A slime mold redesigns the Delhi Metro.

About 12,500 virtual slime agents leave scent trails, follow each other's trails, and are drawn to 34 food sources placed wherever Delhi NCR's people actually live, using the EU's GHSL 2025 population grid. They're never told where the roads, rivers or stations are. Each visit gives you a fresh growth with its own random seed, so your network won't quite match anyone else's.

Who builds the better network?

Per km of track, the real metro reaches about 11% more people than the slime. Even when we let the slime grow to the same length as the metro (386 km against 368 km), DMRC still covers 2 million more people within walking distance. The planners win on density. The slime only knows about 34 hotspots, so it skims past the dense core and spends its tubes on long links out to the edge towns.

Where both maps agree people are missing

The more interesting result is where the slime runs a thick tube through a big crowd that the metro doesn't reach. We checked the top four against current news. All four line up with real plans: three are being built or have been tendered, and one is only a proposal so far.

Mumbai: the sea is a wall

Mumbai is a thin peninsula with creeks cutting into it, so here the slime isn't allowed to swim: sea cells block it, in the research run and in the live sim above (pick the Mumbai tab). The real network is the suburban rail plus the metro, since in Mumbai the local trains are the backbone. Same rules otherwise: GHSL 2025 population (25.9 million people in the box), 6 independent growths of 50,000 agents, and we count people within 800 m of a line.

The slime builds a lean network: half the length, and more people per km. But that's because it only builds the busiest links. Grow it as long as it will go (287 km; it can't stretch to the full 347) and it still reaches 3.3 million fewer people than the real lines. The trains win. For comparison, the metro on its own (85 km) is the densest network of all: 6.8 million people, about 79,900 per km.

Delhi vs Mumbai: in Delhi the real metro beat the slime per km (33,300 vs 29,900); in Mumbai the slime wins per km (56,600 vs 41,100) but loses on total reach. In both cities, even with the slime grown as long as it will go, the real network reaches 2 to 3.3 million more people within walking distance.

Where the slime goes and the trains don't

Crowds the slime connects that are more than 2 km from any line in our station data, checked against the news:

Mumbai caveats. Station data is a Wikidata snapshot (Mumbai Metro, Navi Mumbai Metro, Mumbai Suburban Railway; 220 stations), and it is missing the six Metro 2B stations that opened in 2026, so the Trombay distance is probably somewhat shorter. Monorail is left out, and lines are drawn as straight links between stations. The Natural Earth 10 m coastline misses small creeks (upper Thane Creek, Mahim, Manori), so the slime can sometimes "cross" there. We count residents only, not jobs or trips.

Bengaluru: the closest fight yet

City number three. Same rules as before: GHSL 2025 population (15.7 million people in a 50 × 44 km box), 23 food hotspots, 6 independent growths of 50,000 agents, and we count the people living within 800 m of a line. The real network is Namma Metro's three open lines (Purple, Green and Yellow: 83 stations, 96 km officially). Wikidata already had the Yellow Line, which opened in August 2025. The elevated Pink Line section (Kalena Agrahara to Tavarekere) passed its safety inspection in October 2026 but hasn't opened to passengers yet, so we left it out. Bengaluru has no suburban rail running yet, and no sea, so nothing blocks the slime here. Pick the Bengaluru tab in the live sim above.

Trim the slime to the metro's length and the metro still wins, but only by about 6%: the closest of our three cities. The two networks barely agree, though. Only a fifth of the slime runs within 1 km of a metro line. The slime piles into the dense north and north-west (RT Nagar, Hebbal, Mathikere, Yelahanka) and loops a ring through the east, while the real metro is a cross with a south-east spur to Electronic City. The slime only counts where people sleep. The metro also has to reach where they work, and Bengaluru's tech parks are exactly the places the slime ignores.

Three cities so far: per km of line, the real network beats the slime in Delhi (33,300 vs 29,900) and Bengaluru (35,000 vs 25,700). Mumbai is the exception: there the slime wins per km (56,600 vs 41,100) but loses on total reach.

