Skip to main content

Check Report (ferx check --json)

ferx check <model.ferx> [--data <data.csv>] --json emits a structured JSON report describing every validation finding, instead of the human-readable summary. The report is designed to be consumed programmatically — by editor tooling, CI, or a coding agent authoring model files — so findings carry a stable machine-readable code rather than only prose.

This is the same data the Rust API returns from validate_model_file; see also the CLI reference for the human-readable form and exit codes.

Top-level shape

{
  "valid": false,
  "model": "mymodel",
  "data": "data/mydata.csv",
  "diagnostics": [
    {
      "severity": "error",
      "code": "E_MISSING_COVARIATE",
      "message": "Model references covariate(s) not found in data (case-sensitive): WGT. Available covariate columns: (none).",
      "suggestion": "available covariate columns: (none)"
    }
  ]
}
Field Type Notes
valid bool true when there are no error-severity diagnostics (warnings and info rows alone keep it true).
model string Model file stem.
data string Present only when --data was supplied.
diagnostics array Every finding; see below. May be empty.

Diagnostic object

Field Type Notes
severity "error" | "warning" | "info" Only error affects valid and the exit code. info is a fact about the model or data, not a finding — count it apart from the warnings (#2387). An info row keeps the W_ prefix of its code.
code string Stable identifier (see table below).
message string Human-readable description.
block string Optional. Owning model block, e.g. "error_model". Omitted when not attributable to one block.
line integer Optional. 1-based source line — see the caveat below.
suggestion string Optional. Actionable hint.

Optional fields are omitted entirely when absent (not emitted as null).

Error / warning codes

Grouped by what the finding is about. Codes are stable; new ones may be added over time. Treat an unrecognised code as a generic finding of its given severity.

The tables below list every code the engine emits — the ones ferx check reports, the ones an entry point returns as EngineError::code(), and the ones that open a message in a run’s warnings — and the in-code registry in src/diagnostics.rs lists the same set. A unit test fails when either drifts from the codes in src/ (#1051).

Model-file structure

The file could not be read as a ferx model at all, or names a block ferx does not have.

[fit_options] and [error_model] lines are consumed end to end or reported as E_PARSE with the offending line quoted. A [fit_options] line that is not a key = value pair names the repair (`method saem` is not a `key = value` pair — did you mean `method = saem`?); an [error_model] line matching no statement form lists the accepted forms. Before 0.4.0 both were dropped without a diagnostic, and the method case is the sharp one — the fit then ran the default estimator while ferx check reported the file VALID, so a run the author read as SAEM was FOCEI and its objective was not comparable to the one they asked for. See #1390, Fit options and Error model.

Code Severity Meaning
E_PARSE error The model file failed to parse (catch-all for parser errors).
E_MODEL_READ error The model file could not be read: it does not exist, cannot be opened, or is not UTF-8. The message names the path and the operating system’s reason, and no block is set. Before #1752 this was reported as E_PARSE. The check reads the file once, and parses and binds that one read, so a file moved or edited while the check runs does not change its result.
E_FIT_OPTION error A call-time fit-option override handed to validate_model_file_with() could not be applied: the key is not a [fit_options] / [data_selection] key, or its value is one the key does not accept. One diagnostic per pair, no block; like a parse failure it is terminal, so the report carries only these (#2165). A model file’s own [fit_options] refuses the same values at parse time as E_PARSE.
E_MISSING_BLOCK error A required [block] is absent. block names which one. run_model_simulate() (ferx --simulate) returns it on block simulation for a model file with no [simulation] block (#1788).
E_UNKNOWN_BLOCK error A [block] header is not a recognised block name. The message lists every offending header with its line, the valid set this build accepts, and a did-you-mean when there is a near match; suggestion carries the near match on its own. See #1040.
W_DEPRECATED_BLOCK warning A [block] that was ferx syntax and is no longer read — currently [initial_values], whose contents moved inline into [parameters]. The block was already inert, so it is ignored rather than failing the parse: deleting it cannot change the fit. Reported separately from E_UNKNOWN_BLOCK so the message can name the replacement instead of offering a did-you-mean that does not exist. See #2913.
E_BLOCK_INSTANCE_NAME error A [block NAME] instance name is given for a block that takes none (e.g. [fit_options DOSE]), or omitted for one that requires it ([covariate_nn NAME]).
E_BLOCK_FEATURE_DISABLED error A recognised block needs a cargo feature this binary was not built with — [event_model] / [binary_model] need --features survival, [markov_model] needs --features markov. ([covariate_nn] has its own E_NN_FEATURE_DISABLED.)
E_NN_FEATURE_DISABLED error A [covariate_nn] block requires building with --features nn.
W_PARSE warning A warning the parser recorded while building the model that has no dedicated code. ferx check reports it under W_PARSE with the parser’s own message, which names the block involved. fit() reports the same text in FitResult.warnings.
W_DATA_PATH_OVERRIDE warning A data path was passed explicitly (--data, or from R) and the model file’s [data] block names a different file — compared as written, then as canonicalized paths. The explicit path is used.

Parameters and random effects

A theta, eta, kappa or sigma the model uses is not declared, is declared in the wrong order, or has no data to bind to.

A [parameters] line must also be consumed end to end by exactly one declaration form, at both ends: text before or after a complete declaration is reported as E_PARSE with the offending line quoted, and so is a line matching no form at all. The one shape with a code of its own is a scale tag on a block_* declaration, below. Before 0.4.0 all of it was dropped without a diagnostic — including leading text, which could silently turn block_sigma PROP ~ 0.04 into a diagonal sigma.

Code Severity Meaning
E_ETA_NOT_DECLARED error An unresolved identifier reads as a random effect (ETA… / KAPPA…) but no omega / kappa declaration defines it, and the data carries no such column. Usually a deleted omega line whose exp(ETA_…) term was left behind.
W_ETA_NOT_DECLARED warning As E_ETA_NOT_DECLARED, but reported without --data, where ferx cannot yet know whether the dataset supplies the column.
E_SIGMA_ORDER_MISMATCH error A single-endpoint [error_model] names its sigmas in an order other than the [parameters] declaration order. These sigmas are consumed positionally, so the written names must match the declared order. Fix by reordering the sigma declarations (permuting a block_sigma’s lower triangle with them) — reordering the arguments also parses but swaps the proportional and additive roles, i.e. changes the model. See #1001 and Error model.
E_BLOCK_VARIANCE_ONLY error A scale tag ((sd), (variance) or (var)) on a block_omega, block_sigma or block_kappa declaration. The lower triangle mixes variances and covariances, so one tag cannot say which entry is on which scale. The repair depends on the tag, and the report’s suggestion carries whichever applies: after (sd), square each SD into a variance and write the off-diagonals as covariances; after (variance) or (var), delete the tag - the lower triangle was already on that scale, so the tag said nothing and nothing else needs changing. The code is raised only when deleting the tag leaves a declaration that parses, so the repair is always sufficient; a tag alongside other stray text is E_PARSE. Until 0.4.0 the tag was dropped in silence and the file reported VALID, so an SD-coded block fitted from squared-wrong initial estimates. See #1377 and Parameters.
E_IOV_MISSING_OCC error Model declares kappa (IOV) parameters but no occasion labels were found. Set iov_column to read them from a dataset column, or iov_occasion = dose / time(...) to derive them in the model, both in [fit_options]. On --simulate the [simulation] design has no dataset and gives each subject a single dose, so the message names only iov_occasion = time(...), with a breakpoint inside the times grid. It says that iov_occasion = dose derives a single occasion there, and that an iov_column the model sets has no effect (#2005). Not raised for a model with nested levels, whose groups come from its level columns.
W_IOV_OCC_MISSING warning Some rows in the IOV occasion column had missing or unparseable values; those rows were assigned occasion=0.
E_THETA_LEVEL_BINDING error A theta NAME[COL, ...] level block could not be bound to the data’s levels: a level column missing or non-finite, no row that defines a level (no observation rows, or on data with only binary, Markov, time-to-event or count records, no endpoint that reads the block bare through its implicit index and whose records define levels), or a contrast the data cannot carry. Block parameters. The binders return the same code from a running program, where it also covers binding to a fit’s level layout (a level the fit estimated no θ for), which ferx check cannot reach.
E_COVARIATE_STATS_BINDING error A [covariate_model] statistic could not be bound to the data: a levels = auto relation whose data carry one level, or a covariate with no non-missing value. Block covariate_model. The binders return the same code, including for a fit’s statistics that do not match the model.

Priors, weights and unused parameters

A declared parameter that nothing reads, a prior that names no parameter, or a kappa weight or initial estimate the data make unusable.

Code Severity Meaning
E_PRIOR_UNRESOLVED error A declared prior cannot be applied: it names no theta / omega / sigma / kappa in [parameters], names a FIXed parameter, an element of a block_omega / block_sigma / block_kappa, or a name that resolves to several coordinates; it is declared twice; its central value is not finite (or not positive for a log-packed parameter); or a [priors] from_fit was not expanded. The message names the parameter.
W_UNUSED_PARAM warning A user-declared θ, ω or κ is not referenced by any [individual_parameters] or [event_model] expression, or a σ is not referenced in [error_model]. It does not affect predictions and is not meaningfully estimated. Generated θ (NN weights, diffusion) and the iiv_on_ruv η are not checked.
E_KAPPA_WEIGHT_NONPOSITIVE error A kappa NAME ~ value weight = W sample-size weight evaluates to zero, a negative number or a non-finite value at some observation or dose record, at the model’s initial θ. κ is applied as κ/√W, so such a weight divides an individual parameter by zero. The usual cause is a missing or non-numeric cell in the covariate the weight reads. The first offending record is reported.
W_KAPPA_WEIGHT_VARIES_WITHIN_OCCASION warning A weighted kappa’s weight expression evaluates to different values (relative difference above 1e-9) at two observation records of the same occasion, or of the same group of the kappa’s level when nested levels are declared. κ is drawn once per occasion or group, so its effective variance Ω_IOV/W is ambiguous; each record is scaled by its own weight. Reported once per kappa, naming the first subject.
W_ZERO_DV_MAGNITUDE_ONLY warning Observations with DV = 0 are scored by a residual error model whose variance is zero at a zero prediction: proportional, a power(...) magnitude, or a combined model whose additive SD starts at 0. As the prediction approaches 0 the row’s log-variance keeps falling (to about −708 of −2LL per row), so a fit can stop at a wrong optimum that still reports converged. Rows censored under bloq_method = m3, FREM covariate rows and non-Gaussian endpoint rows are not counted. Reported by fit() and ferx check, not by simulate(), where DV = 0 is a template placeholder. Fix: add an additive component, remove the rows (pre-dose or pre-lag zeros), or code below-LLOQ rows as CENS = 1 with DV = LLOQ under bloq_method = m3.
W_ADDITIVE_INIT_SCALE warning A combined error model’s additive SD initial estimate is below 1% of the data scale (the median abs(DV) of the observations that endpoint covers). From such a start a multimodal fit can settle where the additive term collapses and the proportional one inflates. The value is not changed; if the fit converges with the additive variance near zero, retry from about 10% of the median abs(DV).

Covariate statistics and level blocks

A [covariate_model] statistic or a theta NAME[...] level block that has to be bound to the data, and the data it is bound to.

Code Severity Meaning
E_COVSTAT_UNRESOLVED error A [covariate_model] relation still needs statistics derived from the data — center = median, levels = auto, or a form whose default bounds come from the data — and none were bound. Reported with --data, where the binding was attempted.
W_COVSTAT_UNBOUND warning As E_COVSTAT_UNRESOLVED, but reported without a --data file: the model is fine, it just cannot be built until a dataset supplies the statistics.
E_COV_LEVEL_UNKNOWN error The data carry a value of a categorical [covariate_model] covariate that is not one of the relation’s levels, which the generated factor would silently model as the reference level (#1111, #1740). Raised by fit(), ferx check, the simulate*, adaptive and predict* entry points, compute_npde_npd, run_sir and run_covariance.
E_THETA_GATHER_INDEX_MISSING error A theta NAME[...] block gathered by a bare data column, where some observation or dose record of a subject has no value in that column. Every record that reads the block needs one. Reported at most once per block and subject.
E_THETA_GATHER_INDEX_RANGE error A theta NAME[...] block’s index column holds a value on some record that is not finite, not a whole number, or outside 1..=N for the block’s N levels. The index is 1-based — a 0-based column is the usual cause — or theta NAME[N] declares too few levels. Reported at most once per block and subject.
E_THETA_LEVELS_DATA_UNBOUND error A model whose theta NAME[...] level block is bound was handed a population never bound for it, so no record carries an index into the block’s levels — the usual “fit, then predict or simulate new data” without binding the new data. The message names the binder to call (bind_from_fit after a fit, bind_theta_levels on a fresh parse), or “k of n subjects” and the first such ID when only some subjects carry the index (#1647, #1762). Raised by fit() and the predict*, simulate*, adaptive, NPDE and post-hoc entry points.
E_THETA_LEVELS_DATA_MISMATCH error A population that carries a bound level block’s index was bound for other levels than the model’s (#1762): a record’s index is not the position, in the model’s level table, of the label its own level columns spell, or an index arrives without its level column. Unseen labels are listed, or the first misindexed record is named. Refused, never re-indexed. Raised by the same entry points as E_THETA_LEVELS_DATA_UNBOUND.

