microbiome

This application allows the user to witness the activity of a virtual microbial community.

Play in your browser

The Strata micro-environment (below) runs entirely in your browser, compiled to WebAssembly and drawn with WebGL 2, no server needed. Drag to pan and scroll (or pinch) to zoom; the buttons switch the field, the clock and the statistics overlay:

• https://microbiome.play.danielstephenson.dev
• or find it among the other games at https://danielstephenson.dev/play

The original 2D grid stays available at https://microbiome.play.danielstephenson.dev/classic/ (it also works in browsers without WebGL 2).

This is in addition to the existing server-backed live viewer at https://microbiome.preponderous.org, which stays as it is.

To build the Strata page yourself, see docs/STRATA.md (emcmake cmake -S strata -B build-web). To build the classic page, activate an Emscripten SDK (CI uses 6.0.11) and run make web (or web/build.sh). It writes web/build/index.html, index.js and index.wasm; serve that directory with any static file server, e.g. python3 -m http.server -d web/build 8000, and open http://localhost:8000. The entry point is src/browser.cpp (the page is web/shell.html): it runs the same Microbiome the console app and web server run, one tick every 200 ms, starting a new generation after an extinction just like mb_webapp. It reports no usage, writes no log files, and the native targets never compile it. .github/workflows/browser.yml builds both pages on every pull request and push, smoke-tests the Strata page in headless Chromium and deploys the site (Strata at /, classic at /classic/).

Micro-environment on Strata (opt-in)

mb_strata simulates a much richer micro-environment with the Strata engine: a 128×128 chamber (any size up to 1025×1025) with diffusing nutrient, waste and oxygen fields, solid grains that biofilms grow on, and three species of microorganism as Strata entities:

• aerobes (blue) swim up the nutrient gradient and need oxygen, which only diffuses in from the air interface at the top edge, for a full yield;
• fermenters (orange) are anaerobes that oxygen harms; they settle into biofilms (flat tiles) on grains, on the chamber wall or when crowded, and disperse again when starving;
• cross-feeders (pink) live on the others' waste and swim up the waste gradient.

The console app's traits carry over: starting energy and metabolic rate ranges, binary fission at 900 energy with a cost of 100, biomass returned to the environment on death and an 80% chemotaxis bias, now applied to a run-and-tumble walk up real gradients. Nutrient seeps below the oxic band set up opposing gradients, as in a Winogradsky column.

The overlay is drawn with Strata's unlit overlays (ABI 1.5): the text is Strata's 8×8 bitmap font, the panels, colour bar and sparkline are overlay lines, and the species markers are overlay points. It shows the tick and simulated time, the speed or pause state, each species' population and the biofilm count, births and deaths, the chamber's nutrient, waste, oxygen and cell-energy totals with their change per tick (next to the 1,800 units of nutrient the seeps add each tick), a legend for the current field view and the population of each species over the last 960 ticks. It only reads the simulation, so a run's state hash is the same with it on or off.

It is deterministic: the simulation draws every random number from one seeded generator, runs on a fixed tick and never reads a clock, so a seed gives a bit-identical state hash after the same number of ticks. Measured with mb_strata --bench (simulation only, one thread, GCC 9.4 Release build on an i9-10850K): the default 128×128 chamber, with a mean population of about 1,100, runs at about 1,150–1,250 ticks per second; a 256×256 chamber (--size 256 --per-species 1200, mean population about 2,000) at about 290.