Where the slime goes and the metro doesn't (yet)

The biggest crowds more than 2 km from any open line, checked against the news. All four are already being built or planned:

Bengaluru caveats. Station data is a Wikidata snapshot taken on 25 Oct 2026 (only stations marked "in use"), and lines are drawn as straight links between stations, so we measure 85 km against the official 96 km. The Pink Line's elevated section will add 7.5 km on Bannerghatta Road once it opens, but it doesn't touch any of the four gaps above. We count residents only, not jobs or trips. Lakes and terrain aren't modelled.

Kolkata: the metro's biggest win

City number four. Same rules: GHSL 2025 population (18.3 million people in a 41 × 44 km box), 19 food hotspots, 6 independent growths of 50,000 agents, and we count the people living within 800 m of a line. The real network is every Kolkata Metro line open on 25 Oct 2026: Blue (Dakshineswar to Kavi Subhash), Green (Howrah Maidan to Sector V, including India's first tunnel under a river), Purple (Joka to Majerhat), Orange (Kavi Subhash to Beleghata) and Yellow (Noapara to the airport). That's 55 stations and about 73 km officially. The Orange Line's extension to Sector V is still in trial runs, so we left it out. We let the slime cross the Hooghly, because the Green Line already does. We left out the suburban trains to keep Kolkata metro-against-metro like Delhi and Bengaluru, but see the caveats. Pick the Kolkata tab in the live sim above.

This is the slime's heaviest defeat. Trim it to roughly the metro's length (it still gets 11% more track) and it reaches 31% fewer people. Kolkata's 1984 Blue Line runs straight down the densest strip of the city, Dum Dum to Esplanade to Tollygunge, and almost every km of it has a crowd on both sides: 77,900 people per km, the most of any network in our league. The slime agrees about that spine, but then spends its length on long single arms: west across the river into Howrah and Domjur, north to Serampore, north-east to Barasat and south-west to Maheshtala. Those are the places the metro hasn't reached.

Where the slime goes and the metro doesn't (yet)

The biggest crowds more than 2 km from any open line, checked against the news. Unlike Delhi and Bengaluru, nothing is under construction for any of the top four: there are only plans and proposals.

Kolkata caveats. Station data is a Wikidata snapshot taken on 25 Oct 2026 (only stations marked "in use"), checked line by line against the news. Lines are drawn as straight links, so we measure 66 km against the official ~73 km. Leaving out the suburban trains matters most on the west bank: Serampore and Domjur have railway stations, so these are metro gaps, not rail deserts. We count residents only, so the Esplanade and Sector V office districts count for nothing. The Hooghly and the East Kolkata Wetlands aren't modelled as obstacles.

Chennai: the slime goes south, the metro hasn't (yet)

City number five. Same rules: GHSL 2025 population (11.5 million people in a 37 × 46 km box), 14 food hotspots, 6 independent growths of 50,000 agents, and we count the people living within 800 m of a line. The sea is a wall, as in Mumbai. The real network is every Chennai Metro line open on 25 Oct 2026: Blue (Wimco Nagar to the airport), Green (Chennai Central to St. Thomas Mount) and the first Phase 2 stretch, Poonamallee Bypass to Vadapalani, which opened on 11 Oct 2026 (CMRL press release). That's 51 stations and about 69 km officially. We left out the MRTS (Beach to St. Thomas Mount) and the suburban trains: they're run by Southern Railway, and CMRL's takeover of the MRTS hasn't been signed yet (TOI). So Chennai is metro against metro, like Delhi, Bengaluru and Kolkata. See the caveats. Pick the Chennai tab in the live sim above.