Residual correlation

A residual correlation line, or a block_sigma correlated residual, against the rows it correlates.

Code Severity Meaning
E_RESIDUAL_CORRELATION_RANGE error The θ behind a residual correlation <kind>(RHO) by … line can leave the residual covariance R indefinite: a cs ρ at or below −1/(n−1) for the largest block of n rows, a negative car1 ρ, or a fixed / simulated value with abs(ρ) ≥ 1. A free ρ must also be bounded within ±0.995, since R becomes numerically singular as abs(ρ) approaches 1; FIX it to hold a value closer to ±1.
E_RESIDUAL_CORRELATION_RESET error Under a residual correlation ar1(…) or car1(…) line, one block holds rows on both sides of a TIME reset (an EVID=3/4 record that restarts TIME), so their order and the lag between them are undefined. Add the occasion column to by so each occasion is its own block, or use cs.
E_RESIDUAL_CORRELATION_TIES error Under a residual correlation ar1(…) or car1(…) line, two rows of one block share a TIME. ar1 orders a block by TIME, so tied rows have no order; car1 correlates them by ρ^0 = 1, which makes R singular. Give them distinct times or use cs.
W_RESIDUAL_CORRELATION_INERT warning The model declares a serial residual correlation line, but in every subject each block formed by its by column(s) holds at most one observation. No two rows are correlated, so ρ does not enter the likelihood and cannot be estimated. Check that the by columns group the rows meant to be correlated.
W_BLOCK_SIGMA_L2_ORDER warning Some subjects with no L2 grouping id have a (time, occasion) group in which one observation shares a block_sigma correlation with two or more others, so the cross covariances are paired one-to-one in CSV row order and reordering rows would change the fit. Add an explicit L2 column giving each correlated unit its own id.
W_L2_UNUSED warning The data carry an L2 column, which the reader always treats as the reserved level-2 grouping id for block_sigma pairing, but the model declares no correlated residual. The column therefore does nothing and is not readable as a covariate; rename it if a covariate was meant.

Profiled level blocks

A level block declared profile (#1065) has its levels solved for at every outer step rather than searched by the optimizer. That needs a linear route to the prediction and the FOCE, FOCEI or Laplace objective, so the check refuses anything else; nothing falls back to ordinary estimation.

Code Severity Meaning
E_PROFILE_SCOPE error A profiled block that cannot be fitted that way: profile on a single θ, or with FIX; a box that is a point, is empty or misses the init; a method other than foce / focei / laplace (at any n_agq); a time-to-event or Markov endpoint (a binary one is in scope, #2111); a residual error that is not additive, or log-transformed; M3 censoring; a [mixture], [diffusion] or FREM model; optimizer = trust_region or mma; global_search; a parameter prior. Block parameters.
E_PROFILE_NOT_LINEAR error A profiled block that a prediction does not read as c × block plus terms that do not read it, with c from the data alone, or as one logit link around such a read that is the whole prediction (y = inv_logit(c × block + …), #2111). The message names the construct (a function, power or remainder, a logit link that is not the whole prediction, an inline if or if branch whose condition reads more than data, a product of two reads, a parameter coefficient, a divisor, the states, the residual error, a second profiled block) and where it is, or says that no prediction reads the block. A [binary_model] predictor is a prediction whose logit link is implicit, and a model with an [event_model] or [markov_model] block is left to E_PROFILE_SCOPE rather than called unread (#2562). Block parameters.

Nested variability levels

A model that declares level lines (nested levels, #1082) is checked for how its levels are declared and whether the data can identify them.

Code Severity Meaning
E_LEVEL_MIXED_WITH_IOV_COLUMN error A level declaration together with iov_column or iov_occasion in [fit_options], an iov_occasion rule other than column in the fit options or SimulateOptions of a model with nested levels (block fit_options), or a kappa with no at LEVEL in a model that declares levels. A level replaces the occasion options, and once levels exist every kappa names its own. Block parameters.
E_LEVEL_MISSING_LABEL error A record (observation, dose, EVID=2 / MDV=1 row or reset) has no numeric value in a level column (a blank cell is missing, not carried forward from the row before; a text label such as NCT01 has to be written as a numeric code). Its group at each level is identified by its labels from the outermost level down, so every record needs all of them. The message names the level, the column, the record and its time. Raised by the fit, ferx check --data and every simulate* entry point. Block parameters.
E_LEVEL_SINGLE_MEMBER error A level has at most one group inside every group of its parent, and one of its kappas enters the model through the same parameters as a random effect of the parent level (the subject’s η for the first level). The two variances cannot be told apart. A level where some parents hold several groups is not refused. Block parameters.
W_LEVEL_SUMMARY info Informational, not a finding, so reported with severity info and not counted as a warning (#2387): one row per level of a model with nested levels, outermost first, e.g. level `ARM`: 7 groups in 3 `STUDY` groups (per `STUDY` group: min 1, median 2, max 4); 1 of 3 `STUDY` groups holds a single `ARM` group; 1 kappa at `ARM` (K_ARM). Counted on the observation records. The parent of the first level is the subject. E_LEVEL_SINGLE_MEMBER refuses a level only when every parent holds a single group and a kappa has a twin; a level where some parents hold one group and others several is identified and fits, and this row is where its count shows. Counted on the observation records, as the refusal counts them; a parent with no observation record is left out. Not reported while a level label is missing (E_LEVEL_MISSING_LABEL). Raised by ferx check --data only, not recorded by fit() (#2035). Block parameters.
E_RESIDUAL_CORRELATION_GROUPING_AMBIGUOUS error A residual correlation … by line on a model with nested levels whose grouping a level and a data column would read two ways: a level name inside a list of columns (by (STUDY, ARM)), a level name that is also a different data column, or the by column of a level nested in another level written under another name. by L naming a level groups by that level’s groups, nested by key; a column groups by its value within ID, which merges two groups that share a label in different parents. The message names the level and says what to write. Raised by the fit, ferx check --data and simulate(). Block error_model.
E_RESIDUAL_CORRELATION_CONFOUNDED_KAPPA error A correlation cs(…) by … line whose blocks group the observation rows exactly as the groups of a level that carries a kappa (a nested level, or the occasions of a one-level model), compared as partitions in every subject, so cs by ARMID beside kappa … at ARM counts. A kappa gives each group’s rows one shared random effect, which is the correlation cs estimates: ρ and the kappa variance are not identified together. ar1 and car1 are not refused, and neither is cs over a grouping no kappa is at. Block error_model.
W_RESIDUAL_CORRELATION_PARTLY_CONFOUNDED_KAPPA warning The same comparison holds in some subjects but not all, e.g. cs by STUDY next to a kappa at ARM where some studies have one arm. The model is identified, from the other subjects, and fits; the message gives the count, since ρ and the kappa variance are separated by those subjects alone. A subject whose partitions agree only trivially (every row its own block, e.g. one observation) counts on neither side, here or for the refusal. Raised by the fit and ferx check --data.

Initial estimates against the optimizer box

Every initial estimate is packed and compared against the box the optimizer will search inside, before the first objective evaluation. All of the codes below are data-independent, so ferx check model.ferx reports them without a --data file. The two that are not skipped when maxiter = 0 are called out where they appear, and so is the one that becomes a warning there; for the rest the exemption is because an evaluation-only run clamps the start like any other, it just does not hold the clamped value through a search.

Against the box the optimizer searches
Code Severity Meaning
E_OMEGA_INIT_AT_RAIL error A free omega / kappa / [mixture] omega(k) variance whose initial value packs onto the optimizer’s -6 lower rail — every variance ≤ 6.1e-6, and ~ 0.0 in particular (a declared zero is regularised to 1e-8, packing to ln(L) = -9.21). The start is clamped onto the rail and the coordinate cannot be estimated from there: it stays collapsed or runs away to the opposite rail, moving the θ estimates with it. Either FIX it (that is the supported way to declare no variability) or start it at ≥ 1e-5. NM-TRAN rejects the same stream with error 76, INITIAL ESTIMATE OF VARIANCE CANNOT BE ZERO UNLESS FIXED. Σ is not included — a free sigma ~ 0.0 (sd) is clamped onto its own -8 rail and recovers the optimum. Only reported when an optimizer will actually search: maxiter = 0 (NONMEM MAXEVAL=0, as used by ferx gam --no-fit) is not reported — it clamps the start like any other run, but produces one objective and stops, so nothing is trapped on the rail; note that it therefore evaluates a free ~ 0.0 at exp(-12) rather than at the declared value, which FIX avoids. saem / imp / impmap / bayes / vi carry their own iteration counts, and gn_hybrid a fixed polish budget, so they are still checked. A model used only with predict() / simulate() is not exempt on the check path: ferx check reports what fit() would refuse, so such a model is reported here while still simulating fine. Add maxiter = 0 to [fit_options] if the file is genuinely simulation-only. The message is chosen per eta, not per matrix: an eta declared inside a block_omega / block_kappa is told its block is near-singular (in a block it is the correlations that drive L_ii to zero, so raising a declared variance is not the fix), while a diagonally declared one is told to write ~ 0.0 FIX. Mixing a block and diagonal omega lines in the same model — the whole Ω then packs as one Cholesky — does not change what the diagonal eta is told. A one-eta block_omega (NAME) = [...] is correlated with nothing, so it gets the declared-zero explanation rather than the near-singular one, spelled in its own form (write `block_omega (NAME) = [0.0] FIX`) (#1394). See #1229 and Parameters.
E_THETA_INIT_OUTSIDE_BOUNDS error A theta whose initial value is strictly outside its own declared range. The optimizer clamps the start into its box, so theta TVCL(0.05, 0.1, 10.0) fits from 0.1 — a factor of two — silently, on every run. NM-TRAN refuses the identical stream before any estimation (error 24). Judged against the declared numbers, not the packed ones: theta TVCL(-5.0, 0.0, 10.0) is reported even though -5 and the declared lower 0 both pack onto the same 1e-10 floor, and the message names where the fit really begins (1e-10, which is neither the declared value nor the declared bound). A start exactly on a bound is not reported: there the clamp is a no-op, so nothing moves. Not an error at maxiter = 0: an evaluation-only run is not refused, it gets W_THETA_INIT_OUTSIDE_BOUNDS below instead. See #1251.
W_INIT_OUTSIDE_BOUNDS warning An initial estimate strictly outside one of ferx’s internal rails rather than a declared bound: the hidden 1e9 θ cap, the Ω ±6 (variance ≤ 1.6e5) / off-diagonal ±10 guards, or the Σ [-8, 5] guard. Clamped there before the first objective evaluation. The Σ rails are on the SD scale, which is how ferx stores σ — a plain sigma X ~ v declares the variance v and the message quotes sqrt(v), so sigma PROP_ERR ~ 1e6 reads an SD of 1.000e3. A warning and not an error because these recover: omega ~ 1e8 on examples/warfarin.ferx reaches the base optimum under all eight optimizer × method arms. The Ω lower rail is E_OMEGA_INIT_AT_RAIL instead, where the fit is genuinely trapped. There is no 1e-10 θ counterpart here: a start below the floor packs equal to its floored bound, so it is W_INIT_NOT_REPRESENTABLE below instead, and a range lying entirely below it is E_INIT_BOUNDS_INVERTED.
When the box is empty
Code Severity Meaning
E_INIT_BOUNDS_INVERTED error A coordinate whose packed box is empty, so no start can be placed in it at all. Only a theta can reach this through the parser — every other segment’s bounds are constants — but the code is not θ-specific, and a non-θ coordinate is reported the same way. Three causes: the declared bounds are swapped (theta TVCL(1.0, 5.0, 2.0)); the declared range lies entirely above ferx’s 1e9 packing cap; or it lies entirely below the 1e-10 floor — theta TVCL(1e-12, 1e-13, 1e-11), an ordinary small parameter, is the last. Reported even at maxiter = 0, unlike every other code in this table: an empty box is not clamped, it aborts the clamp, and an evaluation-only run clamps too. Only the affected coordinate is silenced — every other parameter in the file is still reported in the same pass.
When the packed space cannot represent the declared value

A different question from the three above, and the one they are structurally blind to: pack_params applies its own guards before any box exists, and the box is then built from the already-moved value — so a coordinate reported here is, by construction, perfectly in-box.

Code Severity Meaning
W_INIT_NOT_REPRESENTABLE warning A declared value the packer itself altered, before any box existed (#1307): the 1e-10 log floor on a θ / Ω-diagonal / Σ / [mixture] override, or the Fisher-z ±3 rail on a free block_sigma correlation. The message names the declared value and the value the optimizer actually sees. A separate code from W_INIT_OUTSIDE_BOUNDS because every coordinate it names is, by construction, inside its box — the box is built from the moved value — so a silent W_INIT_OUTSIDE_BOUNDS does not mean the declared values reached the optimizer. Not exempt at maxiter = 0: an evaluation-only run scores the substituted value too. A FIXed correlation is never reported, since #1307 holds it exactly.

Method and optimizer compatibility

The chosen estimation method or optimizer cannot serve this model.