Building it needs CMake 3.24 or newer, a C++17 compiler and SDL3's X11 headers (libx11-dev libxext-dev libxrandr-dev libxcursor-dev libxi-dev libxfixes-dev libxss-dev on Debian/Ubuntu); CMake fetches Strata at the commit pinned in strata/strata.ref:

make strata                                   # or: cmake -S strata -B build-strata && cmake --build build-strata -j4
ctest --test-dir build-strata --output-on-failure
build-strata/mb_strata                        # a window; H prints the controls
build-strata/mb_strata --headless --ticks 600 --screenshot out.png    # no window, no X server
build-strata/mb_strata --headless --ticks 960 --frames frames/ --frame-every 12
build-strata/mb_strata --bench --ticks 2000   # ticks per second, no rendering

In the window: WASD/arrows pan, Q/E rotate, mouse wheel or +/- zoom, drag to pan, 1–4 switch the field view, space pauses, ,/. change speed, Tab shows or hides the statistics overlay, R starts a new seed. Options: --seed, --size, --per-species, --view, --zoom, --yaw, --tps, --no-hud, --hud-scale N (glyph size in multiples of 8 pixels; by default it follows the window height), --record run.usda (below).

Shareable replays (--record)

mb_strata --record run.usda writes the run as a replay in the usd-replay/1 contract: one OpenUSD layer that any player can show without knowing the simulation. Each species is one Points set with stable ids, the chamber's heatmap (in the --view colours) is a coarse colour grid, the grains are static cubes, and every sample carries the run's state hash, so anyone can re-run the seed and check it.

build-strata/mb_strata --headless --ticks 2000 --no-hud --record run.usda --record-every 10 --record-field 32
pip install "usd-recorder @ git+https://github.com/Preponderous-Software/[email protected]"
usd-recorder replay validate run.usda && usd-recorder replay convert run.usda run.usdc

Recording only reads the simulation: the state hash is the same with and without --record, which a test checks tick by tick and CI checks again on a full run. Samples are taken every --record-every N ticks (default 10) plus the first and last tick; --record-field N sets the heatmap's columns (default 32). In the window the file is written on quit (R, a new chamber, ends the recording). The writer is usd-recorder's dependency-free C++ header, vendored at v0.2.0 (MIT) in strata/vendor/usd_recorder/, so no OpenUSD is needed. CI uploads each pull request's replay, validated and converted to .usdc, as the mb-strata-replay artifact.

This build is separate from everything else: make, mb_app, mb_webapp (microbiome.preponderous.org), the classic browser build and both test suites are unchanged and do not need CMake or Strata. Built with Emscripten, mb_strata is the arcade page, on Strata's WebGL 2 (OpenGL ES 3.0) backend. .github/workflows/strata.yml builds and tests it in CI. The design, what moved to Strata and what stayed, and the known gaps are in docs/STRATA.md. mb_strata reports no usage and writes no log files.

Example Console Output

==============================
    .        o     o  o       
          .        +     .    
 +  o           +  +        . 
       o        +  +  o  o    
    +           o  .  +  o  . 
    o  !  +  !  o             
 +  +        +              o 
          +     +  .  +  +    
 o  +        .  !     !       
       !  o  .     .        o 
==============================

Name: Simulation 1
Size: 10x10
Microorganisms: 45
Dead Microorganisms: 55
Total Energy: 1858

Ticks elapsed: 16

Testing

The project uses Catch2 for all of its tests, split into two suites: ./mb_tests (the simulation, configuration, logging, usage reporting and replay recording) and ./mb_web_tests (the web server's HTTP endpoints, which link Ulfius).

How to run the tests in the dev container

1. Install Docker
2. Clone the repository
3. Open the project in VSCode.
4. Install the Remote-Containers extension.
5. Click on the green button in the bottom left corner of the window.
6. Select "Reopen in Container".
7. Open a terminal in VSCode.
8. Run the ./run_tests.sh script, or build a suite with make catch2_tests / make catch2_web_tests and run it with ./mb_tests / ./mb_web_tests.

