drmTMB

# Phase 18 Simulation Skeleton

This folder holds optional simulation infrastructure for Phase 18. It is not a
package dependency and it is not run as part of ordinary examples.

The layout follows the Phase 18 blueprint in
`docs/design/41-phase-18-simulation-programme.md`:

- `R/` contains small reusable helpers that CRAN-safe tests can source.
- `dgp/` will contain data-generating mechanisms.
- `fit/` will contain model-fitting wrappers.
- `run/` will contain resumable cell runners.
- `reports/` will contain rendered simulation reports.
- `results/` is reserved for ignored local or scheduled-run output.

Every simulation cell should have a stable cell id, explicit conditions,
replicate-level seeds, fit diagnostics, interval status, and enough metadata to
resume or audit a partial run.

Phase 18 result artifacts now name their grain. Replicate-level tables use
`artifact_grain = "replicate"` and keep one row per simulation replicate and
parameter. Aggregate tables use `artifact_grain = "aggregate"` and carry
`n_replicate`, bias/RMSE, MCSEs, convergence, Hessian, warning, and elapsed
time summaries. Report templates may draw replicate-error clouds only from
replicate-level tables; aggregate-only inputs get points, bars, and MCSE
ranges. Grid-output writers return an `artifact_manifest` table with one row
per CSV artifact, file existence, and row counts; optional interval artifacts
that are present but empty are counted as zero rows. Manifests from multiple
grid-writer outputs can be bound and summarized by surface before a report
tries to read their tables, and
`run/sim_write_first_wave_artifact_status.R` writes those bound manifest and
status tables as first-wave report-staging artifacts. The matching
`reports/phase18-first-wave-status-report.Rmd` template reads those two files
before any larger simulation report tries to consume surface-specific tables.
`run/sim_write_first_wave_table_bundle.R` then combines selected artifact
tables across grid-writer outputs, adding source surface and artifact columns
as the leading columns so the next report can compare surfaces without
hand-binding CSVs. It also writes
`phase18-first-wave-artifact-grain-status.csv`, which classifies each selected
surface/artifact pair as replicate-ready, aggregate-only, missing, empty,
mixed-grain, or missing-grain before any report draws replicate-error clouds.
The
`reports/phase18-first-wave-summary-report.Rmd` skeleton reads those staged
tables and the grain-status preflight into one reader-facing page. It also
derives a per-surface replicate-cloud gate, so aggregate-only surfaces stay on
points, bars, and MCSE intervals until a replicate-ready artifact exists.
The #255 closeout test covers the current `gaussian_ls_grid`, `meta_v_grid`,
`count_mu_random_effect_grid`, `proportion_fixed_effect_grid`, and
`biv_rho12_grid` first-wave surface names under that gate.
`R/sim_gallery_grain.R` provides the shared future-gallery gate: a report may
draw replicate-error clouds only when required plot columns are present and
the table carries either `artifact_grain = "replicate"` or
`replicate_cloud_gate = "replicate_clouds_allowed"` without a conflicting
grain marker.
Priority columns stay first and displayed rows are capped before any
publication-style figures are added.
`run/sim_render_first_wave_summary_report.R` orchestrates the status writer,
table-bundle writer, and optional HTML summary render in one call.

Slice 292 starts the comprehensive design as a blueprint, not as a full grid.
The scenario map in `docs/design/41-phase-18-simulation-programme.md` decides
which continuous, proportion, count, ordinal, meta-analysis, bivariate,
random-slope, shape, phylogenetic, spatial, `animal()`, and `relmat()` lanes
are admitted, opt-in, design-only, or failure-ledger only before new DGP files
are added here.

Current pilot files:

- `docs/design/47-phase-18-gaussian-location-scale-ademp.md` is the first
  one-page ADEMP sheet for the admitted Gaussian location-scale lane.
- `docs/design/48-phase-18-meta-v-ademp.md` is the one-page ADEMP sheet for
  the admitted Gaussian `meta_V(V = V)` lane.
- `docs/design/49-phase-18-count-mu-random-effect-ademp.md` is the one-page
  ADEMP sheet for the paired Poisson/NB2 `mu` random-effect lane.
- `docs/design/50-phase-18-proportion-fixed-effect-ademp.md` is the one-page
  ADEMP sheet for the fixed-effect `beta()` and `beta_binomial()` lane.
- `docs/design/51-phase-18-ordinal-fixed-effect-ademp.md` is the one-page
  ADEMP sheet for the fixed-effect `cumulative_logit()` ordinal lane.
- `docs/design/52-phase-18-bivariate-rho12-ademp.md` is the one-page ADEMP
  sheet for the bivariate Gaussian residual `rho12` lane.
- `docs/design/53-phase-18-student-shape-ademp.md` is the one-page ADEMP
  sheet for the fixed-effect Student-t shape `nu` lane.
- `docs/design/167-model-selection-aic-bic-simulation-design.md` is the
  article-support ADEMP sheet for AIC/BIC candidate-selection comparisons
  across continuous-tail, count-zero, and scale-formula examples.
- `docs/design/54-phase-18-animal-relmat-known-matrix-ademp.md` is the
  one-page ADEMP sheet for the known-matrix animal/`relmat()` intercept and
  matching q=2 bivariate location-covariance lanes.
- `docs/design/58-phase-18-animal-relmat-q4-ademp.md` is the focused addendum
  for the constant all-four animal/`relmat()` q=4 location-scale smoke lane
  and its derived-correlation interval boundary.
- `docs/design/56-phase-18-spatial-q2-ademp.md` is the one-page ADEMP sheet for
  the constant coordinate-spatial q=2 bivariate location-covariance lane.
- `docs/design/73-phase-18-nbinom2-sigma-random-intercept-ademp.md` is the
  one-page ADEMP sheet for the ordinary NB2 log-`sigma` random-intercept smoke
  lane.
- `docs/design/74-phase-18-nbinom2-phylo-q1-ademp.md` is the one-page ADEMP
  sheet for the overdispersion-aware NB2 phylogenetic q=1 `mu` formal-admission
  lane.
- `docs/design/148-phase6c-random-slope-simulation-plan.md` is the #446
  planning handoff for the admitted Phase 6c random-slope pilots. It names the
  run order, allowed claims, required artifact tables, MCSE targets, execution
  limits, and stop rules before larger power grids are dispatched.