Code Severity Meaning
E_SAEM_NO_RANDOM_EFFECTS error method = saem (anywhere in a method chain) on a model with no random effects (n_eta = 0). SAEM is an EM over the random effects; use foce / focei / laplace.
E_METHOD_NO_RANDOM_EFFECTS error method = imp, impmap or bayes (anywhere in a method chain) on a model with no random effects (n_eta = 0). Each integrates over the random effects, so with none declared the marginal collapses to the observation likelihood; use foce / focei / laplace. See #1007.
W_GN_NO_RANDOM_EFFECTS warning method = gn as the last estimating stage on a model with no random effects (n_eta = 0). Pure Gauss-Newton is start-sensitive there — with no inner EBE loop the BHHH step can collapse far from the optimum. Prefer gn_hybrid or focei. Not reported when a later stage re-optimises the GN result (gn_hybrid, or a method = [gn, focei] chain). See #1006.
W_AUTO_OPTIMIZER_FOLLOWS_GRADIENT warning gradient = fd set while optimizer is left at auto, on a model whose analytic outer gradient is in scope. auto resolves off the gradient’s availability, so the one line moved the optimizer (to bobyqa) as well as the gradient. The message names both the optimizer that ran and the one an unforced arm would use; pin optimizer in both arms of a comparison. See #1381 and gradient = fd moves the optimizer too.
E_IMP_CHAIN error imp is mis-placed in a method chain — repeated, or (with imp_eval_only = true) not the terminal stage. The estimating imp may sit anywhere; an evaluation-only imp must be terminal.
E_OPTIMIZER_IOV error optimizer = trust_region used with an IOV model (n_kappa > 0).
E_NESTED_LEVEL_METHOD_UNSUPPORTED error A model with nested levels (more than one level) fitted with a method other than foce / focei at n_agq = 1: SAEM, IMP, IMPMAP, Bayes, VI, Laplace, FOCEI quadrature or GN anywhere in the chain, or with a [mixture] or FREM. The message names the first such stage. A single level is not affected. Block fit_options.
E_OPTIMIZER_AGQ error optimizer = trust_region used with a quadrature stage (method = laplace, or method = focei with n_agq > 1). The trust region scores the quadrature objective but its gradient is the FOCE/Laplace closed form, so it would converge to the FOCE optimum while reporting quadrature OFVs.
E_SDE_INCOMPATIBLE error An SDE ([diffusion]) model used with an incompatible method (saem, gn, gn_hybrid) or gradient setting.
E_AD_RETIRED error gradient_method = ad was requested. The Enzyme automatic-differentiation path was retired; use auto (the analytic sensitivities, with finite differences outside their scope) or fd. The token still parses so an old model file gets this message rather than a parse error.

Quadrature

A quadrature stage — method = laplace, or method = focei with n_agq > 1 — or an n_agq setting the chain cannot use.

Code Severity Meaning
E_AGQ_NODES_ZERO error n_agq = 0, under any method. Zero nodes must not silently disable FOCEI quadrature or be coerced to one-node Laplace, and a method that ignores n_agq still records it in the fit, where a .fitrx holding 0 does not load (#1770); use n_agq = 1.
E_AGQ_NODES_TOO_LARGE error A quadrature stage — method = laplace, or method = focei with n_agq > 1 — with n_agq above MAX_AGQ_NODES (21), where the Golub–Welsch rule loses accuracy. Re-checked here because a FitOptions built in code bypasses the parser’s limit. A chain with no quadrature stage is refused too: it ignores n_agq but records it in the fit, where a .fitrx holding it does not load (#1770).
E_AGQ_GRID_TOO_LARGE error A quadrature stage — method = laplace, or method = focei with n_agq > 1 — whose tensor grid of n_agq^n_eta nodes per subject exceeds MAX_AGQ_GRID (100,000). Lower n_agq, reduce the random effects, or use saem / imp. Under IOV the grid is n_agq^(n_eta + K·n_kappa), with K the most occasions any subject has, so that check needs the data: fit() and ferx check --data raise it (#1788).
E_AGQ_EVAL_ONLY error agq_eval_only = true turns a laplace stage into a likelihood evaluator, so it needs a laplace stage in the chain and that stage must be the last one. Mid-chain, the reported −2 log L would be at parameters the next stage overwrites.
E_FOCEI_NAGQ_UNSUPPORTED error method = focei with n_agq > 1 (the Gauss-Newton-anchored quadrature) on a model outside the analytic sensitivity scope, e.g. an ODE model without sensitivities or a pure TTE endpoint. Its anchor Hessian comes from the adjoint; method = laplace with n_agq > 1 has no such restriction (#251).
E_VI_NAGQ_UNSUPPORTED error n_agq > 1 set on a method chain that contains vi and no stage that uses a quadrature grid (laplace or focei). VI integrates over its variational posterior, so the option would do nothing. Remove n_agq, or chain method = [vi, laplace].

Endpoints, IOV and VI

A method that cannot score the model’s non-Gaussian endpoint or its inter-occasion variability.

Code Severity Meaning
E_BINARY_GN_UNSUPPORTED error gn or gn_hybrid anywhere in the method chain on a model with a [binary_model] (binary or categorical) endpoint. The Gauss-Newton gradient is built from the Gaussian residuals and ignores the logistic likelihood; use focei, saem or imp.
E_CTMM_GN_UNSUPPORTED error gn or gn_hybrid anywhere in the method chain on a model with a [markov_model] (CTMM) endpoint. The Gauss-Newton gradient ignores the Markov transition likelihood; use focei, saem or imp.
E_CTMM_IOV_UNSUPPORTED error A [markov_model] (CTMM) endpoint on a model with inter-occasion variability (n_kappa > 0), whatever the method. The transition generator is built from the between-subject η only, so κ would be ignored.
E_IMP_MIXTURE_IOV_UNSUPPORTED error imp or impmap anywhere in the method chain on a [mixture] model with inter-occasion variability (n_kappa > 0). IMP and IMPMAP estimate IOV, but not in their mixture E-step; use saem or focei.
E_VI_MODEL_UNSUPPORTED error method = vi (anywhere in a method chain) on a model whose likelihood VI’s data term — the fixed-η Gaussian observation NLL — does not cover: one with a non-Gaussian endpoint (time-to-event, binary, categorical; under the survival feature), or an SDE ([diffusion]) model scored by the EKF. Use laplace / saem for the former, foce / focei for the latter.
W_VI_OMEGA_UNANCHORED warning A method chain containing vi sets both vi_kl = mc and vi_omega_update = adam. The closed-form Ω maximiser is disabled and the gradient replacing it is sampled, which is the configuration Janssen et al. (2024) report Ω fluctuating badly in. Keep vi_omega_update = closed_form, or raise vi_mc_samples substantially.

Mixture, LTBS and IIV on residual error

Model features a method or an error model cannot represent. fit() refused each of these before #1788 gave them a code; ferx check now reports them too.

Code Severity Meaning
E_MIXTURE_METHOD error A [mixture] model with a method its objective is not wired for: method = focei with n_agq > 1, an estimator other than foce / focei / saem / bayes / imp / impmap, or bayes with per-class omega(k) / sigma(k) overrides, since its Ω draw is shared across classes. Checked at every stage of the method chain. Block fit_options.
E_IIV_ON_RUV_SHARED_ETA error The iiv_on_ruv random effect is also used in [individual_parameters]. It must be a dedicated omega. Block error_model.
E_IIV_ON_RUV_METHOD error iiv_on_ruv with a non-interaction stage (foce, or gn without interaction) anywhere in the method chain. Y = IPRED + EPS·EXP(ETA) is an interaction model: use focei, imp, impmap or saem. Block fit_options.
E_LTBS_UNSUPPORTED error LTBS (log_transform = true) with a proportional or (natural-scale) combined error on any endpoint, a covariate-selected error model, or an SDE; or a LogCombined error without LTBS. The parser refuses all of these, so only a CompiledModel built in code reaches it, through fit() or check_model_options(). Block error_model.

Correlated residual error

A block_sigma correlated residual, or a residual correlation line, with a method or feature that cannot carry the correlation.

Code Severity Meaning
E_BLOCK_SIGMA_METHOD_UNSUPPORTED error A block_sigma (correlated residual error) model whose method chain contains a stage outside foce, focei, saem, imp, laplace and vi — that is, gn, gn_hybrid, impmap or bayes. Reported once per chain. Not raised for a residual correlation line, which has its own scope check.
E_BLOCK_SIGMA_CHAIN_UNSUPPORTED error A block_sigma model with at least one free (non-FIX) correlation, whose method chain places a stage that cannot read the estimated correlation (anything but foce, focei, laplace) after a stage that estimates it. That later stage would score the declared correlation instead; FIX the block, put the stage first, or drop it (#847).
E_BLOCK_SIGMA_M3_UNSUPPORTED error A block_sigma (correlated residual error) model fitted with bloq_method = m3; the M3 censored-observation likelihood does not support a correlated residual covariance.
E_BLOCK_SIGMA_FREM_UNSUPPORTED error A block_sigma (correlated residual error) model that is also a FREM model with covariate pseudo-observations; the combination is not supported.
E_BLOCK_SIGMA_IOV_UNSUPPORTED error A block_sigma (correlated residual error) model with inter-occasion variability (n_kappa > 0) fitted by a method other than foce / focei, or with n_agq > 1. FOCE and FOCEI score the full residual covariance over the stacked [η, κ] system; the other estimators’ IOV objectives score each row on its own.
E_BLOCK_SIGMA_SDE_UNSUPPORTED error A block_sigma (correlated residual error) model with a [diffusion] block and a saem stage in the method chain. SAEM’s data term omits the EKF process-noise variance, so its E-step and M-step would optimise different likelihoods; foce / focei support the combination.
E_BLOCK_SIGMA_TTE_UNSUPPORTED error A block_sigma (correlated residual error) model that also declares a non-Gaussian endpoint ([event_model], [binary_model] or [markov_model]), whatever the method. The joint likelihood accumulates the residual covariance diagonally and would silently drop the correlation.
E_RESIDUAL_CORRELATION_SCOPE error A residual correlation line is fitted by foce and focei only. Refused with any other method anywhere in the chain, with n_agq > 1, with bloq_method = m3, with FREM covariate pseudo-observations, with a time-to-event / binary / categorical / Markov endpoint, with a mixture model, or with an SDE ([diffusion]) model. The message names the offending method or feature.

Priors

A prior(...) the chosen method or covariance step would silently drop.

Code Severity Meaning
E_PRIOR_METHOD_UNSUPPORTED error A prior(...) is in force but the last stage that estimates (a trailing evaluation-only imp or laplace does not count) is not one that applies parameter priors — only foce, focei, laplace, gn and gn_hybrid do. The fit would return the unpenalized estimates. Make one of those the last estimating stage (e.g. method = [saem, focei]) or remove the prior.
E_PRIOR_COV_METHOD_UNSUPPORTED error A prior(...) is in force and the covariance step runs with covariance_method = s or rsr. S is a sum of per-subject scores and a prior contributes one population-level score, so s reports the unpenalized standard errors and the rsr sandwich reports ones that are too small on every priored coordinate. Use covariance_method = r (the default) or covariance = false.

Data and dosing

The dataset could not be read, or a dose / observation record is inconsistent with what the model expects.

Records and observations

Code Severity Meaning
E_DATA error The --data file could not be read or parsed.
E_MISSING_COVARIATE error The model references a covariate not present in the data (case-sensitive). On block covariates, a column the [covariates] block declares is missing from the file. On block simulation, run_model_simulate()’s [simulation] design does not state a covariate the model reads (#1788).
E_COVARIATE_NOT_NUMERIC error A column the [covariates] block declares holds a value that is not a number. Block covariates.
E_NONFINITE_DV error A subject carries a non-finite (NaN/Inf) observation. Usually a population read by read_population_for_simulation() — a simulation input — passed to fit().
E_NO_SCORED_OBSERVATIONS error The population has no record the likelihood scores: no Gaussian observation and no endpoint (TTE / binary / categorical / Markov) record on any subject. The data contribute nothing to the objective of such a fit, so no estimate can move off its initial value except under a [priors] or regularization penalty. fit() refuses it with the same message, where it used to return OFV 0 with the parameters at their initial values. Reported once for the population, after the more specific data codes, so fit() still names an unrouted endpoint (E_ENDPOINT_NO_RECORDS) or a non-finite DV first. When [data_selection] removed rows classed as observations, the message gives what it removed and names its clauses as the thing to check; otherwise it says what makes a row scored (EVID = 0, MDV = 0 and a DV). Only ferx check --data reaches it; without a dataset there is nothing to count (#1491).
W_MISSING_DV warning One or more EVID=0 rows had a missing DV (./NA/blank) but were not marked MDV=1; they were skipped rather than scored as DV=0.
W_COVARIATE_MISSING warning A covariate column that exists has no value (every cell ., blank or NA) for some subjects, so it is NaN for them. Reported once per covariate, naming it and up to five of the subjects, and only for a covariate the model reads (including [derived], [output] and a numeric [data_selection] test) — a column nothing reads is not reported, even if declared in [covariates] or listed in covariate_columns (#2599). A relation that reads it is then NaN, which makes the objective non-finite, or — through log, sqrt, min, max or a comparison — a finite value with no further warning. Impute the column, exclude the subjects, or guard the relation with present(COV). See Covariate Columns (#2041).
W_DESIGN_DV warning Simulation reader only: one or more EVID=0 rows had a missing DV and were kept as design points to simulate at, so the dataset contributes more rows than a fit of the same file would.
W_CENS_UNEXPECTED warning An observation row’s CENS cell is a whole number other than -1 (above ULOQ), 0 (quantified) or 1 (below LLOQ), such as 7 or -2. Every engine reads only the flag’s sign, so under bloq_method = m3 a positive flag is scored on the lower tail, like 1, and a negative one on the upper tail, like -1; under bloq_method = drop the row is scored as an ordinary observation. The message quotes the cell as written (200, not the 127 it saturates to) and is reported once per subject for each sign. It is not raised on a simulation design row with no DV, which is never scored. A cell that is not a whole number (1.5, abc, inf) is not this warning but a read error, reported as E_DATA, on a Gaussian observation row that carries a DV and that [data_selection] keeps; on a dose row, a row whose DV is missing, or a row the filter removes, the cell is not read (#1496).
W_FILTER_COLUMN_ABSENT warning A [data_selection] condition names a column the dataset does not have. The comparison never matches, so the condition does not select what was meant — and what that costs depends on the key it was written under: an ignore excludes nothing, an accept excludes every row. The message names the missing column(s) and lists the headers the file does have, so a typo (ignore = Coment) is visible rather than silent. ferx check --data reports it since #1465; before that the check never compiled the clauses, so only a fit could raise it.
W_DATA warning A warning from the data reader whose message does not begin with its own W_ code. ferx check relays each reader warning under the code its message opens with, and uses W_DATA only when there is none.