The metro wins clearly. Trim the slime to about the metro's length and it reaches 35% fewer people. The nearest trim is 58 km, a bit shorter than the metro's 63 km, but even the next step, 75 km (longer than the metro), reaches only 2.22M, still 17% short. The two networks barely agree: only 13% of the slime runs within 1 km of a metro line, the lowest overlap in our league. The metro is a cross through the centre plus a new arm to the west. The slime builds a hub near Central, then a long coastal spine north to Tiruvottiyur and a big southern arc through Adyar, Madipakkam and Chromepet towards Tambaram, where the metro doesn't go yet.

Where the slime goes and the metro doesn't (yet)

The biggest crowds more than 2 km from any open metro line, checked against the news. Three of the four already have suburban or MRTS stations nearby, so these are metro gaps, not rail deserts.

Chennai caveats. Station data is a Wikidata snapshot taken on 25 Oct 2026 (only stations marked "in use"), checked against the news. The new Phase 2 stretch runs non-stop from Porur Junction to Vadapalani (the stations in between aren't built yet), so we drew that 6 km link by hand. Lines are drawn as straight links, so we measure 63 km against the official ~69 km. Leaving out the MRTS and suburban trains matters: the slime's coastal spine and southern arc run close to rail lines that already exist. We count residents only, so the Guindy and OMR office districts count for nothing. Chennai is one big blob of people, so it gets only 14 hotspots (lowering the threshold only added tiny peaks at the box edge), and the slime network is fairly tree-like.

Hyderabad: the first city where the slime wins (with an asterisk)

City number six. Same rules: GHSL 2025 population (10.2 million people in a 48 × 41 km box), 13 food hotspots, 6 independent growths of 50,000 agents, and we count the people living within 800 m of a line. The real network is every Hyderabad Metro line open on 25 Oct 2026: Red (Miyapur to LB Nagar), Blue (Nagole to Raidurg) and Green (JBS Parade Ground to MG Bus Station). That's 57 stations and about 69 km officially (HMRL). Nothing new has opened since the Green Line in February 2020: Phase 2 (airport, Old City, Kokapet, Patancheru, Hayathnagar and more) is still waiting for the Centre's approval, and on 24 Oct 2026 the state was still picking a consultant for it (New Indian Express). We left out the MMTS suburban trains, so this is metro against metro, like Delhi, Bengaluru, Kolkata and Chennai. Pick the Hyderabad tab in the live sim above.

For the first time in the league, the blob comes out ahead. With a shorter network (51 km) it already reaches as many residents as the metro, and trimmed to exactly the metro's 56 km it reaches 1.85 million, 6% more. The other 56 km trim reaches 1.78 million, so call it a 2–6% win. Now the asterisk. Hyderabad's metro was built to reach jobs as well as homes: the Blue Line runs out to HITEC City and Raidurg, where few people live but huge numbers work. We count residents only, so this is the city where that rule flatters the slime most. The trimmed slime is also five separate pieces rather than one connected network (Bengaluru's was seven), and its biggest connected piece (31 km) reaches 1.03 million. So read it as "the metro skips some big crowds", not "a slime would run Hyderabad better". Hyderabad also has the lowest metro score per km in the league (31,000), partly because the city has added no new line in six years.

Where the slime goes and the metro doesn't (yet)

The biggest crowds more than 2 km from any open metro line, checked against the news. None of them has a line under construction.

The two biggest crowds of all, about 815,000 people around Mehdipatnam / Asif Nagar and 788,000 around Saidabad / Malakpet, sit 1–2 km from a line, so they just miss our 2 km gap rule. The slime puts hubs on both. The nearest real project is the 7.5 km MGBS–Chandrayangutta Old City line, the one Phase 2 corridor with work on the ground: about 830 of 888 properties have been cleared, and construction tenders will follow the remaining clearances (Hans India, 17 Oct 2026).