Available Test Commands

• make catch2_tests - Build the Catch2 tests that need no Ulfius as ./mb_tests (recommended)
• make catch2_web_tests - Build the web server's Catch2 tests as ./mb_web_tests (needs libulfius-dev and pkg-config, like make webapp)
• make test - Alias for catch2_tests
• ./run_tests.sh - Run both suites; it stops at the first suite that fails to build or fails a test and exits non-zero
• make - Builds both test binaries alongside the applications, so a test suite that stops compiling fails the build rather than going unnoticed
• ctest --test-dir build-strata - after make strata: the opt-in Strata build's Catch2 tests (diffusion mass conservation, chemotaxis direction, same-seed determinism, the console app's fission/death rules), headless render smoke tests, statistics-overlay tests and replay tests (recording leaves the state hash unchanged); see docs/STRATA.md

CI (.github/workflows/docker-build.yml) builds the production image and then executes both suites inside it, so a failing assertion fails the build too, not only a failing compile.

Every build target lists the sources and headers it is built from, so changing any of them rebuilds the affected binaries rather than re-running a stale build after a change elsewhere.

The web server's HTTP tests (GET /api/state, the index page's footer link and share tags, and GET /og.png) are a suite of their own because make catch2_tests deliberately links no Ulfius dependency. Run ./run_tests.sh to cover both.

Catch2 Features

• Better test reporting and failure diagnostics
• Test filtering by tags: ./mb_tests [microorganism]
• Verbose output: ./mb_tests --success
• List all tests: ./mb_tests --list-tests

Running the application

How to run the application with Docker-Compose

1. Install Docker and Docker-Compose
2. Clone the repository
3. Open a terminal in the root directory of the repository
4. Run the following command: docker-compose up --build --remove-orphans

How to run the application in the dev container

1. Install Docker
2. Clone the repository
3. Open the project in VSCode.
4. Install the Remote-Containers extension.
5. Click on the green button in the bottom left corner of the window.
6. Select "Reopen in Container".
7. Open a terminal in VSCode.
8. Run the ./cr.sh script or build the project with make and run it with ./mb_app.

How to view the simulation live in a browser

The webapp target builds a small Ulfius-based web server (mb_webapp) that runs the simulation continuously and serves a live view of it.

With Docker Compose:

1. Run docker-compose up --build microbiome-webapp
2. Open http://localhost:8080 in a browser

Building locally (in addition to make and g++, this needs libulfius-dev and pkg-config, e.g. apt-get install pkg-config libulfius-dev on Debian/Ubuntu):

1. Run make webapp
2. Run ./mb_webapp (set MICROBIOME_WEB_PORT to use a port other than the default 8080)
3. Open http://localhost:8080 in a browser

The page polls GET /api/state a few times a second for the current grid, microorganisms, and biomatter as JSON, and once the population goes extinct the server starts a new generation automatically.

The page carries a description, a canonical link and Open Graph / Twitter card tags for https://microbiome.preponderous.org. Its link-preview image (og:image) is assets/og.png, a 1200×630 card served at GET /og.png: microorganisms and biomatter on a grid in the viewer's four colours, rendered in 3D with Strata, with the title over it. It is drawn from og-card/card.json by community-site-kit's tools/og-card (see og-card/README.md), and its alt text on the page must match that spec's alt. The PNG is compiled into the server as src/header/ogImage.h, so the binary needs no files beside it. After redrawing the PNG, run make og-image-header (needs xxd) and commit both. The web server suite (./mb_web_tests) fails if they differ.

Recording a replay

mb_app --record <path.usda> records the first simulation's history as a time-sampled OpenUSD stage, one time code per tick (10 per second), that can be scrubbed in usdview, Blender or any USD tool:

make mbapp
./mb_app --record run.usda

Every living microorganism and every piece of biomatter is an entity at its grid coordinates (x, y, 0) (y grows downwards, as in the console view), visible while it exists, with kind (microorganism or biomatter) and energy values; microorganisms also carry metabolicRate and timesEaten. The stage carries the alive and dead counts and the totalEnergy per tick. The other simulations run as before. Without --record nothing is recorded and the app behaves exactly as it always has.

The recording is written by usd-recorder's dependency-free C++ writer, vendored at v0.1.0 (MIT) in src/header/vendor/usd_recorder/, so no OpenUSD installation is needed to build or record. src/header/usdRecording.h maps the simulation onto it.