Dose coverage

Whether the dataset delivers any drug at all. These read the dose records as a set, so they say nothing about an individual row — for that see Dose records below.

Code Severity Meaning
W_AMT_NOT_DOSED warning Record(s) carry AMT != 0 but were not read as dose events, because EVID is neither 1 nor 4; their AMT was ignored. In a dataset with no EVID column a nonzero AMT is enough to infer a dose, so this is a file that has one. Reported with the record and subject counts. It wins over W_NO_DOSES, which is the generic backstop for a dataset carrying no AMT signal at all. See #262.
W_NO_DOSES warning Zero dose events parsed across every subject, although scored observations are present. Silent when there are no scored observations (an all-EVID=2, covariate-only file is not a fit), and silent when any subject carries a non-Gaussian observation — a TTE / survival or discrete endpoint legitimately has no PK dose, so warning there would be a false positive. See #262.
W_ALL_DOSES_ZERO warning An AMT column is present and at least one dose event was parsed, but every parsed amount is exactly zero. No drug enters the system, so the objective is flat and the parameters will not move from their initial values. A warning rather than an error because the column is there and a genuinely dose-free-but-AMT-columned dataset is conceivable. See #753.
E_DOSE_NO_AMT error The dataset has dose records (EVID 1 or 4) but no AMT column, so every dose would have amount zero and the fit would run with no drug in the system. Usually the amount column has another name (e.g. DOSE); rename it to AMT. A read error, not a diagnostic of its own: ferx check --data reports it as E_DATA with this text (#753).

Dose records

Code Severity Meaning
W_ADDL_MISSING_II warning ADDL > 0 on a dose row but II is zero or missing; additional doses were not expanded.
W_COMPARTMENT_FREE_DOSES warning The dataset carries dose records but the [structural_model] is compartment-free ($PRED-equivalent), so no dose is applied and AMT / RATE / SS / CMT on those rows have no effect on the prediction. Reported once, with the dose-event count (after ADDL expansion) and the number of dosed subjects. Every dose-level check that would read the model’s compartments is skipped for this class — a coded RATE=-1 / -2 row is not E_MODELED_*_NO_PARAM (a D{n} / R{n} the model cannot consume) and a CMT=2 row is not E_DOSE_CMT_OUT_OF_RANGE (the placeholder one-compartment topology the parser installs) — so the rows land here instead (#1443).
W_CMT_DEFAULTED warning Dose and observation rows were assigned compartment 1 because the dataset did not say which one — no CMT column, or a cell that is missing (./blank/NA) or not a compartment index. The message gives the row counts (plus the ADDL-expanded dose count when a defaulted row carried ADDL) and, for a present column, the offending spellings. Reported whenever CMT selects something on the model: more than one compartment a dose can reach — more than one [odes] state, or an analytical model whose CMT=2 is a real target (an oral model’s depot-bypassing central bolus, a 2-cpt model’s peripheral) — or an observation-side dispatcher, a per-CMT [scaling] block, a CMT=N: error model, or a per-CMT y[CMT=N] readout on either engine, or a declared endpoint the row routes to (unless the model is endpoint-only and that endpoint is compartment 1, where the fallback is provably right), or a [data_selection] clause comparing CMT. Silent where CMT chooses nothing: one_cpt_iv, a one-state [odes] model, a compartment-free model, the transit / inverse-Gaussian absorption models, and a [data_selection] clause on any other column. Rows a CMT-reading filter clause removed are counted and named separately, since they never reach the dose / observation counts (#1409). Reporting a declared endpoint duplicates E_ENDPOINT_NO_RECORDS on an absent column, deliberately: an endpoint routes rows by CMT, and suppressing the warning hid a silent re-routing between endpoints on a bad cell, where that error cannot fire. NONMEM has no equivalent finding because $MODEL lets it declare DEFDOSE; ferx has no such declaration, so an absent column is a genuine ambiguity rather than a default (#1009).
W_STEADY_STATE_II warning SS=1 doses with missing / non-positive II (treated as non-SS).
W_STEADY_STATE_INFUSION warning SS=1 infusion whose length after bioavailability exceeds II (overlapping pulses; SS skipped) — the run-in’s own test (#1281).
W_STEADY_STATE_ABSOLUTE_TIME warning An SS=1 dose (II > 0) on a model whose [odes] right-hand side reads an absolute clock — TAFD, T/t, or the bare TIME built-in. The run-in standing in for the infinite past expands the dose train on a clock local to each cycle, so an absolute clock has no periodic steady state to converge to: T/TIME return a finite number that matches NONMEM’s own steady-state routine but is 67 % from the same model’s explicit dose train, and TAFD reads NaN whatever coefficient it carries. TAD is not included — it is bounded inside one dosing interval, so the run-in does reproduce its train and is anchored against NONMEM (#1139). Counted per dose that actually reaches the run-in, so an SS infusion the run-in itself skips — one whose length after bioavailability exceeds its own II — or a dose on a compartment outside the state vector is not swept up. Asked of the PK block only, so a TIME-reading joint PK-TTE hazard does not trigger it, and TIME read from [individual_parameters] is out of scope; one shape over-reports, see Steady state with TAFD or TIME.
W_STEADY_STATE_INIT warning The model declares an [initial_conditions] baseline and at least one subject has a steady-state (SS=1) dose. The steady-state closed form already assumes an infinite dose history, and the baseline is superposed on top as a separate decaying transient, which double-counts the starting amount. Remove the SS flag or the baseline unless that superposition is intended.

Dose compartments

A dose whose CMT the model has no compartment for, or cannot infuse into.

Code Severity Meaning
E_DOSE_CMT_OUT_OF_RANGE error A dose record — bolus or infusion, zero amount included — whose CMT exceeds the number of compartments: the [odes] block’s declared states, or the analytical model’s compartments. Such a dose used to be silently dropped (ODE) or panic the event walk (analytical). Reported once per (bolus/infusion, compartment); not raised on a compartment-free model, nor on an analytical model whose population feeds no PK prediction.
E_NEGATIVE_CMT error A negative CMT on anything but an EVID=2 row: NONMEM stops on one on a dose or observation, and ferx refuses one on a reset (#2732). See Compartment on/off.
E_CMT_TOGGLE_TIE error A compartment on/off row between two doses, or a reset and a dose, at one TIME: no engine order can place it (#2732). See Compartment on/off.
E_CMT_TOGGLE_UNSUPPORTED error A compartment off row (EVID=2, CMT=-N) on a model that cannot honour it: not an [odes] model, an SDE, a declared endpoint, or a compartment fed by a built-in absorption input (#2732). See Compartment on/off.
E_CMT_TOGGLE_OUT_OF_RANGE error A CMT=-N off row past the [odes] block’s declared states (#2732). See Compartment on/off.
E_CMT_TOGGLE_SS error An SS=1 dose while a compartment is off (#2732). See Compartment on/off.
E_OBS_CMT_OFF error An observation of a compartment that is off at that record, as NONMEM refuses (#2732). See Compartment on/off.
E_DOSE_CMT_NOT_INFUSABLE error An infusion (RATE > 0 or a modeled rate) with a nonzero AMT that names a compartment the model cannot infuse into — in practice CMT=0, NONMEM’s default dose compartment, which is defined for a bolus but not for a zero-order input. Raised on [odes] models and the event-walk analytical models (not transit / inverse-Gaussian); reported once per compartment, with the subject and time of the first row.

Dose attributes

Bioavailability F, lag time ALAG/LAGTIME, and the modeled infusion duration D{n} / rate R{n} behind a coded RATE — the values the engine applies at a dose event.

Code Severity Meaning
E_DOSE_ATTR_DOUBLE_USE error A dose attribute the engine applies at the dose event is also read by the model, so its value is applied twice. Bioavailability and lag are caught at parse time on both engines (message names [odes], [scaling] or [adaptive_dosing]) — on ODE models via the name (F, LAGTIME/ALAG, F{n}/ALAG{n}/LAGTIME{n}), on analytical models via the pk(..., f=…) / lagtime=… mapping, where the remedy is to drop the mapping rather than rename. D{n}/R{n} are reported only when a coded RATE=-2/-1 dose in the data lands on them. Initial conditions are exempt on both engines — [odes] init(state) = … and [initial_conditions] seed the state with the raw value and never consult a dose attribute, so a seed that reads one applies it exactly once (#1046). See #993 and #1004.
E_NEGATIVE_LAGTIME error A dose lag — lagtime / ALAG{n}, or a per-route fn(..., lag=L) on the compartment the dose lands in — is negative at the initial typical-value point (η = 0), at some dose record’s own snapshot. NONMEM refuses the same run (PK PARAMETER FOR ABSORPTION LAG IS NEGATIVE, PROGRAM TERMINATED BY OBJ), and so do fit() and ferx check. The message names the subject, the compartment and the dose record. Use a positive-link parameterisation or a lower bound of 0 on the lag’s theta. A lag driven negative mid-fit is not rejected; it starts that dose’s input earlier, identically on every engine (#1189).
W_NEGATIVE_LAGTIME warning predict() / simulate() only: the same finding as E_NEGATIVE_LAGTIME, on entry points that run no data errors. The lag is applied as written, so the dose’s input starts before its record time (#1189).
E_DOSE_ATTR_NONFINITE error A dose attribute the engine applies at the dose event — lag time (lagtime / ALAG{n}), bioavailability (F / F{n}), or a modeled infusion duration D{n} / rate R{n} behind a coded RATE — is NaN or infinite at typical values for some subject. Checked at every dose record’s own snapshot: that record’s covariate values and its TIME, which is where the engine reads the attribute (#1235). The message names the subject, the compartment, the attribute and the record. Usually a covariate relationship undefined for that subject: a covariate missing at that record (a missing value is NaN, which passes through arithmetic, ^ and exp), a fractional power of a negative value (ALAG1 = TVLAG*(WT/70)^0.75 with a negative WT), or an exponential model on an unscaled covariate (ALAG1 = TVLAG*exp(WT)). exp and ^ are the two DSL operations with no domain guard (/ by ~0 returns 0, and log/sqrt floor their argument; see Values outside an operation’s domain). Every attribute now has the same consequence: the subject’s predictions come back NaN, which the estimator reads as a diverged solve. For a lag or F that has always been so — they make dose.time + lag, and every break derived from it, non-finite, so the timeline cannot be ordered. A non-finite D{n}/R{n} used to be the quiet exception, and it was the dangerous one: it never reached you as a NaN at all, because an infinite D{n} gives rate = amt / D = 0 (not an infusion at all) and a NaN was clamped to the transient-D ≤ 0 floor, so the dose was served as an instantaneous bolus and the fit returned finite but silently wrong numbers — measured 2.03x high against the correct infusion at one elimination half-time. Since #1284 a non-finite D{n}/R{n} repels the subject like every other attribute, so this check is the fit-init front door rather than the only line of defence. See #1189.

Modeled infusion duration and rate

A coded RATE=-2 (modeled duration D{n}) or RATE=-1 (modeled rate R{n}) dose against the parameter that supplies it.

Code Severity Meaning
E_MODELED_DURATION_NO_PARAM error A dose codes RATE=-2 (modeled infusion duration) into compartment n, but the model declares no D{n} in [individual_parameters], so the duration has no value to read. Reported once per compartment, naming the first such dose. Add D{n} = ... or give the dose an explicit positive RATE. Not raised on a compartment-free model, which applies no dose.
E_MODELED_RATE_NO_PARAM error A dose codes RATE=-1 (modeled infusion rate) into compartment n, but the model declares no R{n} in [individual_parameters], so the rate has no value to read. Reported once per compartment, naming the first such dose. Add R{n} = ... or give the dose an explicit positive RATE. Not raised on a compartment-free model, which applies no dose.
W_MODELED_DURATION_NONPOSITIVE warning For a RATE=-2 dose into compartment n, the modeled duration D{n} evaluated at that dose’s typical-value snapshot (initial θ, η = 0) is at or below 1e-8. It is clamped to that floor, so the dose arrives as a bolus-like spike rather than an infusion and the fit may converge to a wrong optimum. One warning per compartment; use a positive-link parameterisation such as D1 = exp(...).
W_MODELED_RATE_NONPOSITIVE warning For a RATE=-1 dose into compartment n, the modeled rate R{n} evaluated at that dose’s typical-value snapshot (initial θ, η = 0) is at or below 1e-8. It is clamped to that floor, so the dose is delivered as a near-zero trickle over a near-infinite duration and the fit may converge to a wrong optimum. One warning per compartment; use a positive-link parameterisation such as R1 = exp(...).