- `phase18_random_slope_registry_preflight()` in
  `run/sim_phase18_structured_workflow_registry.R` prints the random-slope
  registry rows and their gate fields before any pilot is dispatched. It now
  shows the q8 endpoint smoke, recovery, and staged-diagnostic rows as opt-in
  `ready_grid` tasks while keeping their promotion boundary explicit.
- `phase18_biv_gaussian_q8_endpoint_precode_gate()` names the eight q8
  all-endpoint labels, records the 28 implied pairwise correlations, and checks
  that the smoke row is artifact-ready before any recovery, coverage, or power
  claim.
- `docs/design/75-phase-18-nbinom2-phylo-q1-formal-audit.md` records the local
  NB2 q1 all-cell formal sentinel, representative 5-replicate audit, and
  `hold_smoke_only` promotion decision before the 500-replicate gate.
- `docs/design/76-phase-18-nbinom2-phylo-q1-sharded-formal-grid.md` records
  the cancelled single-job formal dispatch, runtime estimate, and shard
  contract for the 500-replicate NB2 q1 formal grid.
- `docs/design/79-supported-nongaussian-evidence-goal.md` records the
  supported non-Gaussian evidence closeout goal, separating fixed-effect family
  evidence, count mixed-model first slices, q=1 phylogenetic formal gates, and
  blocked neighbouring routes.
- `docs/design/109-phase-18-core-family-completion-map-slices-1279-1288.md`
  records the breadth-first Phase 18 order across counts, proportions,
  positive-continuous responses, ordinal responses, and shape families.
- `docs/design/110-phase-18-proportion-fixed-effect-artifacts-slices-1289-1298.md`
  records the DGP, smoke, grid, first-wave, and Actions artifact lane for
  fixed-effect `beta()` and `beta_binomial()` models.
- `docs/design/111-phase-18-positive-continuous-fixed-effect-artifacts-slices-1299-1308.md`
  records the same artifact path for fixed-effect `lognormal()` and
  `Gamma(link = "log")` models.
- `docs/design/112-phase-18-ordinal-fixed-effect-artifacts-slices-1309-1318.md`
  records the same artifact path for fixed-effect `cumulative_logit()` models.
- `docs/design/115-phase-18-zero-one-beta-fixed-effect-artifacts-slices-1339-1348.md`
  records the same artifact path for fixed-effect `zero_one_beta()` models.
- `docs/design/117-phase-18-bounded-response-mu-random-intercept-artifacts-slices-1359-1368.md`
  records the artifact path for ordinary `mu` random intercepts in `beta()`
  and `beta_binomial()` models.
- `docs/design/118-phase-18-positive-continuous-mu-random-intercept-artifacts-slices-1369-1378.md`
  records the artifact path for ordinary `mu` random intercepts in
  `lognormal()` and `Gamma(link = "log")` models.
- `docs/design/119-phase-18-student-mu-random-intercept-artifacts-slices-1379-1388.md`
  records the artifact path for ordinary `mu` random intercepts in
  `student()` models with fixed-effect `sigma` and `nu` formulas.
- `docs/design/134-phase-18-count-structured-q1-artifacts-slices-1721-1728.md`
  records the opt-in artifact path for ordinary Poisson/NB2 q=1
  `spatial()`, `animal()`, and `relmat()` `mu` intercepts, including the
  pre-grid boundary gate that must be checked before larger pilots become
  recovery or coverage evidence.
- `docs/design/135-phase-18-count-structured-q1-next-pilot-slices-1743-1750.md`
  specifies the next 24-cell x 10-replicate diagnostic pilot for that count
  structured q=1 lane, including dispatch inputs, runtime boundaries,
  no-profile interval policy, helper audit, and reporting requirements.
- `docs/design/136-phase-18-count-structured-q1-pilot-audit-slices-1751-1752.md`
  audits that pilot and records the boundary-gate decision
  `hold_diagnostic`, driven by SD-boundary warnings rather than by a Hessian
  rate or unexplained warning-ledger failure.
- `docs/design/137-phase-18-count-structured-q1-followup-condition-sets-slices-1753-1760.md`
  splits the same 24 cells into executable `stable`, `stable_watch`, and
  `boundary_stress` condition sets for the next diagnostic dispatch.
- `docs/design/138-phase-18-count-structured-q1-stable-diagnostic-audit-slices-1761-1762.md`
  audits the first `condition_set = "stable"` Actions run and records the
  `propose_next_pilot` decision for a separate stable formal-pilot design.
- `docs/design/148-phase6c-structured-one-slope-ademp.md` is the #442/#446
  ADEMP sheet for one numeric Gaussian structured `mu` slope in `phylo()`,
  `spatial()`, `animal()`, and `relmat()`, keeping route-specific artifact
  maturity separate from coverage, power, q2/q4 covariance, and multi-slope
  claims.
- `dgp/sim_dgp_gaussian_ls.R` generates Gaussian location-scale data with
  `mu ~ x` and `sigma ~ z`.
- `dgp/sim_dgp_gaussian_mu_random_slope.R` generates Gaussian `mu` data with
  one q=3 grouped random-slope block, `(1 + x1 + x2 | id)`.
- `dgp/sim_dgp_gaussian_sigma_random_slope.R` generates Gaussian
  residual-scale data with one independent `log(sigma)` random slope,
  `(0 + w | id)`.
- `dgp/sim_dgp_spatial_mu_slope.R` generates Gaussian spatial `mu` data with
  independent coordinate-spatial intercept and slope fields,
  `spatial(1 + x | site, coords = coords)`.
- `dgp/sim_dgp_phylo_mu_slope.R` generates Gaussian phylogenetic `mu` data
  with independent phylogenetic intercept and slope fields,
  `phylo(1 + x | species, tree = tree)`.
- `dgp/sim_dgp_animal_mu_slope.R` generates Gaussian animal-model `mu` data
  with independent dense-pedigree intercept and slope fields,
  `animal(1 + x | id, pedigree = pedigree)`.
- `dgp/sim_dgp_relmat_mu_slope.R` generates Gaussian known-matrix `mu` data
  with independent `relmat()` intercept and slope fields,
  `relmat(1 + x | id, Q = Q)`.
