[rsenne]

hey @JoseKling i found a few more things on #92 during review but I don;t want to keep holding it up so I'm doing a stacked PR here with some proposed changes. Sorry this has taken me so long. I think these all make sense but let me know if anything is afoot. I've approved #92 so this can be the next PR after

Bug fixes:

• simulate(::PoissonProcess, tmin, tmax) started its first inter-arrival at 0
  instead of tmin, so simulating on an interval with tmin != 0 produced events
  outside [tmin, tmax) and a positive log-density.
• simulate(::PoissonProcess, ...) and simulate_ogata sampled marks through the
  rng-less sample_mark, so seeded simulations were not reproducible in their marks.
• event_times(h, d) ignored d for univariate histories while event_marks(h, d)
  did not. History(h, d) combines both and passes check_args=false, so
  History(univariate_h, 2) silently returned a history whose times and marks had
  different lengths. Both accessors now agree, and History(h, d) validates d.
• The inner History constructor coerced dims .= nothing when N == 1, which
  throws a convert MethodError for integer dims and mutates the caller's vector
  for Union{Nothing,Int} dims. It now validates instead of coercing.
• The generic multivariate time_change sized its buffers by ndims(h) but looped
  over 1:ndims(pp), giving a BoundsError when the process had more dimensions
  than the history. Mismatched dimensions now throw a DimensionMismatch.

JET:

• fit(::Type{<:Distribution}, ::History) and fit(::Type{<:AbstractPointProcess}, ::History) intersect at Type{Union{}}, so inference put a Distribution in the
  return-type union of fit(PP, h), which is what JET was reporting. Adding a
  Type{Union{}} method removes it, so the ::AbstractPointProcess assertion in
  MonteCarloTest (which turned the inference gap into a possible runtime
  TypeError) is no longer needed.

Docs:

• The multivariate interface docstring listed integrated_ground_intensity(pp, t, h, d)
  and required simulate(pp, h)
• The interface docstrings in abstract_point_process.jl documented the multivariate
  methods as (pp, h, t, d), but the code takes (pp, t, h, d).
• mark_distribution was declared and documented twice, so the first docstring was
  discarded with a Replacing docs warning during precompilation. Merged into one.
• Restored the time_change(h, Λ) docstring, without which @docs time_change in
  api.md rendered the InhomogeneousPoissonProcess method under History.
• Documented the current History constructor forms, including that a plain vector
  of times always builds a univariate history.

[codecov]

Codecov Report

❌ Patch coverage is 93.33333% with 1 line in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
src/mark_distributions.jl	0.00%	1 Missing ⚠️

📢 Thoughts on this report? Let us know!

[JoseKling]

Great, thanks!