Absorption input-rate functions

A built-in absorption input-rate function in [odes] (transit(), igd(), weibull(), zero_order(), first_order()) against the doses that feed it. Checked at typical values at every dose record.

Code Severity Meaning
E_ABSORPTION_DOMAIN error A parameter of a built-in absorption input-rate function in [odes] is out of its domain at typical values (η = κ = 0), evaluated at every dose record’s own snapshot so a covariate or TIME-reading expression is caught at any dose: transit mtt ≤ 0 or n < 0, igd mat or cv2 ≤ 0, weibull td or beta ≤ 0, zero_order dur ≤ 0, first_order ka ≤ 0, or a non-finite per-route lag=. The message names the subject and dose record; only the first offender is reported.
E_ABSORPTION_FRACTION error A dose-splitting pathway fraction (FR*fn(...)) on a built-in absorption input-rate term is not in (0, 1], or the fractions on one compartment sum to something other than 1 (tolerance 1e-4), at typical values evaluated at each dose record. The message names the subject and dose record; only the first offender is reported (#388).
E_ABSORPTION_RATE_ZERO_ORDER error A non-SS infusion (RATE > 0) into a compartment fed by a zero_order() absorption function in [odes]. Infusing a zero-order window is a window-with-window convolution ferx does not yet implement; dose a bolus into it or use a density kernel. Not reported when E_ABSORPTION_SS_INFUSION already is.
E_ABSORPTION_SS_INFUSION error A steady-state infusion (SS=1 with RATE > 0) into a compartment fed by a built-in absorption input-rate function in [odes]. Steady state and infusion-into-kernel are each supported alone but not together; rejected whatever the II, so an II ≤ 0 row is not silently served as a single non-SS infusion.
E_ABSORPTION_SS_LAG error An SS=1 dose with II > 0 into a compartment fed by a built-in absorption input-rate function, where that compartment has a lag time — a compartment lagtime / ALAG, or a per-route lag= on a forcing feeding it. The steady-state seed cannot yet route the dose through the absorption kernel over the previous interval. Not reported when E_ABSORPTION_SS_ZERO_ORDER already is.
E_ABSORPTION_SS_ZERO_ORDER error An SS=1 dose with II > 0 into a compartment fed by a zero_order() absorption function in [odes]. A zero-order input is a spanning window, so its steady state needs a window-based equilibration ferx does not yet have; expand the run-in with explicit dose records or use a density kernel.
E_ABSORPTION_DIFFUSION error An [odes] model whose right-hand side uses a built-in absorption input-rate function (transit(), igd(), weibull(), zero_order(), first_order()) also declares a [diffusion] block. The SDE/EKF propagation does not apply the input-rate forcing, so the absorption term would be silently dropped; remove one of the two.

Transit and inverse-Gaussian closed forms

The analytic one_cpt_transit / two_cpt_transit and one_cpt_ig / two_cpt_ig models, whose closed forms serve fewer dose shapes than an [odes] model and have no solution in the flip-flop regime. A model with an ODE twin is rerouted there; one without is refused.

Code Severity Meaning
E_TRANSIT_UNSUPPORTED error A transit absorption model (one_cpt_transit / two_cpt_transit) with a dose or model feature its closed form cannot serve. Always refused: a dose into a compartment other than the depot (CMT ≥ 2; CMT=0 counts as 1), an EVID=3/4 reset, and a steady-state infusion. Refused only when the model has no ODE twin to reroute to: IOV (n_kappa > 0), a TIME-dependent structural parameter, within-subject time-varying covariates, an infusion, or an SS=1 dose — the last also when the depot carries a lag time, twin or not.
E_TRANSIT_FLIP_FLOP error A transit absorption model (one_cpt_transit / two_cpt_transit) with no ODE twin to reroute to whose typical-value (η = 0) parameters fall outside the closed form’s convergence domain on some subject — the flip-flop regime (disposition rate ≥ KTR = (n+1)/mtt) or, for 2-cpt, coincident disposition eigenvalues — where it returns an identically-zero profile. Rewrite the model with an explicit transit() forcing in [odes], or check the MTT / CL starts. simulate_with_uncertainty() also returns it when every parameter draw landed in that regime although the point estimate did not.
W_TRANSIT_FLIP_FLOP warning A transit absorption closed form (one_cpt_transit / two_cpt_transit) with an ODE twin has, at the initial typical values (η = 0) of at least one subject, a disposition rate (CL/V, or the 2-cpt fast rate α) at or above the transit rate KTR = (n+1)/mtt. ferx evaluates the equivalent transit ODE model for such parameters, which is correct but slower. Reported once, naming the first such subject. A twin-less model in the same regime is the error E_TRANSIT_FLIP_FLOP instead.
E_IG_UNSUPPORTED error An inverse-Gaussian absorption model (one_cpt_ig / two_cpt_ig) with a dose or model feature its closed form cannot serve. Always refused: a dose into a compartment other than the depot (CMT ≥ 2; CMT=0 counts as 1), an EVID=3/4 reset, and a steady-state infusion. Refused only when the model has no ODE twin to reroute to: IOV (n_kappa > 0), a TIME-dependent structural parameter, within-subject time-varying covariates, an infusion, or an SS=1 dose — the last also when the depot carries a lag time, twin or not.
E_IG_FLIP_FLOP error An inverse-Gaussian absorption model (one_cpt_ig / two_cpt_ig) with no ODE twin to reroute to whose typical-value (η = 0) parameters fall outside the closed form’s convergence domain on some subject — the flip-flop regime (disposition rate ≥ 1/(2·MAT·CV²)) or, for 2-cpt, coincident disposition eigenvalues — where it returns an identically-zero profile. Rewrite the model with an explicit igd() forcing in [odes], or check the MAT / CV² / CL starts. simulate_with_uncertainty() also returns it when every parameter draw landed in that regime although the point estimate did not.
W_IG_FLIP_FLOP warning An inverse-Gaussian absorption closed form (one_cpt_ig / two_cpt_ig) with an ODE twin has, at the initial typical values (η = 0) of at least one subject, a disposition rate (CL/V, or the 2-cpt fast rate α) at or above the IG abscissa 1/(2·MAT·CV²). ferx evaluates the equivalent igd ODE model for such parameters, which is correct but slower. Reported once, naming the first such subject. A twin-less model in the same regime is the error E_IG_FLIP_FLOP instead.

Endpoint routing

Code Severity Meaning
E_ENDPOINT_UNROUTED error A compartment the model declares as a non-Gaussian endpoint ([event_model], [binary_model], Markov) carries Gaussian observations: the population was read without endpoint routing (read_nonmem_csv() instead of read_population_for()), so the endpoint’s rows would be scored as concentrations and its likelihood would contribute nothing. Raised by fit(), simulate(), predict(), predict_categorical(), run_covariance() and run_sir() (#1199).
E_ENDPOINT_NO_RECORDS error A declared non-Gaussian endpoint has no row routed to it anywhere in the data — typically the CMT column is missing, so every row reads as CMT 1. Raised by the entry points that run the likelihood — fit(), ferx check, run_covariance(), run_sir() — not by simulate() / predict(): a simulation template has no event rows by construction (#1199).
E_ENDPOINT_TV_COVARIATE error A covariate that varies within a subject — on any row, endpoint rows included (#1854) — is read by an endpoint that evaluates it once, at the subject’s baseline: an analytic hazard’s own parameters, any covariate under an ODE-accumulated (drug-driven) hazard, a [markov_model] CTMM transition intensity, or — on a drug-driven (time-inhomogeneous) CTMM — an individual parameter the [odes] solve feeds into the states the intensities read (#1854). A [binary_model] predictor and a DTMM transition probability read each record’s own covariates, so they are refused only for a subject whose records cannot be lined up with its rows — one that also carries time-to-event records (#2562). The block is the endpoint’s (event_model, binary_model or markov_model), and the message names the covariate and the first subject, or a θ level block whose level changes between records by its name. Raised by fit(), ferx check, predict(), predict_survival(), predict_categorical(), simulate(), and the post-fit re-scorers run_covariance() and run_sir() (#1854); ferx check reports it since #1774, so a model fit() refused is no longer valid: true (#741).
E_ENDPOINT_TIME_DEPENDENT_PK error The TIME axis of the same freeze: an ODE-accumulated (drug-driven) hazard integrates the individual parameters at their TIME=0 values, so a [individual_parameters] value that reads TIME and reaches the hazard (through the [odes] right-hand side or an init() seed) would put the PK and hazard halves of the objective on different concentration curves (#1853). Also raised, fail-closed, for a hand-built CompiledModel missing the compiled programs the analysis reads. Block event_model. A function of the model alone, so ferx check reports it without --data (since #1774). Raised by fit(), ferx check, predict(), predict_survival() and simulate(); not by predict_categorical(), which never reads the frozen snapshot. The same code on block markov_model is the freeze on a drug-driven (time-inhomogeneous) CTMM generator, whose [odes] solve runs under the same TIME=0 snapshot (#1918); raised by fit(), ferx check, run_covariance() and run_sir().

Endpoint records

Records of a non-Gaussian endpoint that its likelihood cannot score, each of which used to fold into a silent sentinel objective. fit() refuses them, and ferx check --data reports them, since #1788.

Code Severity Meaning
E_RTTE_RECORDS error A repeated-event (type = rtte) endpoint’s records: interval-censored, a non-finite TIME, an entry time after the record’s TIME, or out of time order; under clock = reset also a right-censored row before a later record, or two events at one TIME. The message names the subject. Block event_model. The reader drops some of these rows on load, so mostly a Population built in code reaches it.
E_BINARY_STATE error A [binary_model] record whose DV is not 0 or 1. The reader accepts any non-negative integer on a discrete compartment, since it cannot tell binary from ordinal. Block binary_model.
E_CTMM_RECORDS error A [markov_model] (CTMM) record whose DV is not one of the declared states, or a subject whose CTMM observation times decrease (the reader sorts doses, not observations). Block markov_model.

Scaling, output and derived columns

An observed compartment is missing a scaling or error-model entry, a declared per-CMT entry matches no observation, or an [output] / [derived] name does not resolve.

Code Severity Meaning
E_ODE_METHOD_LINEAR error ode_method = linear was passed in a caller’s fit options (fit(), run_covariance, run_sir) for a model that cannot take the exact linear path: it has no [odes] block, or an equation is nonlinear in the states, reads model time, or uses an absorption input function, or it is a [diffusion] model. The message names which. (The same option in the model file is a parse error with the same message.)
E_PER_CMT_SCALING error An observed compartment lacks a per-CMT scaling entry.
E_PER_CMT_ERROR_MODEL error An observed compartment lacks a per-CMT [error_model] entry.
E_RUV_MAGNITUDE_NONPOSITIVE error A custom residual-error magnitude in [error_model] — an expression multiplying a sigma, such as a weight = ... modifier — or a power(SIGMA, P) exponent evaluates to zero, a negative number or a non-finite value at some observation, at the model’s initial θ. A zero magnitude collapses that row’s variance onto the floor so it dominates the fit; a non-positive exponent changes the error model’s form. The usual cause is a missing or non-numeric covariate cell. The first offending observation is reported. An adaptive-dosing run raises the same code at a Dv reading whose magnitude is not positive and finite at the decision, which this check does not see (magnitudes and adaptive dosing).
W_PER_CMT_UNMATCHED warning The other direction of the two above: a compartment the model declares an entry for, that no observation is recorded on. One code for all three per-CMT channels — [scaling] obs_scale[CMT=N], [error_model] CMT=N:, and a y[CMT=N] readout on either engine — with the channel and the dead compartments named in the message. A warning rather than an error because the entry is inert rather than wrong: nothing goes NaN, and a scaling block shared across studies may legitimately declare more compartments than one dataset exercises. The usual cause is on the data side, a missing or mis-mapped CMT column keying every observation to compartment 1; on a column-less dataset W_CMT_DEFAULTED also fires, for the defaulting itself, while this code names the entries the defaulting left dead. That reading is offered only when the observed set is consistent with it (every observation on compartment 1), so a model observed on compartment 2 is not sent to audit a column that is present and correct. [data_selection] is named as a candidate cause — rather than suppressing the finding, since a filter on any column can remove exactly the rows an entry needed — but only for the dead compartments the clause actually emptied, which ExclusionSummary::obs_cmts_excluded records; a clause that removed rows on some other compartment is not blamed. Silent on an observation-free population, so ferx check without --data does not report every declared entry as dead (#1405).
E_READOUT_DEPOT_UNSUPPORTED error An analytic [scaling] y = … readout reads the oral depot amount on a subject whose depot dose superposition cannot reconstruct: one with an EVID=3/4 reset (#650), or one dosing a compartment other than the default (#375). Block scaling. Raised by fit(), ferx check and predict(); ferx check reports it since #1774. Reference only central, or use an ODE model.
E_DERIVED_NAME_CONFLICT error A [derived] name clashes with a built-in sdtab column, theta, eta, or individual-parameter name.
W_DERIVED_COVARIATE_SHADOW warning A [derived] name shadows a covariate (allowed but may be confusing).
W_DERIVED_STEP_IGNORED warning step= given for a DV-based integral (ignored; DV integrals always use observation times).
E_OUTPUT_UNKNOWN_COLUMN error A name in [output] is not recognised as a covariate, individual parameter, or derived expression.
W_OUTPUT_DUPLICATE warning A name in [output] is already written to sdtab automatically (e.g. TAFD, TAD, an eta name).
W_OUTPUT_MIXNUM_IS_CLASS warning [output] MIXNUM on a [mixture] model whose data carry a column of that name. The sdtab column is the mixture class — the simulated class where there is one, else the fitted MIXEST — as NONMEM’s $TABLE MIXNUM is, not the data column; rename the column to table it (#2784).

