v8.0.0

8.0.0 (2026-10-15)

🚀 Features

• numpy 2.x compatibility (#240)
• Add distribution based placement (#242)
• Rust acceleration for segment intersection touch detection and Poisson-disk placement (#256)
• bsb-core: add an after_simulation hook (#257)
• ⚠️ bsb-core: give the prepared simulation a hook of its own (#263)
• ⚠️ bsb-core: drop the adapter's post_prepare argument (#278)
• ⚠️ bsb-core: parameters answer what a model's values are, per arity (#260)
• bsb-core: configured randomness, with replicates and reproduction (#264)
• ⚠️ bsb-core: draw placement from the configured randomness (#276)
• ⚠️ bsb-core: draw connectivity from the configured randomness (#277)
• ⚠️ bsb-core: a run ends with one results file, whatever it took to make it (#266)
• ⚠️ bsb-core: record one signal per cell, annotated the same way everywhere (#272)
• ⚠️ bsb-core: a device records every cell it watched (#290)
• ⚠️ bsb-core: recordings say what they recorded, and read_results traces it back to the network (#293, #288)
• ⚠️ bsb-nest: seed the kernel from the network's randomness (#275)

🩹 Fixes

• ⚠️ Let writing data to file at flush (#213)
• Python 3.13 and 3.14 compatibility, arbor 0.12 (#249)
• code dependency config for dotted paths. (#255)
• ⚠️ seed postprocessing hooks and Distribution draws; require Python 3.11 (#291)
• ⚠️ bsb-arbor: draw the poisson generators from the network's randomness (#280)
• bsb-core: parse config package requirements into a slotted subclass (#268)
• bsb-core: attribute configuration errors to the node they happened on (#273, #67, #45, #59)
• bsb-core: stop re-parsing pyproject.toml per report, and survive an offline fetch (#269)
• ⚠️ bsb-core: store structured metadata so it survives being written (#262)
• ⚠️ bsb-core: draw device targets the same way on every rank (#265)
• ⚠️ bsb-core: the soma is every location labelled soma (#295)
• ⚠️ bsb-core: multiplex per-device recordings instead of one Neo object per cell (#297)
• bsb-hdf5: count placement sets from the tracked chunk stats (#270, #26, #28)
• bsb-nest: install NEST without a synced environment (#279)
• ⚠️ bsb-neuron: default synapse delay to 1 ms, and drop the unrunnable Convergence strategy (#271, #99, #136)
• release: bump sibling pins on release and stop skipping the release run (#259, #193)

⚠️ Breaking Changes

• bsb-core: multiplex per-device recordings instead of one Neo object per cell (#297)
• seed postprocessing hooks and Distribution draws; require Python 3.11 (#291)
  BSB requires Python 3.11 or newer. Every package's
  requires-python moves from >=3.10,<4 to >=3.11,<4.
• bsb-core: the soma is every location labelled soma (#295)
  SomaTargetting targets the locations labelled soma instead of
  location (0, 0). A cell whose morphology labels no soma has no soma locations, so
  devices that default to it (voltage recorder, current and voltage clamp, spike
  generator) target nothing on such a cell.
• bsb-core: recordings say what they recorded, and read_results traces it back to the network (#293, #288)
  the cell_type annotation of recordings and
  Recording.cell_type are renamed to cell_model. On NEST and arbor,
  cell_id is the cell's placement set id instead of the simulator's node id
  or gid.
• bsb-core: a device records every cell it watched (#290)
  recordings no longer carry gids or pop_size. A device's
  recordings are the cells it watched, so their number is the population and their
  cell_ids are which cells.
  Claude-Session: https://claude.ai/code/session_017Zcwom5VRiA5WMreKncmVg
• bsb-core: record one signal per cell, annotated the same way everywhere (#272)
• bsb-core: a run ends with one results file, whatever it took to make it (#266)
  bsb simulate writes one file per run rather than one per rank,
  and SimulationResult gains finalize, which an adapter driving the result
  itself has to call.
  Claude-Session: https://claude.ai/code/session_01BHScuq1Hoemb5AUeofueWP
• bsb-nest: seed the kernel from the network's randomness (#275)
  seed is no longer configured on a NEST simulation. Pin the
  network's rng.seed, or declare an rng.settings entry of strategy nest and
  name it on the simulation's rng.
  Claude-Session: https://claude.ai/code/session_017Zcwom5VRiA5WMreKncmVg
• bsb-core: draw connectivity from the configured randomness (#277)
• bsb-core: draw placement from the configured randomness (#276)
• bsb-arbor: draw the poisson generators from the network's randomness (#280)
• bsb-core: draw device targets the same way on every rank (#265)
  a device targetting a fraction or count of a population
  selects a different subset than before, since the selection is now seeded.
• bsb-core: parameters answer what a model's values are, per arity (#260)
  Parameter no longer has a value attribute and
  ParameterValue is removed. CellModel.parameters is a dict keyed by the model
  parameter, not a list. constants blocks no longer accept a computed parameter;
  write it in parameters. resolution is configured on any simulation, not only
  per backend.
• bsb-core: store structured metadata so it survives being written (#262)
  the config annotation on a result block is a JSON string rather
  than a dict, in memory as well as on file. Keeping it a dict in memory and a
  string on file would reintroduce exactly the trap that caused this. Read it with
  read_simulation_config.
• bsb-core: drop the adapter's post_prepare argument (#278)
• bsb-core: give the prepared simulation a hook of its own (#263)
  post_prepare on a simulation is replaced by after_prepare,
  whose entries are named dynamic nodes with a strategy rather than a list of
  dotted function paths.
  Claude-Session: https://claude.ai/code/session_01BHScuq1Hoemb5AUeofueWP
• bsb-neuron: default synapse delay to 1 ms, and drop the unrunnable Convergence strategy (#271, #99, #136)
  bsb.connectivity.Convergence and the bsb.Convergence
  public API entry are removed. Configurations using
  strategy: bsb.connectivity.Convergence did not work; use
  bsb.connectivity.FixedIndegree or a custom strategy instead.
  Closes #99
• Let writing data to file at flush (#213)

❤️ Thank You

• Claude Opus 4.6
• Claude Opus 4.8 (1M context)
• Claude Opus 5
• Claude Opus 5 (1M context)
• Claude Sonnet 4.6
• Claude Sonnet 5
• Dimitri Rodarie
• Dimitri RODARIE
• Filippo @filimarc
• Robin De Schepper