Hyderabad caveats. Station data is a Wikidata snapshot taken on 25 Oct 2026 (only stations marked "in use"), checked against the news. Lines are drawn as straight links, so we measure 56 km against the official ~69 km, and a shorter measured metro reaches fewer people. Residents only: HITEC City, Gachibowli and the Financial District count for nothing, which hurts the metro more here than anywhere else. Hussain Sagar and the other lakes aren't walls (they have almost no residents, so the slime has little reason to cross them). The MMTS suburban trains are left out. The box stops short of the airport (Shamshabad) and the far outskirts. The slime network is in pieces, and with 13 hotspots it is fairly sparse.

Pune: a near-draw, with the slime just ahead

City number seven. Same rules: GHSL 2025 population (6.9 million people in a 32 × 31 km box covering Pune and Pimpri-Chinchwad), 9 food hotspots, 6 independent growths of 50,000 agents, and we count the people living within 800 m of a line. The real network is every Pune Metro line open to passengers on 25 Oct 2026: the Purple Line (PCMC to Swargate) and the Aqua Line (Vanaz to Ramwadi), meeting at Civil Court. That's about 33 km officially and 28 stations in Wikidata (status table, 13 Oct 2026). The 23 km Hinjewadi–Shivajinagar Line 3 is not counted: it is largely built and 17 of its 23 stations have provisional safety clearance, but it still has no opening date after the Prime Minister's 3 Nov visit was deferred (Hindustan Times, 24 Oct 2026). Pick the Pune tab in the live sim above.

Pune's metro is the shortest in the league: two lines that cross in the old city. Left alone, the slime builds more than twice as much (67 km), so per km it scores lower, which is normal. The fair test is to trim it to the metro's length. At 30 km it reaches 1.16 million residents against the metro's 1.12 million, 4% more. Per km that's 39,100 against 38,800, under 1% apart, so call it a near-draw with the slime just ahead, the league's second slime win after Hyderabad. The trims on either side don't settle it: the next shorter one (19 km) reaches 0.98 million, and the next longer one (32 km, about 10% longer than the metro) reaches 1.38 million. The two networks barely agree on where to go. Only 7% of the slime runs within 1 km of the metro, the lowest overlap in the league. The metro runs north–south through PCMC and east–west through Kothrud and the old city. The slime runs a spine from Nigdi down through Rahatani to the centre and then out to Kondhwa, plus an eastern line through Kharadi and Hadapsar, the crowds no open line reaches yet. Caveats: the 30 km trim is two pieces (the bigger one, 20 km, reaches 0.93 million); residents only, so Hinjewadi's and Kharadi's IT parks count for nothing (that cuts both ways, since the metro doesn't reach Hinjewadi yet either); and lines are straight links, 29 km against about 33 km officially.

Where the slime goes and the metro doesn't (yet)

The biggest crowds more than 2 km from any open metro line, checked against the news. Two have a line under construction, one has a line built but not open, and the biggest has none of its own.

The biggest crowd of all, about 962,000 people around Mandai and the old city peths, is already on the Purple Line (0.6 km away). Further out, about 75,000 people around Wagholi wait on Line 2B (Ramwadi–Wagholi), approved in June 2025 with a four-year target (status table).

Pune caveats. Station data is a Wikidata snapshot taken on 25 Oct 2026 (only stations marked "in use"; Line 3 stations are ignored), checked against the news. Range Hill, between Khadki and Shivaji Nagar, has no Wikidata entry, so we join those two stations directly (trains run through either way). Lines are drawn as straight links, so we measure 29 km against the official ~33 km, and a shorter measured metro reaches fewer people. We count residents only, so Hinjewadi, Kharadi and Magarpatta's offices count for nothing. The Mula-Mutha river isn't a wall. The box stops short of Hinjewadi Phase 3, Chakan and Talegaon. With only 9 hotspots the slime is sparse, and the 30 km trim is in two pieces. Pune is also the only city in the league with a major line that is built but not open, so this result will go out of date when Line 3 opens.

League table: seven cities, one slime

"Which Indian metro looks most like a slime mold designed it?" Same model, same settings, seven cities, sorted by how close the slime gets in the equal-length fight. "People" means residents within 800 m of a line (GHSL 2025). "Gap" is how many fewer people the slime reaches than the real network when it's trimmed (or grown) to about the same length. A minus sign means the slime reaches more people. Every number is recomputed from the saved run results.