- `dgp/sim_dgp_spatial_q2.R` generates bivariate Gaussian spatial `mu1`/`mu2`
  data with matching coordinate-spatial q=2 fields and residual `rho12` kept as
  a separate layer.
- `dgp/sim_dgp_biv_gaussian_mu_slope.R` generates bivariate Gaussian
  `mu1`/`mu2` data with matching ordinary slope-only random-effect blocks,
  `(0 + x | p | id)`, and residual `rho12` kept as a separate layer.
- `run/sim_summary_biv_gaussian_mu_slope_recovery.R` and
  `run/sim_write_biv_gaussian_mu_slope_recovery_grid.R` promote that slope-only
  `mu1`/`mu2` lane from a single-replicate smoke check to a multi-replicate
  recovery lane over its 10 estimands: bias, RMSE, MCSE, and fixed-effect Wald
  coverage, with the two slope SDs and the derived slope-slope correlation kept
  `derived_interval_unavailable`. It dispatches through the opt-in
  `biv_gaussian_mu_slope_recovery` Actions task.
- `dgp/sim_dgp_biv_gaussian_q6_location.R` generates bivariate Gaussian
  `mu1`/`mu2` data with matching ordinary q=6 location random-effect blocks,
  `(1 + x + z | p | id)`, and residual `rho12` kept as a separate layer.
- `run/sim_summary_biv_gaussian_q4_location_recovery.R` and
  `run/sim_write_biv_gaussian_q4_location_recovery_grid.R` promote the q4
  `mu1`/`mu2` location lane (`(1 + x | p | id)` in both location formulas) from
  a single-replicate smoke check to a multi-replicate recovery lane. They reuse
  the smoke DGP, fit, and runner, run at recovery-scale `n_rep`, and add bias,
  RMSE, MCSE, and Wald interval coverage. Wald coverage is reported only for the
  fixed `mu1`/`mu2` endpoints; the four location SDs and six derived
  location-location correlations stay `derived_interval_unavailable`. The lane
  dispatches through the opt-in `biv_gaussian_q4_location_recovery` Actions task;
  its evidence design follows the bivariate-slope ADEMP
  `docs/design/145-phase6c-bivariate-slope-ademp.md` and the recovery-lane
  pattern in `docs/design/156-phase-18-bivariate-scale-q2-recovery-ademp.md`.
- `R/sim_structured_re_ademp.R` and
  `run/sim_write_structured_re_ademp_scaffold.R` stage the structured
  random-effect q1/q2/q4 ADEMP scaffold for `phylo()`, coordinate `spatial()`,
  `animal()`, and `relmat()` cells. This is a registry, seed, MCSE-policy, and
  denominator-accounting scaffold only: it does not run a calibrated grid, it
  does not promote R-via-Julia bridge support, and it does not make a coverage
  claim.
- `run/sim_summary_biv_gaussian_q6_location_recovery.R` and
  `run/sim_write_biv_gaussian_q6_location_recovery_grid.R` give the q6
  `mu1`/`mu2` location lane (`(1 + x + z | p | id)` in both location formulas)
  the same recovery treatment over its 30 estimands: bias, RMSE, MCSE, and
  fixed-effect Wald coverage, with the six location SDs and fifteen derived
  location-location correlations kept `derived_interval_unavailable`. It
  dispatches through the opt-in `biv_gaussian_q6_location_recovery` Actions task.
- `dgp/sim_dgp_biv_gaussian_q2_scale.R` generates bivariate Gaussian data with
  a matching residual-scale random-intercept covariance block,
  `sigma1 = ~ 1 + (1 | p | id)` and `sigma2 = ~ 1 + (1 | p | id)`, so the two
  log-`sigma` intercepts share a q=2 scale-scale block while residual `rho12`
  stays a separate layer. This is the fittable scale-covariance prerequisite
  for the q8 endpoint gate.
- `run/sim_summary_biv_gaussian_q2_scale_recovery.R` and
  `run/sim_write_biv_gaussian_q2_scale_recovery_grid.R` promote that same
  q2 scale-intercept lane from a single-replicate smoke check to a
  multi-replicate recovery lane. They reuse the smoke DGP, fit, and runner, run
  at recovery-scale `n_rep`, and add bias, RMSE, MCSE, and Wald interval
  coverage. Wald coverage is reported only for the fixed `mu1`/`mu2` endpoints;
  the random-effect SD and derived scale-scale correlation rows stay
  `derived_interval_unavailable`. The lane dispatches through the opt-in
  `biv_gaussian_q2_scale_recovery` Actions task; its design sheet is
  `docs/design/156-phase-18-bivariate-scale-q2-recovery-ademp.md`.
- `dgp/sim_dgp_biv_gaussian_q2_scale_slope.R` generates bivariate Gaussian data
  with a matching residual-scale slope-only covariance block,
  `sigma1 = ~ x + (0 + x | p | id)` and
  `sigma2 = ~ x + (0 + x | p | id)`. The two log-`sigma` slopes share a q=2
  scale-scale block while residual `rho12` stays a separate row-level
  correlation. `run/sim_summary_biv_gaussian_q2_scale_slope_smoke.R`,
  `run/sim_write_biv_gaussian_q2_scale_slope_grid.R`,
  `run/sim_summary_biv_gaussian_q2_scale_slope_recovery.R`, and
  `run/sim_write_biv_gaussian_q2_scale_slope_recovery_grid.R` provide smoke and
  recovery artifacts for the two scale-slope SDs, their group-level
  correlation, fixed scale slopes, and residual `rho12`. Wald coverage is
  reported only for fixed-effect endpoints that carry standard errors.
- `dgp/sim_dgp_biv_gaussian_mu_sigma_slope.R` generates bivariate Gaussian data
  with a same-response location-scale slope covariance block,
  `mu1 = y1 ~ x + (0 + x | p | id)` and
  `sigma1 = ~ x + (0 + x | p | id)`. The `mu1` slope and log-`sigma1` slope
  share a q=2 group-level block while response 2 and residual `rho12` stay
  separate. `run/sim_summary_biv_gaussian_mu_sigma_slope_smoke.R`,
  `run/sim_write_biv_gaussian_mu_sigma_slope_grid.R`,
  `run/sim_summary_biv_gaussian_mu_sigma_slope_recovery.R`, and
  `run/sim_write_biv_gaussian_mu_sigma_slope_recovery_grid.R` provide smoke and
  recovery artifacts for the fixed endpoints, the two same-response slope SDs,
  their derived `mu_sigma` correlation, and residual `rho12`. Wald coverage is
  reported only for fixed-effect endpoints that carry standard errors.