Experimental and degraded paths

The model runs, but on a path that is experimental or that quietly drops a capability.

Code Severity Meaning
W_EXPERIMENTAL_SDE warning The model uses stochastic differential equations ([diffusion]), an experimental feature. The filter is covariance-only — the state mean is never corrected by the data — so see What the filter does not do.
W_EXPERIMENTAL_NN warning The model uses neural-network components ([covariate_nn]), an experimental feature.
W_NN_BARE_OUTPUT_START warning A [covariate_nn] block declares no init, so every output starts at the output activation’s value at 0 (0.693 for softplus) whatever parameter it feeds. From that start a fit can report convergence in a far worse optimum. See Start the network at plausible values.
W_NN_UNSCALED_INPUT warning A [covariate_nn] input whose data, after the block’s center/scale, reach past where that input alone saturates a hidden unit at the first layer’s initial (Glorot) weight scale: 3/a for tanh, 6/a for sigmoid, a = √(6/(fan_in+fan_out)). Saturated units start the fit with no gradient. See Normalize your inputs.
W_SDE_RESET warning EVID=3/4 resets under an SDE [diffusion] model are not honoured.
W_ABSORPTION_TWIN_DECLINED warning An analytic transit / inverse-Gaussian model’s ODE twin could not be built, so the model stays closed-form with no ODE fallback. Subjects needing the fallback (time-varying covariates, a TIME-dependent parameter, IOV, SS or infusion doses, the flip-flop regime) are then rejected with an explicit error. The message carries the twin parser’s own reason — most often an individual parameter named after a twin state (CENTRAL, PERIPH). See #1008.
W_SDE_LAGTIME warning An absorption lag time on an SDE ([diffusion]) model is not honoured by the filter.
W_NN_WEIGHT_DIRECT_REFERENCE warning An expression references a generated [covariate_nn] weight or bias θ directly rather than through the network’s declared outputs. The analytic NN θ-gradient assumes the weights reach the likelihood only through the network output, so it is disabled for the model and every θ uses finite differences (slower, still correct). ferx check currently shows this under W_PARSE.

Warnings a run records

ferx check never reports these: they describe what happened while a run estimated or predicted, so they exist only as the code that opens a message in FitResult.warnings (or, where noted, in the warnings a predict* / simulate* call returns).

Estimation

Code Severity Meaning
W_STALLED_AT_INIT warning The estimates never left the initial values. The test is the optimizer’s own escape test when the result carries one (no packed coordinate moved by 1% of its scaled range), otherwise a comparison of the free θ, Ω and σ against their initial values at 1% relative. The reported objective is the objective of the initial values, even if converged is true. Suppressed for an evaluation-only run (outer_maxiter = 0, or no estimating stage).
W_NO_FEASIBLE_EVAL warning The outer optimizer finished without one feasible objective evaluation: every trial was rejected by the EBE convergence guard (max_unconverged_frac) or returned a non-finite objective. Nothing was minimised, the reported estimates are the ones the fit started from, and the fit is converged: false. The message names the cause and the remedy (inner_maxiter / max_unconverged_frac, or the record that poisons the objective). Not raised for a cancelled run.
W_ABOVE_START warning The fit reported an objective materially above (the plateau check’s relative \(10^{-3}\) bound) the one its starting estimates scored — after any restart, which is measured against the fit’s start, not its own, so the fit is converged: false. The message gives both objectives and, when the EBE convergence guard rejected the starting evaluation (so it was never a candidate for the reported estimates), how many subjects’ inner loops did not converge there against max_unconverged_frac; and, when the starting evaluation failed outright (a profiled fit at a small inner_maxiter), that the start was re-scored once and at which inner_maxiter. See When the fit ends above its start.
W_NONFINITE_MARGINAL warning The importance-sampling marginal −2 log L at the final estimates is NaN, infinite, or past the divergence cutoffs, so the fit is demoted to converged: false. Covers an imp_eval_only readout as well as an estimating IMP / IMPMAP stage.
W_NONFINITE_OBJECTIVE warning The objective at the final estimates is NaN, infinite, or at or above the 1e14 divergence cutoff, so the fit is demoted to converged: false whatever the optimizer’s stop rule said; the message names which. Exempt: method = vi with vi_final_ofv = none, which publishes no objective by design.
W_NON_STATIONARY_OPTIMUM warning The covariance step’s Newton decrement at a FOCE / FOCEI / Laplace / Gauss–Newton estimate exceeds 1 OFV unit — one standard error — so the estimate is not a stationary point of its objective and the fit is demoted to converged: false whatever the optimizer’s stop said. The message names the coordinates with the largest Newton steps, in standard errors (#2785, details).
W_DEGENERATE_CL_V warning The final estimates put a non-positive or non-finite CL or V on at least one dose, observation, EVID=2 or EVID=3/4 record of a 1-/2-/3-cpt IV / oral closed-form model, each record read under its own covariates and occasion. Both closed-form engines serve a non-positive or infinite value a concentration of 0, and a NaN one NaN once a dose has arrived; NONMEM would reject the record. The message names the subjects, the record count and the first offending TIME. See Warnings and #2104.
W_EBE_START_DEPENDENT warning Three triggers. A cold re-solve of the EBEs at the final estimates scores more than 0.05 OFV units worse than the reported objective, or returns a non-finite objective (#833, #2646); the message names the subjects it missed on. A cold re-solve scores more than 0.05 OFV units below the objective the optimizer minimised, and no restart replaced the run; the fit is then reported converged: false (#1565). Or the inner_screen Sobol screen moved some subjects’ EBEs to a lower individual objective than the descent from η = 0 reached; the message names them, the first ten and then a count (#2013).
W_VI_BAD_BASIN warning At the end of a method = vi run, the excess of the sampled data term over its value at the variational means is more than 25 times the Σ dᵢ/2 a posterior-shaped q would give. The ELBO is not a usable bound there, the optimizer has almost certainly settled in a bad basin, and the fit is reported converged: false. Re-run from better starting values, or chain method = [focei, vi].

Mixture

Code Severity Meaning
W_MIXTURE_MERGED_CLASSES warning A [mixture] fit ended with fewer distinct classes than nsub. Two classes are merged when both carry posterior mass and no subject’s data favours one over the other by more than 1 log-likelihood unit, max_i |nll_ia − nll_ib| < 1, compared on the class objectives themselves, so a subject whose posteriors of both classes underflow to 0 still counts; the test is on the objective, so it also catches classes equal but written two different ways. A class is empty when no subject’s posterior probability of it reaches 1e-2. The message names each merged pair and empty class. Such a fit is often a local optimum: refit with the class starts further apart, or with optimizer = bobyqa, and compare the objective. See Mixture models and #2907.

IMP and IMPMAP

Code Severity Meaning
W_MCEM_OBJECTIVE_ROSE warning An IMP / IMPMAP final marginal sits more than 10% of the objective’s scale above the lowest value its iterations reached, so the fit is demoted to converged: false.
W_MCEM_MSTEP_FROZEN warning A θ or σ owned by the IMP / IMPMAP weighted M-step never left its starting value over the whole run although a step along it still lowers −2 log L by more than 3.84, so that value is not an estimate and the fit is demoted to converged: false.
W_MCEM_MSTEP_FAILED warning NLopt failed at least one IMP / IMPMAP weighted M-step, for a reason other than roundoff, so the fit is demoted to converged: false.
W_MCEM_NOT_SETTLED warning An IMP / IMPMAP run stopped at its iteration limit with its objective’s floor still falling, or well above the final marginal, or with too few finite iterations to read, so the fit is demoted to converged: false.
W_MCEM_SETTLING_UNTESTED warning An IMP / IMPMAP run of fewer than 30 iterations was too short to test whether the objective settled. The fit keeps its verdict, which then only says the run did not diverge.

Covariance step

Code Severity Meaning
W_COV_ANALYTIC_SALVAGE warning The covariance step assembled the exact analytic R matrix for the subjects inside the analytic scope and finite-differenced the information terms of the subjects outside it, rather than moving the whole population onto the finite-difference stencil (#1514). The message names those subjects.
W_COV_QUADRATURE_OBJECTIVE warning A saem / imp / impmap estimate on a model with a non-Gaussian endpoint had its covariance R and S matrices taken on the 7-node adaptive Gauss–Hermite marginal (method = laplace, n_agq = 7) rather than on the FOCEI objective, whose minimum can sit far from that estimate (#1781).
W_COV_QUADRATURE_DECLINED warning The same kind of fit, where the quadrature marginal could not be used — a mixture, inter-occasion variability, or more than three random effects — so the standard errors are the FOCEI objective’s curvature at a point that objective need not minimise (#1781).
W_COV_ODE_TOLERANCE_NOT_TIGHTENED warning An integrated model’s covariance step, which finite-differences its R stencil or score matrix S, could not be run at ode_reltol = 1e-6 / ode_abstol = 1e-8 and ran at the fit’s looser tolerance, so the standard errors may be inflated by integration noise (#520). A step that then fails reports this in its Covariance step failed message instead, without the code.
W_COV_NN_PENALTY warning A regularized [covariate_nn] fit (nn_l2 / nn_smooth): under covariance_method = r / rsr the standard errors include the penalty’s curvature (penalized-ML SEs, Info); under s they cannot represent it (Warning) (#1352).
W_COV_RSR_ONE_SUBJECT_LEVELS warning A successful covariance_method = rsr step on a model with a θ level block, profiled or not, where some estimated levels are read by a single subject (an MBMA block over [STUDY, TIME] with one subject per study). A FIX block and a contrast = ref reference level are constants with no SE, and are not counted. At the estimate such a level’s own score is zero or nearly so, so its rsr SE is not usable: on mbma_placebo they are 1.5–4.8× below r’s. The message names the block, the count and an example level, and says to read those SEs from r. Raised by fit() and run_covariance, which replaces it with the step it reruns.
W_PROFILE_FD_GRADIENT warning A fit with a profiled θ level block, on the analytic gradient route, where the adjoint could not give some subjects the exact gradient of their terms over the profiled levels. Those subjects’ level gradients used reconverged central differences, re-solving their EBEs at every point: the levels solved are the same minimum, only the solve is slower. Raised once for the fit’s profile solves (during this fit) and once for the covariance step’s (during the covariance step); the message counts the subjects and names the first. In warnings_structured it is gradient_fallback at info severity, a route note rather than a problem with the fit (#2405).
W_COV_PROFILED_LEVELS_RESOLVED warning run_covariance on a profiled fit whose level estimates are not the profile solve’s at the other estimates, because they were edited after the fit. The step re-profiles the levels before it runs, and computes its matrix where that re-solve put them; the reported estimates and their SEs are then those re-solved levels, so a level moved onto its declared bound reports the bound as its estimate next to at_bound: true. The message counts the levels that moved, names the largest move, and says the objective and the EBEs are still the incoming fit’s. Raised by run_covariance only, which replaces it with the step it reruns.
W_COV_NON_STATIONARY_ESTIMATE warning The covariance R matrix needed an eigenvalue floor (indefinite, or an eigenvalue under max(1e-10·λ_max, 1e-12)) at a saem / imp / impmap / vi estimate, which those methods never minimise the R objective at. A requested r / rsr covariance was replaced by S⁻¹, or the step failed when S was unavailable too (#1781). Also emitted when R was positive definite and returned, but the gradient’s Newton decrement at the estimate exceeds 1 OFV unit, so the estimate is not a stationary point of R’s objective (#2541, details).

ODE solver

Code Severity Meaning
W_ODE_SOLVER_DIAGNOSTICS warning The ODE solver statistics of that pass (post-fit predictions at the final estimates, or the predict/simulate call at the supplied parameters) show an unclean integration. That means steps clamped at the minimum step size that were kept, ode_method = auto stiff escalations that were rejected or whose fallback failed, unfinished segments kept, stiff-aborted or diverged segments, or a timeline abandoned as non-finite. The message carries the counters and names the pass.
W_FD_JACOBIAN_NOT_TIGHTENED warning Some subjects’ finite-difference Jacobian at the final estimates could not be integrated cleanly at no looser than ode_reltol = 1e-8 / ode_abstol = 1e-10 with 32 times ode_max_steps, and was taken at the fit’s own ODE settings, so the OFV may be off. Raise ode_max_steps or ode_stiff_abort_after, or name a stiff ode_method (#2701).
W_ODE_SOLVER_ESCALATION_NOTE warning Informational (severity Info in warnings_structured). ode_method = auto escalated one or more integration segments to a stiff stepper, and nothing else went wrong: no escalation was rejected, no step clamped at the minimum step size, no segment was unfinished. Set ode_method explicitly to pin either stepper.
W_ODE_RHS_INTERPRETED info Informational, not counted as a warning (severity Info in warnings_structured, category ode_rhs_route). The [odes] right-hand side runs on the bytecode interpreter, because a construct in it is outside what the RHS compiler translates. The message names the construct (an if (…) { … } statement, tanh, …) and the statement it is in; an inline if compiles (#2680), and so does sqrt (#2844). Raised by ferx check (block odes) and once per fit() (#2681, details).
W_ODE_RHS_NOT_NATIVE info Informational, as above. The [odes] right-hand side compiled but runs as register code rather than native code: FERX_NO_RHS_JIT=1 is set, Cranelift cannot target the host, or the native compiler did not build the program. The message names which. Raised by ferx check and once per fit() (#2681).

Compartment reads in [derived]

Code Severity Meaning
W_DERIVED_CMT_IOV_UNSUPPORTED warning A [derived] expression reads compartments[i], and at least one subject of the fit carries IOV (kappa) parameters. The IOV prediction path computes no compartment states, so the expression is NaN for those subjects.
W_DERIVED_CMT_TV_ANALYTICAL warning The model has no [odes] block, a [derived] expression references compartments[i], and at least one subject carries time-varying covariates, or an individual parameter reads TIME (#1855). Compartment states are not reconstructed for those subjects (baseline-parameter states would disagree with the time-varying IPRED), so the expression evaluates to NaN for them.
W_DERIVED_CMT_TV_ODE warning The model has an [odes] block, a [derived] expression references compartments[i], and at least one subject carries time-varying covariates, or an individual parameter reads TIME (#1855). For those subjects the compartment states come from a pass that holds the disposition parameters at the first observation, so they are approximate; IPRED is exact. The message names which of the two causes the fit has. A reset alone, or an [odes] right-hand side that reads TIME, does not trigger it: those states are exact.
W_DERIVED_CMT_RESET_ANALYTICAL warning A [derived] expression reads compartments[i] on an analytical (non-[odes]) model where some subject has an EVID=3/4 reset. Superposition is not valid across a reset, so those subjects’ compartment states and compartment-based integrals are NaN; use an ODE model if states across resets are needed.
W_DERIVED_CMT_ORAL_DEPOT_INFUSION_ANALYTICAL warning A [derived] expression reads compartments[i] on an analytical (non-[odes]) model where some subject has a dose the closed-form superposition cannot place: an infusion into a compartment other than central (including a zero-order input into the oral depot) or a bolus into a compartment other than 1. Predictions stay exact, but those subjects’ compartment states are NaN; use an ODE model for per-compartment amounts.
W_DERIVED_INIT_ANALYTICAL warning The model has no [odes] block, declares an [initial_conditions] baseline, and a [derived] expression references compartments[i]. The superposition state reconstruction does not seed the baseline, so compartment states are left empty and the expression evaluates to NaN; predictions are exact. Use an [odes] model with init(...) if compartment amounts are needed.

Codes only an entry point raises

ferx check never reports these: they refuse an option, a parameter set, a fit or a call that a model file and its data do not contain, or a model fit() accepts but the entry point cannot run. Each is the code() of the EngineError the entry point returns (#1774). They are not the only codes ferx check cannot reach: the θ length, parameter shape, fit-binding and population codes are too, and matching on the code lists every code an entry point returns.

Code Severity Meaning
E_CTMM_UNSUPPORTED error A [markov_model] endpoint reached predict() with no continuous observation to predict, or simulate() or run_model_simulate(): neither has a state-occupancy path yet (#820). fit() supports the model. Block markov_model.
E_TTE_NOT_SIMULATABLE error A TTE endpoint the event-time simulator cannot draw for this population: more than one repeated-event stream on a subject, a repeated-event cause beside a single-event one, an EVID=3/4 reset under a repeated-event or a drug-driven hazard, or left truncation under a drug-driven hazard. Block event_model.
E_SIMULATE_HORIZON error The simulation horizon is missing where a repeated-event or drug-driven TTE endpoint needs one, is not finite and positive, or lies below a subject’s entry time. The message names the option of the entry point that was called: [simulation] horizon / SimulateOptions.horizon, or SimulateUncertaintyOptions.horizon on simulate_with_uncertainty() (#2003). run_model_simulate() returns it for a [simulation] block with a TTE endpoint of any kind and no horizon, the time its template rows are censored at (#1788).
E_PROPENSITY_INPUT error Propensity-score matching was requested on a population it cannot compute posthoc etas from: empty, a subject with no observations, a subject with a non-finite DV, or a [mixture] model.
E_NESTED_LEVEL_SIMULATE_UNSUPPORTED error A model with nested levels (more than one level) reached an adaptive-dosing entry point. Its kappa stream draws one kappa per occasion, and a kappa at an outer level has to be drawn once per group. fit(), simulate() and simulate_with_uncertainty() support the model. Block parameters.
E_THETA_LENGTH error An entry point taking a parameter vector was handed a θ whose length is not the model’s, and refused it rather than read it by position. Raised by fit() on its initial parameters, the predict*, simulate* and adaptive entry points, compute_npde_npd, and run_sir / run_covariance on a fit whose θ does not fit the model they run on (#1614, #1615, #1774). On a level-block model the message says the θ count is set by the bound data.
E_PARAM_SHAPE error An entry point was handed an Ω, σ or ΩIOV whose dimension is not the model’s, an ΩIOV on a model with no κ, or none on a model with κ, and refused it rather than read it by position (#1764, #1774). Raised by fit(), the simulate* and adaptive entry points, simulate_with_uncertainty, compute_npde_npd, fitted_params_from_result (ΩIOV), and run_sir / run_covariance (σ and ΩIOV). fit() also returns it for init_params with no per-class Ω/Σ on a [mixture] model (#1788). Not by predict*, which read none of the three.
E_THETA_LEVELS_UNBOUND error A theta NAME[...] level block was never bound to data, so it has no levels and every value gathered from it is NaN — a CompiledModel assembled in code that skipped the binder. Call bind_theta_levels after reading the population (bind_from_fit with the fit’s data_bindings when predicting with a fit’s θ), use a file entry point that binds for you, or declare the block as theta NAME[N](...) indexed by your own column.
E_FREM_NON_GAUSSIAN_ENDPOINT error prepare_frem() or transform_dataset_for_frem() was handed a model that declares a non-Gaussian endpoint ([event_model], [binary_model] or [markov_model]). The FREM dataset is written from Gaussian rows only, so the endpoint’s records would be dropped; the message names the endpoint CMTs. Run FREM on the PK model without that block.

A missing ΩIOV on simulate() reuses an existing code, since the cause is the same: it is E_PARAM_SHAPE, the parameter-shape code the other entry points return (see matching on the code).

Kaplan–Meier VPC

Code Severity Meaning
E_VPC_TTE_INPUT error An input a TTE VPC (vpc_tte() for a single event, vpc_rtte() for repeated events, vpc_kmmc() for the Kaplan–Meier mean covariate of a single event) would draw a wrong curve or band from: several TTE endpoints and no cmt, a cmt that is not a TTE endpoint, an endpoint of the wrong kind (repeated events for vpc_tte() and vpc_kmmc(), a single event for vpc_rtte()), left truncation, an interval-censored record, a time that is not finite and non-negative, a duplicated subject id, no record on the endpoint, band probabilities not strictly inside (0, 1) and increasing, too few replicates for them, a grid time that is not finite and non-negative, or a stratification covariate a subject lacks or with more than 10 values. For vpc_tte() and vpc_kmmc() also more than one record per subject on the endpoint; for vpc_rtte() also a subject’s records out of time order and max_order = 0; for vpc_kmmc() also an empty covariate, a subject on the endpoint that lacks the covariate or whose baseline value is missing, a level no such subject has, and stratifying on the covariate itself (#1886). See Kaplan–Meier VPC.
W_VPC_RTTE_ENDS_ON_EVENT warning In vpc_rtte(), some subjects’ last record on the repeated-events endpoint is an event, so their follow-up ends at that event. Such an observed subject has an event exactly at its follow-up end, which a simulated subject almost never has. The message gives the count. If follow-up went on after the last event, add a censoring record at its end (#1886).
W_VPC_KMMC_TIME_VARYING_COVARIATE warning In vpc_kmmc(), the covariate changes over time within some subjects. The curve uses each subject’s first value, not its value at the risk time. The message gives the count (#1886). See The curve.

Time-to-event npde

Code Severity Meaning
E_NPDE_TTE_INPUT error An input the TTE npde (compute_tte_npde()) would score wrongly: no TTE endpoint, more than one (competing risks), a repeated-events endpoint, left truncation, an interval-censored record, more than one record per subject on the endpoint, a time that is not finite and non-negative, a duplicated subject id, no record on the endpoint, or nsim = 0. A fit with npde_nsim > 0 reports the same input as the warning W_NPDE_TTE_SKIPPED and still succeeds.
W_NPDE_TTE_SKIPPED warning A fit with npde_nsim > 0 could not score its time-to-event records: a repeated-events endpoint, competing risks, left truncation, an interval-censored record, or a failed reference simulation. The warning quotes the reason. The NPDE / NPD cells of the TTE rows are blank; every other row is scored as usual and the fit succeeds.

File entry points

Refusals of the file entry points (fit_from_files(), prepare_run(), run_model_with_data(), run_model_simulate() and their variants, which ferx runs) and of fit() itself, for inputs ferx check does not read (#1788).

Code Severity Meaning
E_DATA_PATH_MISSING error No data path was given and the model file has no [data] path (no [data] block, or one that only maps columns). ferx check without --data skips its data checks instead of refusing. resolve_data_path() returns the same text as a String.
E_SIMULATION_DESIGN error A [simulation] design that gives the model nothing to observe: no times for a continuous (residual-error) or [binary_model] endpoint, or no times and no TTE endpoint to observe at a horizon. Block simulation. A design that does not state a covariate the model reads is E_MISSING_COVARIATE on block simulation, and one with a TTE endpoint and no horizon is E_SIMULATE_HORIZON.
E_IOV_SINGLE_OCCASION error A derived iov_occasion rule (dose or time(...)) gives every subject at most one occasion, so the per-occasion κ cannot be told from the η. Raised by fit() once it has derived the occasions, by ferx check --data (which derives the model’s rule on the data with the same message), and by run_model_simulate() on its [simulation] template before simulating (#2052), where the message names only iov_occasion = time(...). Block fit_options.

Simulation warnings

Warnings a simulate* call (or compute_npde_npd_diag, which simulates its reference distribution) returns in its warnings, prefixed with the code. The simulation still completes.

Code Severity Meaning
W_IOV_NO_OCCASIONS warning The model declares kappa (IOV) parameters and an Ω_IOV is supplied, but some subjects carry no occasion label once iov_occasion has been applied. Those subjects get no κ draw and are simulated (or, from compute_npde_npd_diag, have their NPDE/NPD reference distribution simulated) at κ = 0. Silent when no Ω_IOV is supplied.
W_FREM_DRAWN_ETA warning A FREM model was simulated with each subject’s η drawn from Ω, covariate η included: by simulate_with_options_diag() without match_method, by simulate_with_uncertainty_diag(), or by an adaptive simulation for the subjects with no supplied η. A FREM model links its covariates to the PK only through Ω, so those subjects’ PK follows the drawn covariate η, not their own covariate values, and their FREMTYPE rows report the drawn covariate value. Silent on a non-FREM model (#2245).
W_SIMULATED_FLIP_FLOP warning On a transit or inverse-Gaussian absorption closed form with no ODE twin, some simulated individuals with a positive dose before their last observation were evaluated at a sampled η inside the flip-flop regime (or the 2-cpt confluent-eigenvalue corner). There the closed form clamps the dose-driven concentration to zero, so their rows carry no contribution from their doses and still enter every summary. The message gives the count out of the eligible individuals; rewrite the model as an explicit ODE to simulate them. The NPDE/NPD reference distribution reports the same count for its own replicates (#1928).
W_UNCERTAINTY_DRAWS_SKIPPED warning Some parameter draws of simulate_with_uncertainty put a transit or inverse-Gaussian closed form with no ODE twin into the flip-flop regime at typical values, where it has no solution, so those draws were not simulated. The message lists them. The remaining draws exclude that tail of the parameter uncertainty, so prediction intervals built from them are biased. If every draw is skipped the call is refused instead.
W_RTTE_DEGENERATE warning While simulating a repeated-event (RTTE) stream, a subject’s expected number of events before the horizon is non-finite or exceeds 1,000,000, so the stream is skipped; or a stream that passed that check produced more than 1,000,000 events, so it is truncated. Either way the subject is censored at the horizon and the rest of the population still simulates. Check the hazard parameters and covariate values.
W_TTE_DEGENERATE_HAZARD warning While simulating a time-to-event endpoint with an analytic hazard, a subject drew a non-positive or non-finite effective hazard rate. No event is generated and the subject is censored at the observation window. Under competing risks it is raised for each analytic cause whose draw degenerated when the subject ends censored. A run of several replicates reports each subject and CMT on one line, in population order, with the number of replicates in which the subject ended censored and their sim indices (the first 10); a run with parameter uncertainty names the affected draws instead. Draws are numbered as in the run’s rows, so a skipped draw keeps its number and the count of simulated draws excludes it (#1648). Check the hazard parameters and covariate values.
W_TTE_SOLVE_FAILED warning While simulating a time-to-event endpoint with an ODE-accumulated (drug-driven) hazard, the event-time solve failed for a subject: an [odes] state or the cumulative hazard became non-finite, the hazard went negative, or the solver ran out of steps. The message names the subject, the CMT and the solver’s reason. A diverged [odes] state is named by its [odes] name, and a cumulative hazard as “the cumulative hazard” (of another cause: “of CMT=k”). No event time is simulated: the subject is censored at the observation window (every cause, under competing risks) and is not simulated past the failure, so its censoring overstates survival. The rest of the population still simulates. The advice names the block to check: the [odes] right-hand side for a diverged state, the [event_model] hazard expression for a negative hazard, and both otherwise, together with the parameter values, including the drawn random effects.