The slime now wins two: Hyderabad and Pune. Trimmed to Hyderabad's metro length (56 km), the slime reaches 1.85M residents against the metro's 1.74M, 6% more (a neighbouring trim, also 56 km, gives 1.78M, 2% more). Read that carefully: Hyderabad's metro was built to reach offices too (HITEC City, Raidurg), and we count only where people live, so it's the city where "residents only" flatters the slime most. The trimmed slime is also five separate pieces, not one connected network. Pune is closer to a draw: at 30 km against the metro's 29 km, the slime reaches 1.16M against 1.12M, 4% more, but per km it's under 1% ahead, and the trim is two pieces. Pune's metro is short (two lines meeting in the old city), so the slime simply heads for the big southern and eastern crowds (Kondhwa, Hadapsar) that no open line reaches yet. The real network wins the other five fights. Bengaluru is the closest of those (6%): its metro is young and still a sparse cross, so a blob that simply chases crowds almost matches it. Kolkata (31%) has the densest line of all: its 1984 Blue Line runs down a strip with 77,900 people per km, and no amount of slime wandering can beat that. Chennai (35%) has the biggest gap on paper, but part of that is length: the nearest equal-length slime is 58 km against the metro's 63 km, and the next step up (75 km, longer than the metro) still loses by 17%. Notes: Mumbai's slime tops out at 287 km, short of the 347 km rail network, so that fight isn't quite equal-length; Mumbai's real network includes the suburban trains, the other five are metro only (Chennai's MRTS and suburban trains and Hyderabad's MMTS are left out). Real km are straight lines between stations, a little under the official figures.

How it works

1. Population. GHSL 2025 residential population (EU Joint Research Centre), 1 km cells resampled to 150 m, over a 72 × 61 km box around Delhi holding 33.3 million people.
2. Food. Hotspots are local peaks of smoothed density, at least 3 km apart. Each one's scent grows with the square root of the people living within 3 km of it. There are 34 in total.
3. Slime. Jones-style Physarum agents. Each one senses ahead, left and right, turns toward the strongest scent, and leaves a deposit. The trail spreads out and evaporates. The offline research run used 50,000 agents for 1,500 steps, and the live version above is a lighter 12,500-agent cut.
4. Consensus. We ran six independent growths. A tube only counts if at least 3 of the 6 runs grew it, so the result isn't one lucky run.
5. Compare. Metro segments are straight lines between adjacent stations (275 stations, 292 links, from Wikidata). Coverage counts the residents within 800 m and 1.5 km of each network. For the equal-length test, the slime was grown to 386 km against the metro's 368 km.

Caveats

• Station data is a Wikidata snapshot, so very recently opened lines may be missing. The metro is drawn as straight lines between stations, not the actual track.
• Food is just 34 point hotspots, which is why the slime under-serves the dense core.
• There's no terrain yet. The slime will happily cross the Yamuna, the Ridge forest and the airport, whereas Tero's original experiment used light to mark areas the slime should avoid.
• Coverage counts residents only, not jobs, trips or real travel demand. This is a toy that turned out to be surprisingly well calibrated, not a transport plan.

Data & credits

• Tero, A. et al. (2010). Rules for biologically inspired adaptive network design. Science 327, 439–442. That's the oat-flake paper the studio is named after.
• Population: GHSL GHS-POP R2023A, epoch 2025, European Commission Joint Research Centre.
• Stations and links: Delhi Metro , Namma Metro and Kolkata Metro on Wikidata (CC0).
• Agent model after Jeff Jones (2010), "Characteristics of pattern formation and evolution in approximations of Physarum transport networks".
• Reality checks come from public news reports on the Golden Line (Tughlakabad to Sangam Vihar), Gurugram Metro Phase 1 and the Rithala to Kundli corridor.