  `run/sim_audit_biv_gaussian_mu_sigma_slope_hardening.R` reads those recovery
  artifacts, rebuilds the weak-replicate table, optionally robust-refits weak
  seeds, recomputes fixed-effect Wald coverage after replacement, and optionally
  profiles clean direct q2 targets for local promotion audits.
- `dgp/sim_dgp_biv_gaussian_q8_endpoint.R` generates bivariate Gaussian data
  with matching all-four endpoint blocks,
  `mu1 = y1 ~ x + (1 + x | p | id)`,
  `mu2 = y2 ~ x + (1 + x | p | id)`,
  `sigma1 = ~ x + (1 + x | p | id)`, and
  `sigma2 = ~ x + (1 + x | p | id)`. The eight endpoint SDs are direct
  targets, the 28 endpoint correlations are derived group-level summaries, and
  residual `rho12` stays separate. `run/sim_summary_biv_gaussian_q8_endpoint_smoke.R`,
  `run/sim_write_biv_gaussian_q8_endpoint_grid.R`,
  `run/sim_summary_biv_gaussian_q8_endpoint_recovery.R`,
  `run/sim_write_biv_gaussian_q8_endpoint_recovery_grid.R`, and
  `run/sim_write_biv_gaussian_q8_endpoint_staged_diagnostic_grid.R` provide
  diagnostic smoke/recovery/staged-start artifacts. The 2026-06-07 local
  recovery audit is recorded in
  `docs/design/161-phase-18-bivariate-q8-recovery-audit.md` and keeps the lane
  at `hold_diagnostic`, not coverage or power evidence.
- `dgp/sim_dgp_poisson_mu_random_effect.R` generates non-zero-inflated Poisson
  count data with ordinary log-mean random intercepts and independent numeric
  slopes, `(1 | id) + (0 + x | id)`, and its condition helper can cross
  group count, observations per group, fixed effects, and true random-effect
  SDs.
- `run/sim_write_poisson_mu_re_recovery_grid.R` gives the ordinary Poisson `mu`
  random-effect surface its own standalone, dispatchable recovery artifact set.
  The recovery contract (bias, RMSE, MCSE, Wald coverage for the fixed mean
  coefficients, and profile coverage for the random-effect SD) is already
  computed by `phase18_summarise_poisson_mu_re_smoke()`; this writer runs that
  summary at recovery-scale `n_rep` and emits isolated CSV artifacts through the
  opt-in `poisson_mu_re_recovery` Actions task, instead of only riding the
  combined `first_wave_summary`. This is the first standalone non-Gaussian
  recovery artifact lane.
- `dgp/sim_dgp_poisson_phylo_q1.R` generates non-zero-inflated Poisson count
  data with one q=1 phylogenetic log-mean intercept,
  `phylo(1 | species, tree = tree)`, and its condition helper can cross
  species count, observations per species, true phylogenetic SD, mean count,
  slope, and tree shape.
- `dgp/sim_dgp_nbinom2_mu_random_effect.R` generates non-zero-inflated NB2
  count data with ordinary log-mean random intercepts and independent numeric
  slopes, `(1 | id) + (0 + x | id)`, plus fixed-effect overdispersion
  `sigma ~ z`; its condition helper can also cross true overdispersion
  settings.
- `run/sim_write_nbinom2_mu_re_recovery_grid.R` gives the ordinary NB2 `mu`
  random-effect surface its own standalone recovery artifact set, the same way
  the Poisson lane does: it runs the already-recovery-capable
  `phase18_summarise_nbinom2_mu_re_smoke()` at recovery-scale `n_rep` through the
  opt-in `nbinom2_mu_re_recovery` Actions task and writes isolated bias/RMSE/MCSE,
  Wald-coverage, and profile-coverage CSVs instead of only riding
  `first_wave_summary`. (The truncated-NB2 `mu` random-intercept surface already
  has an equivalent standalone coverage-emitting lane through its existing
  `truncated_nbinom2_mu_random_intercept` task and
  `run/sim_write_truncated_nbinom2_mu_random_intercept_grid.R` writer, so it
  needs no separate recovery writer.)
- `dgp/sim_dgp_nbinom2_sigma_random_effect.R` generates non-zero-inflated NB2
  count data with fixed-effect log-mean `mu` and an ordinary grouped
  log-`sigma` random intercept, `sigma ~ z + (1 | id)`. Its condition helper
  crosses group count, repeats, mean count, baseline overdispersion, and the
  true grouped overdispersion SD.
- `dgp/sim_dgp_nbinom2_phylo_q1.R` generates non-zero-inflated NB2 count data
  with one q=1 phylogenetic log-mean intercept,
  `phylo(1 | species, tree = tree)`, fixed-effect log-`sigma`
  overdispersion, and tree-shape conditions for the formal-admission lane.
- `dgp/sim_dgp_count_structured_q1.R` generates non-zero-inflated Poisson and
  NB2 count data with one q=1 coordinate-spatial, animal-model, or `relmat()`
  log-mean intercept and fixed-effect NB2 log-`sigma` when needed.
- `dgp/sim_dgp_truncated_nbinom2_mu_random_intercept.R` generates positive
  zero-truncated NB2 count data with one ordinary grouped `mu` random
  intercept, `mu ~ x + (1 | id)`, and fixed-effect `sigma ~ z`.
- `dgp/sim_dgp_proportion_fixed_effect.R` generates strict continuous
  `beta()` proportions and denominator-aware `beta_binomial()` successes with
  fixed-effect `mu ~ x` and `sigma ~ z`.
- `dgp/sim_dgp_bounded_response_mu_random_intercept.R` generates strict
  continuous `beta()` proportions and denominator-aware `beta_binomial()`
  successes with one ordinary grouped `mu` random intercept,
  `mu ~ x + (1 | id)`, and fixed-effect `sigma ~ z`.
- `dgp/sim_dgp_positive_continuous_fixed_effect.R` generates positive
  `lognormal()` and `Gamma(link = "log")` responses with fixed-effect `mu ~ x`
  and `sigma ~ z`.
