v0.6.5 (September 7th, 2026)
----------------------------

API changes
###########

* Removed the wrapper that still allowed initialising the `OffsetConverter`
  using a tuple named `coefficient`. (This breaks compatibility with solph
  API before v0.5.3 from June 25th, 2024.)

New features
############

* Added support for the solver HiGHS. If installed, you can use it by calling
  `Model.solve(solver="highs")`.

Bug fixes
#########

* Fix TSAM result disaggregation for ``NonConvex`` and ``InvestNonConvex``
  flows, which silently dropped per-timestep variables from the output.
* Fix inconsistency in the assignment of sequences using scalar variables;
  e.g. `flow = Flow(variable_costs=5)` worked, but `flow.variable_costs = 5`
  failed when constructing the model.
* Name the TSA-disaggregated storage SOC result ``storage_content`` to match
  non-clustered ``GenericStorage`` results, so result readers (e.g. ``views``)
  find it by the same key regardless of clustering.
* Fix ``limit_active_flow_count`` and ``limit_active_flow_count_by_keyword ``
  constraints for flows of type ``INVEST_NON_CONVEX_FLOWS``
* Fix ``NonConvex`` ``minimum_uptime``/``minimum_downtime`` being skipped for
  the first time steps of the optimisation horizon. The constraints were only
  built for ``t > first_flexible_timestep``, which is derived from
  ``minimum_downtime`` if ``initial_status=0`` and from ``minimum_uptime``
  otherwise. A unit could therefore start up (or shut down) in the first time
  step it was free to, and reverse that one time step later, ignoring the
  other of the two minimum times. Both constraints now use ``initial_status``
  as the status preceding the horizon, the same convention the startup and
  shutdown constraints already used. The rules are now also built for the last
  time step, so ``minimum_uptime`` is no longer ignored for a startup there.
  Note that this can change the result of existing models that combine
  ``initial_status`` with ``minimum_uptime`` or ``minimum_downtime``.

Other changes
#############

* Updated respective examples and tutorials to use current TSAM API (v3).
* Internally, sequences of 'None' are now avoided and a simple 'None'
  is used instead. We expect no effect for end users.
* ``_FakeSequence`` can no longer have a defined length.
  They have "every length" by convention. Thus, it can now be reliably cast;
  not to an array but to a scalar float. We expect no effect for end users.
* ``_FakeSequence`` have operators defined. We expect no effect for end users.

Contributors
############

* Jelmer de Wit
* Patrik Schönfeldt
* Pierre Francois Bachibouzouk
* Stephan Günther
* Diana Maldonado Castro
* Jonas Freißmann
* Uwe Krien
* Francesco Witte
* Dylan Pulver