Skip to content

About

SKALA: an interactive reconstruction of the Chernobyl unit 4 process computer and its RBMK-1000 core, built from INSAG-7, Karpan and the surviving DREG record.

Topics

Resources

Contributing

Stars

0 stars

Watchers

0 watching

Forks

Repository files navigation

СКАЛА: Chernobyl unit 4 process computer, reconstructed. A VFD clock reads 01.23.40 beside a lit АВАР ЗАЩИТА 5 lamp.

Open the simulator

Runs in any browser No install 211 real rod positions Validation checks Code license: Apache 2.0 Content license: CC BY-SA 4.0

SKALA (СКАЛА) was the process computer of each RBMK-1000 reactor. At Chernobyl unit 4 it watched thousands of sensors, ran the PRIZMA core calculation every five minutes and logged events to its diagnostic recorder, DREG. It did not control the reactor. The operators did, asking SKALA for numbers by keying codes on a request device and waiting for the answer to light up.

This project rebuilds that experience in a web page: the rod-position wall, the unit control board, the SKALA console and its printers, driven by a reactor model built on the real control-rod map of unit 4. You can run the plant from routine operation, or step into the 01:22:30 state of 26 April 1986 and follow, or change, what happened next.

Dotted and linked terms show a short definition when you hover over them.

Important

Read this first. This is a reconstruction, not an emulation and not a reactor simulator.

  • SKALA's software was never published, nor the V-3M computer's instruction set. No original code runs here. The console's behaviour is rebuilt from surviving printouts, messages, timings and testimony.
  • Most request codes are invented, in a real grammar. The code structure, the two display sets and a handful of real codes come from the Chornobyl Family reconstruction, built with former Chernobyl staff. That reconstruction follows the modernized SKALA of unit 3, which ran until 2000, so unit 4's codes in 1986 may have differed in detail. The full code table is not public, so the other codes follow the same rules but are this simulator's own. Printout layouts are reconstructions too.
  • The physics is simplified and tuned. It is a one-dimensional model along the core's height. The explosion timing matches the record because one parameter was tuned to make it match, and a 7 % change to that parameter changes the outcome. See validation.
  • Use it to understand the mechanisms, such as why AZ-5 raised power first, or why the operating margin mattered. Do not treat it as evidence for exact numbers.

Every screen element is labelled on the page as documented or reconstructed. Corrections with sources are very welcome (how).

Animated capture: AZ-5 is pressed at the end of the run-down test. Rod dials begin to turn, the teletype types АВАР ЗАЩИТА 5, power climbs on the VFD readouts, and the session ends with signals lost.
The last seconds of a 26 April run: AZ-5, the overpower trip, the teletype cut off when SKALA loses power.


What you can do with it

Try this What you get
Read the plant as the shift did 211 selsyn-style dials on the real channel grid, eight VFD readouts, a power trend and axial power bars.
Query SKALA Key a seven-character code (system, part, quantity, address, such as П0N0000 for thermal power), flip ЗАМЕР, and wait for the value, its deviation from the nearest setpoint, and the code on one of two display sets. ORM, the margin that mattered most that night, was never on a continuous display. You had to ask.
Watch the printers The RTA-80 teletype types at its real 7 characters per second. DREG prints in numbered cycles. PRIZMA arrives on fanfold paper at the end of its 5-minute cycle, as a channel map of the core printed in two halves. The teletype also logs every setpoint a measured parameter crosses. With sound on you hear them print, and every emergency protection sounds the alarm horn.
Replay 26 April 1986 Start the run-down test at 01:23:04, press AZ-5 at 01:23:40, and compare your run with the surviving record in the archive tab.
Ask "what if?" Run the same test with the rod pattern of 00:39, try AZ-5 without the test, or try to shut the reactor down safely. Pick rods by sector or draw your own group on the dial panel, and a short debrief tells you how the run went.
See what nobody could see A fold-out panel shows the reactivity balance and the axial flux, steam and xenon profiles in real time.
Full simulator: rod-position dial panel, SKALA request device with VFD tubes, and printer bay. Close-up of the SKALA request device: two display sets showing value, setpoint deviation and code, above the seven-column code keyboard.
The control-room wall: rod dials, SKALA console, printers. The request device, laid out after the Chornobyl Family reconstruction.

