Issues

docs: the fit.toml dt_check sub-table's keys are not in any embedded guide

vsbuffalo/camdl#976 · opened

on Oct 15, 2026

docs: chain-binomial simulates R = β·dt/(1 − e^(−γ·dt)), not β/γ; say so, and name it when the dt check fails

vsbuffalo/camdl#975 · opened

on Oct 15, 2026

starts = { from_posterior } draws rows with replacement, so a short table seeds several chains from one row

vsbuffalo/camdl#974 · opened

on Oct 15, 2026

docs: the origin row of a restarted forecast is not a structural zero for starting_on streams

vsbuffalo/camdl#973 · opened

on Oct 15, 2026

fit summary's per-chain loglik-eval table numbers chains from 0; every other artifact numbers from 1

vsbuffalo/camdl#972 · opened

on Oct 15, 2026

fit summary passes the Â leg for a single IF2 chain (max Â = 0.000 ✓) instead of saying not assessed

vsbuffalo/camdl#971 · opened

on Oct 15, 2026

pfilter --trace leaves the observed column empty on every row ({:.0} applied to a String)

vsbuffalo/camdl#970 · opened

on Oct 15, 2026

E107 fires on a parenthesised unit literal after /, the form its own fix line recommends

vsbuffalo/camdl#969 · opened

on Oct 15, 2026

sinusoidal forcing peaks at phase + period/4, but the language reference says phase is the time to the peak

vsbuffalo/camdl#968 · opened

on Oct 15, 2026

A duration parameter with a negative lower bound gets a log transform under IF2: log(-20) = -690 and a 1e300 random-walk step

vsbuffalo/camdl#967 · opened

on Oct 15, 2026

pfilter runs at dt = 1 unless --dt is given, ignoring the model's simulate { dt } and the --fit [config] dt

vsbuffalo/camdl#966 · opened

on Oct 15, 2026

A typed constant let becomes a parameter that fit run demands in [fixed], and trips E512 when another let uses it

vsbuffalo/camdl#965 · opened

on Oct 15, 2026

simulate --init-state refuses every restart time above 8192: an absolute 1e-12 epsilon is below one ulp of t

vsbuffalo/camdl#964 · opened

on Oct 15, 2026

Test issue: notification check (will be deleted)

vsbuffalo/camdl#963 · opened

on Oct 10, 2026

A scheduled intervention at exactly t_start never fires (particle filter and forward chain-binomial alike, silently)

vsbuffalo/camdl#958 · opened

on Oct 4, 2026

2

Off-grid events: the particle filter fires them at the exact time, forward chain-binomial snaps them to the grid (−0.355 nats)

vsbuffalo/camdl#957 · opened

on Oct 4, 2026

1

PGAS: report θ-sweeps stalled by a non-finite gradient separately from NUTS divergences

vsbuffalo/camdl#954 · opened

on Oct 3, 2026

1

simulate --design-from silently ignores --scenario

vsbuffalo/camdl#948 · opened

on Sep 28, 2026

Store integrity check ignores the recorded SHA-256 (gh#583 item H)

vsbuffalo/camdl#946 · opened

on Sep 28, 2026

fit summary --params-only (and likely compare) fails on nl-* fits: reads final_params.toml, NLopt writes mle_params.toml

vsbuffalo/camdl#942 · opened

on Sep 27, 2026

nl-sbplx objective swallows structural errors as -inf (gh#475 sibling)

vsbuffalo/camdl#939 · opened

on Sep 27, 2026

obs_sample_nan is process-global, so counter-only sampler tests can miss a regression under parallel test runs

vsbuffalo/camdl#937 · opened

on Sep 27, 2026

θ-dependent failures raised as structural errors abort fits instead of rejecting the θ

vsbuffalo/camdl#934 · opened

on Sep 27, 2026

fit predict replays the free-forward horizon at the model's dt, not the dt the posterior was fitted at

vsbuffalo/camdl#933 · opened

on Sep 24, 2026

An NLopt chain that ends at -inf with status xtol_reached counts as converged

vsbuffalo/camdl#931 · opened

on Sep 23, 2026

Navigation Menu

Repository navigation

Issues

docs: the fit.toml dt_check sub-table's keys are not in any embedded guide

docs: chain-binomial simulates R = β·dt/(1 − e^(−γ·dt)), not β/γ; say so, and name it when the dt check fails

starts = { from_posterior } draws rows with replacement, so a short table seeds several chains from one row

docs: the origin row of a restarted forecast is not a structural zero for starting_on streams

fit summary's per-chain loglik-eval table numbers chains from 0; every other artifact numbers from 1

fit summary passes the Â leg for a single IF2 chain (max Â = 0.000 ✓) instead of saying not assessed

pfilter --trace leaves the observed column empty on every row ({:.0} applied to a String)

E107 fires on a parenthesised unit literal after /, the form its own fix line recommends

sinusoidal forcing peaks at phase + period/4, but the language reference says phase is the time to the peak

A duration parameter with a negative lower bound gets a log transform under IF2: log(-20) = -690 and a 1e300 random-walk step

pfilter runs at dt = 1 unless --dt is given, ignoring the model's simulate { dt } and the --fit [config] dt

A typed constant let becomes a parameter that fit run demands in [fixed], and trips E512 when another let uses it

simulate --init-state refuses every restart time above 8192: an absolute 1e-12 epsilon is below one ulp of t

Test issue: notification check (will be deleted)

A scheduled intervention at exactly t_start never fires (particle filter and forward chain-binomial alike, silently)

Off-grid events: the particle filter fires them at the exact time, forward chain-binomial snaps them to the grid (−0.355 nats)

PGAS: report θ-sweeps stalled by a non-finite gradient separately from NUTS divergences

simulate --design-from silently ignores --scenario

Store integrity check ignores the recorded SHA-256 (gh#583 item H)

fit summary --params-only (and likely compare) fails on nl-* fits: reads final_params.toml, NLopt writes mle_params.toml

nl-sbplx objective swallows structural errors as -inf (gh#475 sibling)

obs_sample_nan is process-global, so counter-only sampler tests can miss a regression under parallel test runs

θ-dependent failures raised as structural errors abort fits instead of rejecting the θ

fit predict replays the free-forward horizon at the model's dt, not the dt the posterior was fitted at

An NLopt chain that ends at -inf with status xtol_reached counts as converged