- `dgp/sim_dgp_tweedie_fixed_effect.R` generates non-negative
  semicontinuous `tweedie()` responses with fixed-effect `mu ~ x`,
  `sigma ~ z`, intercept-only `nu ~ 1`, and low- or high-zero regimes.
- `dgp/sim_dgp_positive_continuous_mu_random_intercept.R` generates positive
  `lognormal()` and `Gamma(link = "log")` responses with one ordinary grouped
  `mu` random intercept, `mu ~ x + (1 | id)`, and fixed-effect `sigma ~ z`.
- `dgp/sim_dgp_student_mu_random_intercept.R` generates Student-t responses
  with one ordinary grouped `mu` random intercept, `mu ~ x + (1 | id)`,
  fixed-effect `sigma ~ z`, and fixed-effect `nu ~ 1` using
  `nu = 2 + exp(eta_nu)`.
- `dgp/sim_dgp_ordinal_fixed_effect.R` generates ordered-category
  `cumulative_logit()` responses with latent `mu ~ x`, ordered cutpoints, and
  no free location intercept.
- `dgp/sim_dgp_zero_one_beta_fixed_effect.R` generates continuous
  proportions on `[0, 1]` with fixed-effect interior `mu ~ x`,
  interior `sigma ~ z`, exact-boundary probability `zoi ~ w`, and
  conditional-one probability `coi ~ v`.
- `dgp/sim_dgp_meta_v.R` generates Gaussian meta-analysis data with vector or
  dense known sampling covariance via `meta_V(V = V)`.
- `dgp/sim_dgp_biv_rho12.R` generates bivariate Gaussian data with
  response-specific `mu`, `sigma`, and residual-correlation `rho12 ~ w`
  predictors.
- `dgp/sim_dgp_student_shape.R` generates Student-t data with `mu ~ x`,
  `sigma ~ z`, and `nu ~ w`, using the fitted `nu = 2 + exp(eta_nu)` shape
  transform and optional mean-shape predictor correlation.
- `dgp/sim_dgp_skew_normal_fixed_effect.R` generates skew-normal responses
  with fixed-effect `mu ~ x`, `sigma ~ z`, and `nu ~ w`. It uses the public
  moment contract, then transforms to native skew-normal `xi`, `omega`, and
  `alpha = nu` for simulation so `mu` remains the response mean and `sigma`
  remains the response standard deviation.
- `dgp/sim_dgp_model_selection.R` generates paired model-selection cells for
  Gaussian versus Student-t tails, NB2 versus ZINB2 structural zeros, and
  constant versus predictor-dependent Gaussian `sigma` formulas. It records a
  `selection_target` for each cell so AIC/BIC truth-selection summaries can be
  computed without treating the article-support run as a formal power grid.
- `dgp/sim_dgp_animal_relmat_q2.R` generates bivariate Gaussian data with a
  known animal or lower-level relatedness matrix, matching q=2 `mu1`/`mu2`
  structured effects, and residual `rho12` kept as a separate layer.
- `dgp/sim_dgp_animal_relmat_q4.R` generates bivariate Gaussian data with a
  known animal or lower-level relatedness matrix, matching q=4 `mu1`/`mu2`/
  `sigma1`/`sigma2` structured effects, and residual `rho12` kept separate
  from the six latent endpoint correlations.
- `fit/sim_summarise_gaussian_ls.R` converts one fitted pilot model into a
  parameter-level truth/estimate/standard-error/error table.
- `fit/sim_summarise_meta_v.R` does the same for the `meta_V(V = V)` pilot,
  including standard errors for estimated `mu` coefficients and fitted
  residual `sigma`.
- `fit/sim_summarise_gaussian_mu_random_slope.R` summarises fixed `mu`
  coefficients, public residual `sigma`, q=3 random-slope SDs, and q=3
  derived random-effect correlations for the ordinary Gaussian `mu`
  random-slope pilot.
- `fit/sim_summarise_gaussian_sigma_random_slope.R` summarises fixed `mu` and
  fixed `sigma` coefficients plus the direct residual-scale random-slope SD on
  the modelled `log(sigma)` scale.
- `fit/sim_summarise_truncated_nbinom2_mu_random_intercept.R` summarises fixed
  zero-truncated NB2 `mu` and `sigma` coefficients plus the direct ordinary
  positive-count `mu` random-intercept SD on the public SD scale.
- `fit/sim_summarise_proportion_fixed_effect.R` summarises fixed `beta()` and
  `beta_binomial()` `mu` and `sigma` coefficients on their link scales.
- `fit/sim_summarise_bounded_response_mu_random_intercept.R` summarises
  fixed `beta()` and `beta_binomial()` `mu` and `sigma` coefficients plus the
  direct ordinary logit-mean random-intercept SD on the public SD scale.
- `fit/sim_summarise_positive_continuous_fixed_effect.R` summarises fixed
  `lognormal()` and `Gamma(link = "log")` `mu` and `sigma` coefficients on
  their documented link scales.
- `fit/sim_summarise_tweedie_fixed_effect.R` summarises fixed `tweedie()`
  `mu`, public-`sigma`, and intercept-only `nu` coefficients on their formula
  scales while carrying response-scale power and observed-zero diagnostics.
- `fit/sim_summarise_positive_continuous_mu_random_intercept.R` summarises
  fixed `lognormal()` and `Gamma(link = "log")` `mu` and `sigma` coefficients
  plus the direct ordinary positive-continuous `mu` random-intercept SD on the
  public SD scale.
- `fit/sim_summarise_student_mu_random_intercept.R` summarises fixed
  Student-t `mu`, `sigma`, and `nu` coefficients plus the direct ordinary
  Student-t `mu` random-intercept SD on the public SD scale.
- `fit/sim_summarise_ordinal_fixed_effect.R` summarises the identifiable
  cumulative-logit `mu` slope and ordered cutpoints.
- `fit/sim_summarise_zero_one_beta_fixed_effect.R` summarises fixed
  `zero_one_beta()` `mu`, `sigma`, `zoi`, and `coi` coefficients on their
  documented link scales.
- `fit/sim_summarise_spatial_mu_slope.R` summarises fixed `mu` coefficients,
  public residual `sigma`, and the two direct coordinate-spatial `mu` SDs for
  the intercept and slope fields.