Adaptive simulation warnings

Warnings an adaptive simulation (simulate_adaptive(), simulate_adaptive_from_spec()) returns in its warnings, besides the simulation warnings above that apply to it.

Code Severity Meaning
W_ADAPTIVE_DECISION_BEFORE_MOVED_RESET warning An adaptive simulation has a decision point after an EVID=3/4 reset as written, but before that reset on the engine clock. The data reader moves a reset written at or before the end of a running infusion or ADDL train, or at the TIME of the record before it, to just past that point, with every later row (#1609). Decision times are read on the engine clock, so the dose issued at that point lands before the reset and is removed by it. The warning names the subject, the decision times, the reset’s written and engine times, and the shift to add. Silent for a reset whose TIME goes back (a restarting period) and every reset after it, where a written decision time is ambiguous (#2559).
W_ADAPTIVE_DOSE_ORACLE warning Under dose_knowledge = oracle, some subjects of an adaptive simulation were dosed from their drawn individual parameters or random effects (#944): a perfect-information comparison arm, since no prescriber has these values. The warning counts those subjects and names each dose key and the value it reads; each one’s metrics row has dose_oracle = true. Silent when no subject was dosed from its drawn values.
W_ADAPTIVE_STEP_ABSORBED warning With round_to, the smallest increase N% of an adaptive simulation rounds back to the same dose at a dose some subjects can be at (#2810): the start dose as placed on the grid, or the dose the decrease rules bring it down to. Each time that rule fires there, the dose stays where it is. The warning counts those subjects and names each with its dose (an oracle subject with its replicates), gives the lowest dose the step can raise, and says what to change. A dose at the upper bound is not reported. See Rounding doses to a grid.

Adaptive dosing

simulate_adaptive() and simulate_adaptive_from_spec() refuse most of these before any subject is simulated. The rest come from the driver once it runs, when they depend on the resolved doses, the break timeline or what the controller returned (#1983).

Code Severity Meaning
E_ADAPTIVE_UNSUPPORTED error A model the adaptive-dosing driver cannot simulate faithfully (no [odes] block, an SDE, a covariate-selected [error_model], a residual correlation line, a log-transformed (LTBS) error model — #2129), refused before any subject is simulated. Once the driver runs (#1983): a base dose under a time-varying covariate or IOV that is not a plain bolus or infusion into a plain compartment, a dose lag or infusion duration that is not finite, or an [odes] right-hand side that reads TAD / TAFD before any dose. The block is the one at fault, when there is one.
E_ADAPTIVE_OPTIONS error AdaptiveSimulateOptions with no usable schedule (decision_times empty, not finite, not strictly increasing, or longer than max_decisions), a monitor on a compartment the model does not have, a Dv monitor on a compartment no [error_model] covers, or decision_times / monitors set alongside an [adaptive_dosing] spec that owns them. These are refused before any subject is simulated. Once the driver runs (#1983): two of a subject’s times (decision, observation, EVID=2, reset or base-dose) within 1e-15 of each other without being equal, which the driver cannot tell apart.
E_ADAPTIVE_CONTROLLER_ACTION error The controller returned an action the driver refuses (#1983): a dose into compartment 0 or one past the model’s states, an amount that is negative or not finite, an infusion rate that is not positive and finite, a dose into a compartment fed by a built-in input-rate function or carrying a lag, or a Stop followed by another action. The controller is the caller’s code, so this is a refusal of the input, although only a simulation can reach it. On simulate_adaptive_from_spec() the controller is compiled from the spec’s route, so the same refusal is E_ADAPTIVE_SPEC there. A non-finite amount computed from a diverged ODE state is not this code: the driver refuses a non-finite state, or a monitor that reads a non-finite value, before the controller runs, as an uncoded runtime error naming the states or the monitor (#1578).
E_ADAPTIVE_SPEC error An [adaptive_dosing] spec that does not compile against the model: an invariant the block parser checks (on a spec built in code), an assay_cmt out of range or with no [error_model] to draw the assay from, a route compartment out of range or fed by a built-in input-rate function, or an observe expression that does not compile. On a spec built in code, a dose expression (start_dose, dose_bounds, levels) that reads a state, TIME, an individual parameter, a random effect, a κ or MIXNUM (#944); in a model file these are parse errors. Once the driver runs, a route dose it refuses, such as one into a compartment with a dose lag. Block adaptive_dosing.
Per-subject values

These concern one subject: its dose values, or the η supplied for it.

Code Severity Meaning
E_ADAPTIVE_DOSE_VALUE error An [adaptive_dosing] dose expression whose value for one subject breaks a dose invariant (#944): not finite or negative, bounds with low > high, a start dose outside the bounds or not one of the levels, a rung outside the bounds, a ladder that does not increase, no multiple of round_to within the bounds, or a low bound of 0 with round_to under percentage steps. Every subject is checked before any is simulated, except a subject dosed from its drawn values under dose_knowledge = oracle, which is checked in each replicate after its draw. The message names the first subject that fails (and, for an oracle subject, the replicate), the expression, its value and the covariate and individual values it read. Block adaptive_dosing. Raised by simulate_adaptive_from_spec.
E_ADAPTIVE_DOSE_KNOWLEDGE error An [adaptive_dosing] dose expression reads a subject’s individual parameter or random effect, and some subjects’ η are drawn by the simulation, under dose_knowledge = known (the default) (#944). A drawn subject’s individual values are the simulated truth, which no prescriber knows at dosing time. The message names the key and the value read, lists the drawn subjects (never a supplied one), and names the three routes: a data column carrying the known value, supplied η, or dose_knowledge = oracle. Checked before any subject is simulated. Block adaptive_dosing. Raised by simulate_adaptive_from_spec.
E_SUPPLIED_ETAS error Per-subject η supplied to an adaptive run, or to predict_survival_at_etas_diag, that cannot be used as given (#1594, #1885). From SuppliedEtas::from_table or from_fit: IDs and rows that differ in count, a row that is not one finite value per named η, a repeated subject ID, or a fit whose η names are not the model’s. From simulate_adaptive_with_inputs / simulate_adaptive_from_spec_with_inputs, before any subject runs: a [mixture] model (the run draws each subject’s mixture class, so a fit’s η̂ could be simulated under another class), η names that are not the model’s in the model’s order, or a simulated subject with no supplied η under UnmatchedSubjects::Refuse (the default), listing up to ten IDs. From predict_survival_at_etas_diag, after predict_survival_diag’s checks: the same three, except that a subject with no supplied η is always refused. See Supplied individual values and Individual survival curves at the EBEs.

An [adaptive_dosing] observe or dose-expression covariate the data lack reuses E_MISSING_COVARIATE (block adaptive_dosing), the code ferx check gives a model covariate missing from the data.

Post-hoc steps on a fit

run_sir() and run_covariance() take a fit, and refuse one whose inputs they cannot recover or verify (#1774). ferx check has no fit, so it raises none of these. simulate_with_uncertainty() also rebuilds a fit’s parameters, and refuses a [mixture] fit’s override values with the same two codes. It draws from the fit’s covariance matrix (asymptotic method) or its kept SIR resamples, and refuses one the fit does not carry, or an empty pool, as E_FIT_INPUT_MISSING, and one that is not the size of the model’s packed parameters as E_FIT_INPUT_MODEL (#2044).

Code Severity Meaning
E_FIT_INPUT_MISSING error An input the step needs was neither passed nor recorded on the fit: the model file’s path (no model passed), the data file’s path (no population passed), SIR’s covariance matrix, the population of an IOV model passed as a model on a fit that records no reader settings, the occasion rule of an IOV fit saved before ferx recorded one, on a population that carries no occasion labels, or the fitted values of a [mixture] fit’s per-class overrides, on a fit that records neither them (mixture_overrides) nor a packed estimate holding them.
E_FIT_INPUT_CHANGED error The model or data file the step re-reads no longer has the SHA-256 the fit recorded.
E_FIT_INPUT_MODEL error The model the step would run on is not the one the fit was made with: a model passed in bound with other data-derived bindings than the fit’s, a model passed in whose covariate statistics the population does not reproduce (on a fit that records no bindings), a model whose η count is not the width of the fit’s per-subject η̂, a [mixture] fit whose recorded override values are malformed, or which records none and whose packed estimate has another layout or no longer reproduces its reported estimates, or a covariance matrix that is not the size of the model’s packed parameters.
E_SIR_OPTIONS error SIR settings run_sir() cannot run with: sir_resamples above sir_samples, a sir_df that is not a valid Student-t degrees of freedom, or sir_scale = natural on a model with prior(...). A sir_df below 1 is refused as [fit_options] refuses it, and is refused by fit() and ferx check up front whenever sir = true or covariance_fallback = sir (#1770); otherwise fit() with sir = true reports the same text as a SIR failed: warning instead.
E_SIR_NOTHING_TO_SAMPLE error every estimated parameter of the fit is FIX (θ, Ω and σ, and Ω_IOV, block_sigma correlations and [mixture] overrides where the model has them), so run_sir() has nothing to sample. Block parameters. fit() with sir = true reports the same text as a SIR failed: warning instead.
E_ODE_SOLVER_OPTIONS error ODE solver settings on a FitOptions built in code that no solver can run with: an ode_reltol or ode_abstol that is not finite and positive, ode_max_steps = 0 (#2044), or ode_stiff_abort_after = Some(0), which [fit_options] spells as off (#1770). Refused on an analytic model too, since every fit records these settings. Raised by fit(), run_sir() and run_covariance(); run_sir_core() returns the same text as a String. A model file’s [fit_options] never gets this far: the parser refuses the same values as E_PARSE.
E_LOGIT_THETA_OUT_OF_RANGE error A free θ used as inv_logit(logit(θ) + ETA) whose estimate is not strictly inside (0, 1), so it has no logit-scale draw (#2044). Only a declared upper bound above 1 lets the estimate get there; the suggestion is to declare one below 1 and refit. Block parameters. Raised by simulate_with_uncertainty() with the asymptotic method and by run_sir(). Its to_string() ends with the suggestion; run_sir_core() returns that text as a String, and fit() with sir = true reports it as a SIR failed: warning.

The θ count and the Ω, σ and ΩIOV shape of a fit against the model are E_THETA_LENGTH and E_PARAM_SHAPE, the codes every entry point gives a parameter vector of the wrong shape. A population that is not the fit’s is E_POPULATION_MISMATCH, or E_FIT_SUBJECTS_MISMATCH when the fit’s per-subject rows are what is wrong.

When the step re-reads the model or data file, a failure is the one ferx check reports on the same file, with the same code and block: E_MODEL_READ, the parse codes, E_DATA, E_MISSING_COVARIATE, E_COVARIATE_NOT_NUMERIC, and E_PARSE on data_selection for a recorded selection clause that no longer compiles. The message is the read’s own: an E_DATA refusal has no Failed to read data file lead.

The population a step runs on
Code Severity Meaning
E_POPULATION_MISMATCH error A population handed to, or re-read by, a post-hoc step is not the one the fit was given (#1685): another subject list, other covariate columns, or a subject whose records, doses or covariate values differ. Raised by run_sir, run_covariance and PopulationFingerprint::verify; the message names what differs. Pre-empted by E_ENDPOINT_UNROUTED and E_THETA_LEVELS_DATA_UNBOUND, whose fix is a reader or a binder.
E_FIT_SUBJECTS_MISMATCH error The population matches the fit’s fingerprint, but the fit’s per-subject results (fit.subjects), which run_sir / run_covariance read by position, are another count or another order (#1868). Reachable when a front end rebuilds a fit from separate fields; a fit() result carries both from one population. Without a current fingerprint the same mismatch is E_POPULATION_MISMATCH.

Line-number caveat

line is currently block-level: when present it points at the [block] header that owns the finding, not the exact offending token or column. A finding that is not attributable to a single block (for example a missing-covariate error, where the reference may appear in several blocks) omits line. A missing required block omits line too, since the block has no header in the source. Token/column-level spans are a possible future enhancement.