Run it

  • Online: open the live simulator.
  • Offline: download this repository and double-click index.html. It is a single file with no install and no server. An internet connection is only used for the PT Sans and PT Mono typefaces.

The training manual walks through every panel, lists the request codes and gives eight graded exercises, from reading the plant to replaying the accident: PDF · read on GitHub. In Russian (на русском): PDF · read on GitHub.

Documented or reconstructed

documented taken from the sources   inferred worked out from the sources   invented made up so the simulator works

Element Status Basis
Positions of all 211 control rods at 00:39 and 01:22:30 documented Karpan's cartograms, checked cell by cell against INSAG-7 Fig. II-6
Which rods are АР (automatic) and УСП (from below), and the 12 detector cells documented Karpan's map typography
Which of the other 175 rods are РР, ЛАР or АЗ unknown Not stated in any source found. Shown as one group.
Teletype speed, message format and cut-off documented Karpan pp. 359–361
DREG cycle numbering, "=0" convention, message texts documented DREG transcript, Karpan, INSAG-7
DREG raw signal codes documented Only К06L005 is known. Other lines show text only.
PRIZMA 5-minute cycle, odd-hour printouts, restart time documented Karpan pp. 338, 350–351
Pump numbering, coast-down and trip times documented Karpan App. 3 oscillogram chronology
Fuel-channel positions inferred Every lattice cell not used by a rod or detector
Xenon distribution at 01:22:30 inferred Solved backwards from the measured flux shape (INSAG-7 Fig. II-5)
Request device layout, code grammar, display sets, error codes 301/303, and codes П0N0000, Т0N7100, К0S0000, К5N0000, К0S/К0G + channel reconstruction Chornobyl Family, built with former Chernobyl staff, after the modernized unit 3 system (in service to 2000). Channel coordinates are octal, which the plant grid confirms.
All other request codes, the subprogram numbers except 0301, and ПС/АС as the falling setpoints invented Same grammar, but the full table is not public
PRIZMA channel cartogram form (two halves, rulers, four-character cells) reconstruction Chornobyl Family printer video. Per-channel flow and boiling margin left out: not modelled
PRIZMA summary lines, rod cartogram layout, teletype setpoint wording invented No original sheet reproduced
Rod "sectors" and the SIUR assist invented Simulator conveniences, labelled as such

How close it gets

Check Record Simulator
Operating margin at 01:22:30, from the real rods (not tuned) 6–8 rods 6.4
Operating margin with the 00:39 rods 16.5 14.3
Coolant flow just before AZ-5 51 000 m³/h 50 400
Overpower trip after AZ-5 about 3 s 2.7 s
Reactor destroyed after AZ-5 about 5 s 5.4 s

Warning

The last two rows were tuned. Change the water-column absorption by 7 % and the 01:22:30 state either stops exploding or the safer 00:39 case starts to trip. INSAG-7 (p.142) reports the same extreme sensitivity in the Soviet three-dimensional codes. Full results, the reasoning and how to reproduce them: docs/validation.md.

How it fits together

flowchart LR
  subgraph data["references/ · from the sources"]
    map["Rod map<br/>211 rods · 3 АР banks"]
    rec["Record<br/>flux shape · timings · messages"]
  end
  subgraph model["src/sim.js · reactor model"]
    core["Axial core<br/>28 slices · xenon · void · displacers"]
    plant["Plant<br/>8 pumps · drums · run-down test"]
    prot["Protections<br/>AZ-5 · АЗМ · АЗСР"]
  end
  subgraph skala["src/sim.js + ui.js · SKALA"]
    prizma["PRIZMA<br/>5-min cycle"]
    dreg["DREG<br/>2–4 s cycles"]
    req["Request device"]
  end
  subgraph out["what you see"]
    desk["Desk and dials"]
    tty["Teletype"]
    lpt["Drum printer"]
  end
  map --> core
  rec --> core
  core <--> plant
  core --> prot
  core --> prizma
  plant --> dreg
  prot --> dreg
  prot --> tty
  prizma --> lpt
  dreg --> lpt
  req --> prizma
  core --> desk
  classDef src fill:#e6e1cf,stroke:#9d988a,color:#1b221e
  classDef mod fill:#c4cdc0,stroke:#5c655c,color:#1b221e
  classDef sk fill:#0c1d10,stroke:#46f070,color:#b2ffa8
  classDef ui fill:#23292a,stroke:#23292a,color:#eef0e6
  class map,rec src
  class core,plant,prot mod
  class prizma,dreg,req sk
  class desk,tty,lpt ui
  style data fill:#f3f1e6,stroke:#9d988a,color:#485349
  style model fill:#eef1ea,stroke:#8d9a8a,color:#485349
  style skala fill:#d6d3c8,stroke:#9d988a,color:#485349
  style out fill:#eef1ea,stroke:#8d9a8a,color:#485349
Loading