- `fit/sim_summarise_phylo_mu_slope.R` summarises fixed `mu` coefficients,
  public residual `sigma`, and the two direct phylogenetic `mu` SDs for the
  intercept and slope fields.
- `fit/sim_summarise_animal_mu_slope.R` summarises fixed `mu` coefficients,
  public residual `sigma`, and the two direct dense-pedigree `animal()` `mu`
  SDs for the intercept and slope fields.
- `fit/sim_summarise_relmat_mu_slope.R` summarises fixed `mu` coefficients,
  public residual `sigma`, and the two direct known-matrix `relmat()` `mu` SDs
  for the intercept and slope fields.
- `fit/sim_summarise_spatial_q2.R` summarises fixed `mu1`/`mu2` coefficients,
  public residual scales, coordinate-spatial SDs, the spatial q=2 correlation,
  and residual `rho12` for bivariate spatial smoke fits.
- `fit/sim_summarise_poisson_mu_random_effect.R` summarises fixed Poisson
  `mu` coefficients and direct ordinary log-mean random-effect SDs.
- `fit/sim_summarise_poisson_phylo_q1.R` summarises fixed Poisson `mu`
  coefficients, the direct phylogenetic `mu` SD, profile-target status, and
  phylogenetic diagnostic status for the q=1 route.
- `fit/sim_summarise_nbinom2_mu_random_effect.R` summarises fixed NB2 `mu`
  and `sigma` coefficients plus direct ordinary log-mean random-effect SDs.
- `fit/sim_summarise_nbinom2_sigma_random_effect.R` summarises fixed NB2
  `mu`, fixed NB2 `sigma`, the direct ordinary log-`sigma` random-intercept
  SD, direct `log_sd_sigma` profile-target status, and `check_drm()`
  replication status.
- `fit/sim_summarise_count_structured_q1.R` summarises fixed Poisson/NB2 `mu`
  coefficients, fixed NB2 `sigma` coefficients when present, direct
  structured `mu` SDs, direct `log_sd_phylo` profile-target status, and
  marker-specific diagnostic status for q=1 `spatial()`, `animal()`, and
  `relmat()` count routes. It also carries fit-level Hessian and
  random-effect-SD boundary status so boundary-sensitive smoke replicates are
  visible in the replicate table.
- `fit/sim_summarise_biv_rho12.R` summarises bivariate Gaussian fixed
  `mu1`, `mu2`, `sigma1`, `sigma2`, and `rho12` coefficients on their fitted
  formula scales, adds optional profile and parametric-bootstrap interval
  columns, and includes a helper for named response-scale truth grids.
- `fit/sim_summarise_biv_gaussian_mu_slope.R` summarises bivariate Gaussian
  fixed `mu1`/`mu2` coefficients, public residual scales, residual `rho12`,
  ordinary slope-only SDs, and the slope-slope `corpairs()` correlation.
- `fit/sim_summarise_biv_gaussian_q6_location.R` summarises bivariate
  Gaussian fixed `mu1`/`mu2` coefficients, public residual scales, residual
  `rho12`, six direct q=6 location SDs, and 15 derived q=6 location
  correlations.
- `fit/sim_summarise_biv_gaussian_q2_scale.R` summarises bivariate Gaussian
  fixed `mu1`/`mu2` coefficients, public residual-scale intercepts, residual
  `rho12`, the two direct q=2 scale SDs from `sdpars$sigma`, and the single
  derived scale-scale correlation from `corpars$sigma`.
- `fit/sim_summarise_student_shape.R` summarises fixed Student-t `mu`,
  `sigma`, and `nu` coefficients on their fitted formula scales, adds optional
  profile and parametric-bootstrap interval columns, and includes a helper for
  named response-scale truth grids.
- `fit/sim_summarise_skew_normal_fixed_effect.R` summarises fixed
  skew-normal `mu`, `sigma`, and `nu` coefficients on their formula scales,
  adds optional profile and parametric-bootstrap interval columns, and includes
  a helper for named public-moment truth grids.
- `fit/sim_summarise_model_selection.R` summarises candidate-level AIC, BIC,
  log likelihood, degrees of freedom, convergence, Hessian status, warnings,
  errors, and target-selection indicators for the model-selection lane.
- `fit/sim_summarise_animal_relmat_q2.R` summarises fixed `mu1`/`mu2`
  coefficients, public residual scales, structured SDs, structured
  correlations, and residual `rho12` for known-matrix animal/`relmat()` q=2
  smoke fits.
- `fit/sim_summarise_animal_relmat_q4.R` summarises fixed `mu1`, `mu2`,
  `sigma1`, and `sigma2` coefficients, the four endpoint structured SDs, the
  six q=4 structured correlations, and residual `rho12`; requested q=4
  correlation intervals are marked as derived-unavailable rather than
  profile-ready.
- `R/sim_correlation_targets.R` classifies fitted `corpairs()` rows by their
  current profile route, keeping residual `rho12`, ordinary group, and
  phylogenetic correlations separate before simulation coverage is claimed.
- `R/sim_bootstrap.R` provides a private Phase 18 parametric-bootstrap refit
  harness, percentile interval summariser, and summary-column adapter for
  simulation studies; it records requested and actual core counts in draw and
  interval tables, and it does not change public `confint()` bootstrap support.
- `R/sim_runner.R` runs one cell replicate, captures warnings/errors, can save
  or resume an RDS result, can reload saved result directories, can bind
  replicate-level summaries, and can reduce result lists to compact manifests
  or warning/error ledgers. Its private bounded execution contract supports
  serial runs and Unix `multicore`, with actual workers capped at 10 and at the
  number of replicate jobs.
- `R/sim_aggregate.R` reduces parameter-level replicate summaries to grouped
  bias, RMSE, convergence, Hessian, warning, and elapsed-time summaries.
- `R/sim_uncertainty.R` adds Monte Carlo uncertainty and explicit
  interval-coverage summaries, plus generic Wald, profile, and interval-
  evidence table helpers for summaries that already contain estimates and
  standard errors, including a Fisher-z back-transformed path for correlation
  summaries, a failure ledger for interval rows whose status is not usable
  evidence, and interval-diagnostics summaries that separate usable coverage,
  misses, failures, and not-requested rows.
- `R/sim_plot_data.R` prepares plot-ready data tables for Phase 18 outputs,
  starting with paired Poisson/NB2 `mu` random-effect pilot summaries.
- `R/sim_gallery.R` writes plot-ready count-pilot CSV inputs and renders the
  first Florence-facing count-pilot gallery artifact from a pilot object. Its
  bias panel draws replicate-error points only when the replicate CSV carries
  `artifact_grain = "replicate"`.
- `run/sim_run_gaussian_ls_smoke.R` wires the Gaussian location-scale DGP,
  `drmTMB()` fit, summariser, registry, and replicate runner into one
  end-to-end smoke surface.
- `run/sim_run_gaussian_mu_random_slope_smoke.R` does the same for the ordinary
  Gaussian `mu` q=3 random-slope surface.
- `run/sim_run_gaussian_sigma_random_slope_smoke.R` does the same for the
  Gaussian `sigma` independent one-slope surface.
- `run/sim_run_spatial_mu_slope_smoke.R` does the same for the coordinate
  spatial Gaussian `mu` one-slope surface.
- `run/sim_run_phylo_mu_slope_smoke.R` does the same for the phylogenetic
  Gaussian `mu` one-slope surface.
- `run/sim_run_animal_mu_slope_smoke.R` does the same for the dense-pedigree
  `animal()` Gaussian `mu` one-slope surface.
- `run/sim_run_relmat_mu_slope_smoke.R` does the same for the known-matrix
  `relmat()` Gaussian `mu` one-slope surface.
- `run/sim_run_spatial_q2_smoke.R` does the same for the coordinate-spatial
  q=2 bivariate location-covariance surface.
- `run/sim_run_poisson_mu_random_effect_smoke.R` does the same for the
  non-zero-inflated Poisson `mu` random-effect surface.
- `run/sim_run_poisson_phylo_q1_smoke.R` does the same for the
  non-zero-inflated Poisson phylogenetic q=1 `mu` surface.
- `run/sim_run_nbinom2_mu_random_effect_smoke.R` does the same for the
  non-zero-inflated NB2 `mu` random-effect surface.
- `run/sim_run_count_structured_q1_smoke.R` does the same for the ordinary
  non-zero-inflated Poisson/NB2 q=1 `spatial()`, `animal()`, and `relmat()`
  `mu` intercept surface.
- `run/sim_run_nbinom2_sigma_random_effect_smoke.R` does the same for the
  non-zero-inflated NB2 log-`sigma` random-intercept surface.
- `run/sim_run_truncated_nbinom2_mu_random_intercept_smoke.R` does the same
  for the zero-truncated NB2 ordinary `mu` random-intercept surface.
- `run/sim_run_proportion_fixed_effect_smoke.R` does the same for the
  fixed-effect `beta()` and `beta_binomial()` proportion surface.
- `run/sim_run_bounded_response_mu_random_intercept_smoke.R` does the same for
  the ordinary bounded-response `mu` random-intercept surface for `beta()` and
  `beta_binomial()`.
- `run/sim_run_positive_continuous_fixed_effect_smoke.R` does the same for the
  fixed-effect `lognormal()` and `Gamma(link = "log")` positive-continuous
  surface.
- `run/sim_run_tweedie_fixed_effect_smoke.R` does the same for the
  fixed-effect `tweedie()` semicontinuous surface with intercept-only `nu`.
- `run/sim_run_positive_continuous_mu_random_intercept_smoke.R` does the same
  for the ordinary positive-continuous `mu` random-intercept surface for
  `lognormal()` and `Gamma(link = "log")`.
- `run/sim_run_student_mu_random_intercept_smoke.R` does the same for the
  ordinary Student-t `mu` random-intercept surface.
- `run/sim_run_ordinal_fixed_effect_smoke.R` does the same for the fixed-effect
  `cumulative_logit()` ordinal surface.
- `run/sim_run_zero_one_beta_fixed_effect_smoke.R` does the same for the
  fixed-effect `zero_one_beta()` exact-boundary bounded-response surface.
- `run/sim_run_meta_v_smoke.R` does the same for vector and dense
  `meta_V(V = V)` smoke cells.
- `run/sim_run_biv_rho12_smoke.R` does the same for the bivariate Gaussian
  residual-correlation surface.
- `run/sim_run_biv_gaussian_mu_slope_smoke.R` does the same for the matching
  bivariate Gaussian `mu1`/`mu2` slope-only surface.
- `run/sim_run_biv_gaussian_q6_location_smoke.R` does the same for the
  matching bivariate Gaussian q=6 `mu1`/`mu2` location surface.
- `run/sim_run_biv_gaussian_q2_scale_smoke.R` does the same for the matching
  bivariate Gaussian q=2 residual-scale intercept covariance surface.
- `run/sim_run_student_shape_smoke.R` does the same for the Student-t
  fixed-effect shape `nu` surface.
- `run/sim_run_skew_normal_fixed_effect_smoke.R` does the same for the
  fixed-effect skew-normal residual-slant surface, using moderate default
  sample sizes because stochastic skewness recovery is sample-size dependent.
- `run/sim_run_model_selection_smoke.R` does the same for the article-support
  AIC/BIC model-selection surface, writing one candidate row per replicate and
  preserving warning-bearing candidates. The file name keeps the Phase 18 smoke
  runner convention, but the article table is generated at 200 replicates.
- `run/sim_run_animal_relmat_q2_smoke.R` does the same for the known-matrix
  animal/`relmat()` q=2 bivariate location-covariance surface.
- `run/sim_run_animal_relmat_q4_smoke.R` does the same for the constant
  all-four known-matrix animal/`relmat()` q=4 location-scale surface, using
  point-estimate smoke fits by default.
- `run/sim_summary_spatial_q2_smoke.R` reduces the coordinate-spatial q=2
  smoke run into aggregate, replicate, manifest, failure-ledger, fixed-effect
  Wald interval, profile-status, interval-evidence, interval-diagnostics, and
  interval-failure tables.
- `run/sim_summary_animal_relmat_q2_smoke.R` reduces the animal/`relmat()`
  q=2 smoke run into aggregate, replicate, manifest, failure-ledger,
  fixed-effect Wald interval, profile-status, interval-evidence,
  interval-diagnostics, and interval-failure tables.
- `run/sim_summary_animal_relmat_q4_smoke.R` reduces the animal/`relmat()`
  q=4 smoke run into aggregate, replicate, manifest, failure-ledger,
  profile-status, interval-evidence, interval-diagnostics, and
  interval-failure tables. The default smoke writes no Wald rows because the
  q=4 fits use `se = FALSE`.
