Requant research
Live broadcasts
A broadcast is a network that plays by the server's clock, one for everyone who watches, and that viewers steer together. It turns the lab's questions into something to watch: what happens to a chain when the network splits, a pool hides its blocks or a crowd keeps pulling at it. How to watch and step in is in the guide to broadcasts; this page is how they work and what they showed.
Channels
| Channel | Network | Pace | A block on screen |
|---|---|---|---|
| Bitcoin | Bitcoin profile: 10-minute blocks, 2016-block window | ×20 | about every 30 s |
| Bitcoin ×30 | the same network | ×30 | about every 20 s |
| Bitcoin ×50 | the same network | ×50 | about every 12 s |
| Crowd | the same network with 50 bots stepping in | ×20 | about every 30 s, until the crowd steps in |
- Seasons. A season lasts 6 hours of real time; season
nstartsn × 6hours after the Unix epoch, and its seed isn × 16plus the channel's number. At the start of a season the world begins anew and impacts are cleared. - The history is built ahead. When the plan changes, the server rebuilds the season's history from the moment of the change; the moments before it do not change. Rewinding is free: the past of the season is already in the history.
- Pace. The pace is world seconds per real second. The three Bitcoin channels show the same network at three speeds; before 3 November 2026 the faster two were Litecoin and Dogecoin profiles, which looked too much alike on screen.
Stepping in
| Impact | What it does | Parameters |
|---|---|---|
| Split | cuts the network in two | by default the pools split about evenly by power |
| Switch off | switches nodes off; a pool's node takes its miners' power with it | 1-6 nodes |
| Delay | a delay on every link | 1-30 s |
| Hide blocks | a pool mines in secret (selfish mining) and releases its blocks when the network catches up, cancelling honest ones | one pool |
| Power | a pool's miners surge | ×2-5 |
| Loss | a share of messages is lost on every link | 1-50% |
- Timing. An impact starts 15 s after the press, so every open tab has the new history in time, and lasts 90 s of real time (a split and hidden blocks 180 s, so each side has time to find its own blocks), multiplied by the pace in world time.
- Impacts add up. The network splits along every running split at once (two splits give up to four parts, each with its own pools), the longest delay and the largest loss apply, power surges multiply up to ×10.
- Balance. Each viewer has 3 charges per channel and gets one back every minute; a split and hidden blocks cost two, the rest one. At most 8 impacts run at once, with places per kind (2 splits, 3 switch-offs, one delay and one loss, one hiding and one surge per pool). When there is no place, an impact waits in a queue and the viewer sees when it starts.
- Storms and calm. An impact joins the running ones while their storm, with it, stays within 240 s; otherwise it waits until 90 s after the storm ends, so the network converges and the result is on screen.
- Support. A viewer can support someone else's impact once: it lasts 15 s longer, up to 240 s and not past its storm.
- Names, not addresses. Each viewer appears under a name made from a salted hash of the address (two words and a number); the address cannot be recovered from the name.
What a crowd does to a network
The server can play a channel's bots through a season without waiting and compare the history with the same season without impacts. A whole crowd season (6 hours of broadcast, 120 hours of the world, Bitcoin profile, pace ×20), November 2026:
| Bots | Impacts | Network split | Parts | Blocks cancelled | A main-chain block | Longest wait | Waited in the queue | History build |
|---|---|---|---|---|---|---|---|---|
| 10 | 254 | 45% of the time | up to 4, 1.6 on average | 21% | every 12.1 min | 103 min | 78, 45 s on average | 0.11 s |
| 50 | 369 | 56% | up to 4, 2.3 on average | 28% | every 12.7 min | 80 min | 103, 50 s on average, up to 5.5 min | 0.14 s |
Before the balance above (places per kind, the queue, storms and calm), the same simulation showed what an unbounded audience does:
| Bots | Network split | Parts on average | Blocks cancelled | A main-chain block | Longest wait |
|---|---|---|---|---|---|
| 0 | 0% | 1 | 0% | every 10.1 min | 84 min |
| 5 | 76% of the time | 2.8 | 36% | every 15.4 min | 88 min |
| 10 | 92% | 6.6 | 46% | every 19.3 min | 2 h 50 min |
| 25 | 100% | 24 | 66% | every 28.3 min | 3 h 53 min |
| 50 | 86%* | 30 | 54% | every 22.2 min | 3 h 31 min |
* The season's 3000 impacts ran out an hour before its end; the last hour the network was left alone.
- Five active viewers were enough to keep the network split most of the time. A split lasts 3 minutes of broadcast, an hour of the world, and the next one began before the last ended; a third of the blocks were cancelled.
- With dozens of viewers a single impact stopped being visible. The network was cut into dozens of parts, most of them without miners, and the race went on between the few parts that had pools.
- The chain did not stop; it slowed two to three times, with hours of the world without a main-chain block.
- With the balance the network splits into 2-4 understandable parts, each impact's effect can be seen, and a season's history builds tens of times faster.
Boundaries
- Bots are scripted viewers: they act through the same limits as people, with fixed probabilities per kind. The tables show what such a crowd does to this network, not what people would do.
- The future is in the data. The season's history is built to its end under the current plan, so whoever reads the tab's data can see what will happen if nobody steps in. Predictions in the broadcast need a history that ends at "now" with a margin, and are planned on the roadmap.
- A model network. Broadcasts play the lab's model of a Bitcoin-profile network scaled for a demonstration (60 nodes); they are not the Requant testnet.