The replay above is a recorded run of 1,061 ticks (a 2.5 MB stage) rendered top-down with headless Blender's USD importer and Cycles, coloured by kind.

Notable Classes

Microbiome

The Microbiome class represents a virtual microbial community. It is an extension of the Environment class provided by env-lib-cpp. Within the microbiome, there are a number of microbes that are able to interact with each other and the environment. The Microbiome class is responsible for managing the microbes and the environment they exist in.

Microorganism

The Microorganism class represents a single microbe. It is an extension of the Entity class provided by env-lib-cpp. The Microorganism class is responsible for managing the microbe's energy and metabolism.

Biomatter

The Biomatter class represents decomposing biomass left behind when a microorganism dies. It is an extension of the Entity class provided by env-lib-cpp. Living microorganisms bias their movement toward it (chemotaxis) and can forage it for energy, modeling nutrient recycling instead of dead microorganisms simply vanishing from the energy budget.

WebServer

The WebServer class (built on Ulfius, see #19) runs a Microbiome simulation continuously in the background and exposes its state over HTTP, so it can be viewed live at http://localhost:8080 instead of only in the console.

Simulated Mechanics

See RESEARCH.md for the microbiology background behind these mechanics and what's still missing.

• Metabolism - every microorganism loses energy each tick and dies once its energy reaches zero.
• Decomposition - a dead microorganism's biomass becomes Biomatter in the environment rather than an inert corpse.
• Chemotaxis - microorganisms bias their movement toward neighboring locations that have Biomatter, rather than moving with pure uniform randomness.
• Foraging - a microorganism sharing a location with Biomatter can consume it for energy, depleting it over time.
• Reproduction - a microorganism that accumulates enough energy divides via binary fission into two daughter cells, each inheriting half its remaining energy (minus the energetic cost of dividing) and its metabolic rate.

Usage reporting

The console app (mb_app) reports to trace by default: one startup event per launch, carrying the program name (microbiome), its version from version.txt, and a random installation ID. Nothing about you, your machine or the simulation is sent (the trace server sees the IP address of the request, as every web server does). The live-view web server (mb_webapp) and the test suites report nothing.

The first run prints one line saying so on stderr and writes a small settings file, usage-reporting.conf, to $XDG_CONFIG_HOME/microbiome/ (by default ~/.config/microbiome/; ~/Library/Application Support/microbiome/ on macOS, %APPDATA%\microbiome\ on Windows). To turn reporting off:

• set enabled=false in that file, or
• set TRACE_USAGE_REPORTING=off or DO_NOT_TRACK=1 in the environment (this turns it off for every trace-reporting program, and nothing is printed or written).

The installation ID is a random UUID, made the first time reporting is on and kept in trace-install-id in $XDG_DATA_HOME/microbiome/ (by default ~/.local/share/microbiome/; ~/Library/Application Support/microbiome/ on macOS, %APPDATA%\microbiome\ on Windows). It is not derived from anything about you or your machine; it only lets trace count installations rather than launches. Delete the file to reset it. Setting TRACE_INSTALL_ID sends that value instead, and the file is left alone. Every opt-out above also stops the ID: with reporting off, the file is never created, read or sent.

The event is sent in the background by the vendored trace-client-cpp header (src/header/trace_client.hpp) through the system curl; if curl is missing or the machine is offline, nothing is sent and the simulation is unaffected. MICROBIOME_USAGE_REPORTING_ENDPOINT points it at another server, e.g. a local one while testing. Details: https://danielstephenson.dev/usage-reporting

📄 License

This project is licensed under the Preponderous Non-Commercial License (Preponderous-NC).
It is free to use, modify, and self-host for non-commercial purposes, but commercial use requires a separate license.

Disclaimer: Preponderous Software is not a legal entity.
All rights to works published under this license are reserved by the copyright holder, Daniel McCoy Stephenson.

Full license text:
https://github.com/Preponderous-Software/preponderous-nc-license/blob/main/LICENSE.md