Repository layout

skala-simulator/
├── index.html              the simulator, built from src/ (GitHub Pages serves this)
├── src/
│   ├── template.html       page layout and styles
│   ├── sim.js              reactor physics, plant, protections, SKALA programs
│   ├── ui.js               dials, VFD readouts, request device, printers
│   ├── data.js             core map in JavaScript (generated, do not edit)
│   ├── glossary.js         definitions behind the dotted hover terms
│   ├── gen_data.py         rebuilds data.js from references/
│   └── build.py            rebuilds index.html from the files above
├── references/             data and notes recreated from the sources, with page numbers
│   ├── rbmk4_cps_map.csv           all 211 rods + 12 detector cells, both maps
│   ├── insag7_fig_II-6_rods_012230.csv
│   ├── fuel_lattice_inferred.csv
│   ├── oscillogram_chronology_karpan_app3.csv
│   ├── dreg_table_karpan_app3.csv
│   ├── message_catalogue.csv
│   ├── validation_points.csv
│   ├── insag7-findings.md · karpan-findings.md
│   └── sources.md                  where every source can be found
├── tests/check.js          checks against the historical record (prints PASS/FAIL)
├── docs/
│   ├── SKALA-Training-Manual.pdf   the training manual (also as .md)
│   ├── SKALA-Training-Manual.ru.pdf  the same manual in Russian (also as .md)
│   ├── validation.md
│   └── assets/             images for this README
├── .github/workflows/check.yml     runs the checks and verifies index.html on every push
├── CONTRIBUTING.md · CHANGELOG.md · CITATION.cff · LICENSE · NOTICE · LICENSES/
└── README.md

Changing it

You need Python 3 to rebuild, and Node.js to run the checks.

cd src
python3 gen_data.py      # only if you changed a CSV in references/
python3 build.py         # rewrites ../index.html
cd ..
node tests/check.js      # every line should say PASS

Sources

Page-by-page notes are in references/. No books or reports are redistributed here.

Thank you

This reconstruction rests on the shoulders of experts and their work (and some impressive AI to tie it together).

  • Nikolai Karpan, whose book Чернобыль. Месть мирного атома preserved the rod maps, shift logs and recorder data the core model is built on.
  • The IAEA's International Nuclear Safety Advisory Group, whose INSAG-7 report supplied the measured flux shape, the rod positions to cross-check against, and a frank account of the reactor's design flaws.
  • Chornobyl Family, whose restoration of real SKALA hardware shows how the console, printers and DREG actually behaved.
  • Linus Sturm and R. Plaskov, for transcribing the DREG protocol of the last 40 seconds.
  • Claude (Anthropic), the AI assistant that read the Russian sources side by side with the English ones, extracted the rod map from the book's typography, cross-checked it cell by cell against INSAG-7, and ran many simulation runs to test the model against every figure in the record.

Note

Working this way was a lesson in what AI can do to aid historical reconstruction. Its value was not knowing the answers, but holding scattered Russian and English sources together, reconstructing them into tabulated data, and testing them against each other far faster than one person could. Every finding was checked against the source documents, and the gaps are marked accordingly.

License

Built by Fernand Gouveia (2026).


On 26 April 1986 reactor No. 4 of the Chernobyl Nuclear Power Plant was destroyed at about 01:23:45.
Two plant workers died that night. 134 people were later diagnosed with acute radiation syndrome.
This project is made in their memory and for the people who worked to understand what happened.

About

SKALA: an interactive reconstruction of the Chernobyl unit 4 process computer and its RBMK-1000 core, built from INSAG-7, Karpan and the surviving DREG record.

Topics

Resources

Contributing

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages