Gesar cults are. Nobody has ever located the capital, and it is unlikely that anybody will.

Where Ganzi is. This county belongs to a different name layer altogether: Hor. The district was the country of the five Hor houses — Khangsar, Machu, Beri, Donggu and Druwo — and the county itself was carved out of the first two. Ganzi is dkar mdzes, “white and beautiful”, after the white cliffs above the town. There is no Ling in the name.

The awkward part. In the epic, Hor is the enemy — the northern kingdom whose king carries off Dromo and whose capital Gesar burns. Local tradition in Ganzi nonetheless holds that the Hor–Ling war was fought here, and the city you are standing in is built on what it calls the battlefield. That is a local claim, sincerely held and impossible to verify. It is also, on the epic’s own map, the wrong side of the war.

Why that is not a scandal. Oral epic does not work like a deed. Stories attach to the places that will keep them, and a country that sings Gesar for four hundred years eventually becomes part of the poem. The claim is not a forgery; it is what happens when a story outlives the map it came with.