- `run/sim_write_spatial_q2_grid.R` writes those coordinate-spatial q=2 tables
  as repeatable CSV artifacts beside resumable per-replicate RDS files. Profile
  requests are optional; with `profile_parameters = character()`, spatial SDs,
  the spatial correlation, residual `rho12`, and residual `sigma1`/`sigma2`
  rows stay visible as `not_requested`.
- `run/sim_write_animal_relmat_q2_grid.R` writes those tables as repeatable
  CSV artifacts beside resumable per-replicate RDS files. Profile requests are
  optional; with `profile_parameters = character()`, structured SDs,
  structured correlations, residual `rho12`, and residual `sigma1`/`sigma2`
  rows stay visible as `not_requested`.
- `run/sim_write_animal_relmat_q4_grid.R` writes the q=4 smoke tables as
  repeatable CSV artifacts. If a q=4 structured-correlation row is requested
  as a profile parameter, the artifact records
  `derived_interval_unavailable` instead of treating the derived correlation
  as a direct profile target.
- `run/sim_summary_gaussian_ls_smoke.R` runs a tiny Gaussian location-scale
  summary smoke grid and returns grouped bias, RMSE, MCSE, manifest,
  warning/error ledger, formula-coefficient Wald interval, and Wald coverage
  outputs.
- `run/sim_write_gaussian_ls_grid.R` writes a repeatable Gaussian
  location-scale grid output folder with aggregate, replicate-level, manifest,
  failure, Wald-interval, and Wald-coverage CSVs beside the per-replicate RDS
  results, forwarding the private runner `cores` and `backend` settings.
- `run/sim_write_meta_v_grid.R` writes the same repeatable artifact set for
  the Gaussian `meta_V(V = V)` lane with vector and dense known sampling
  covariance cells.
- `run/sim_write_count_mu_random_effect_grid.R` writes the paired Poisson/NB2
  `mu` random-effect artifact set with Wald and direct-SD profile interval
  tables.
- `run/sim_write_truncated_nbinom2_mu_random_intercept_grid.R` writes the
  zero-truncated NB2 `mu` random-intercept artifact set with aggregate,
  replicate-level, manifest, failure-ledger, fixed-effect Wald interval, Wald
  coverage, direct-SD profile interval, and profile coverage CSVs.
- `run/sim_write_proportion_fixed_effect_grid.R` writes the fixed-effect
  `beta()` and `beta_binomial()` artifact set with aggregate, replicate-level,
  manifest, failure-ledger, fixed-effect Wald interval, and Wald coverage CSVs.
- `run/sim_write_bounded_response_mu_random_intercept_grid.R` writes the
  bounded-response `mu` random-intercept artifact set with aggregate,
  replicate-level, manifest, failure-ledger, fixed-effect Wald interval, Wald
  coverage, direct-SD profile interval, and profile coverage CSVs.
- `run/sim_write_positive_continuous_fixed_effect_grid.R` writes the
  fixed-effect `lognormal()` and `Gamma(link = "log")` artifact set with
  aggregate, replicate-level, manifest, failure-ledger, fixed-effect Wald
  interval, and Wald coverage CSVs.
- `run/sim_write_tweedie_fixed_effect_grid.R` writes the fixed-effect
  `tweedie()` semicontinuous artifact set with aggregate, replicate-level,
  manifest, failure-ledger, fixed-effect Wald interval, and Wald coverage
  CSVs.
- `run/sim_write_positive_continuous_mu_random_intercept_grid.R` writes the
  positive-continuous `mu` random-intercept artifact set with aggregate,
  replicate-level, manifest, failure-ledger, fixed-effect Wald interval, Wald
  coverage, direct-SD profile interval, and profile coverage CSVs.
- `run/sim_write_student_mu_random_intercept_grid.R` writes the Student-t
  `mu` random-intercept artifact set with aggregate, replicate-level,
  manifest, failure-ledger, fixed-effect Wald interval, Wald coverage,
  direct-SD profile interval, and profile coverage CSVs.
- `run/sim_write_ordinal_fixed_effect_grid.R` writes the fixed-effect
  `cumulative_logit()` ordinal artifact set with aggregate, replicate-level,
  manifest, failure-ledger, fixed-effect Wald interval, Wald coverage,
  cutpoint, and cutpoint-ordering CSVs.
- `run/sim_write_zero_one_beta_fixed_effect_grid.R` writes the fixed-effect
  `zero_one_beta()` artifact set with aggregate, replicate-level, manifest,
  failure-ledger, fixed-effect Wald interval, and Wald coverage CSVs.
- `run/sim_write_poisson_phylo_q1_grid.R` writes the ordinary Poisson
  phylogenetic q=1 `mu` smoke artifact set with aggregate, replicate-level,
  manifest, failure-ledger, fixed-effect Wald interval, Wald coverage, and
  direct `log_sd_phylo` profile-target CSVs beside per-replicate RDS results.
  Optional profile requests add profile-interval, profile-coverage,
  interval-evidence, interval-diagnostics, and interval-failure CSVs. The same
  file also provides the formal-grid wrapper, read-back QA, and promotion
  decision helpers; formal recovery or coverage claims still require the
  500-replicate gate and artifact review.
- `run/sim_write_nbinom2_phylo_q1_grid.R` writes the ordinary NB2 phylogenetic
  q=1 `mu` smoke artifact set with fixed-effect `sigma`, aggregate,
  replicate-level, manifest, failure-ledger, fixed-effect Wald interval, Wald
  coverage, direct `log_sd_phylo` profile-target, optional profile-interval,
  interval-evidence, interval-diagnostics, and interval-failure CSVs. Each
  replicate also records an ordinary grouped NB2 species-intercept comparator
  row, so overdispersion and unstructured species heterogeneity stay visible
  before formal recovery claims. Slices 541-555 wrote ignored local artifacts
  for a 288-cell one-replicate sentinel and a 24-cell x 5-replicate audit; both
  passed artifact QA but kept the promotion state at `hold_smoke_only` because
  the 500-replicate formal gate remains unmet. Slices 561-575 add condition
  sharding for the formal task; shard artifacts cannot allow coverage claims
  by themselves.
- `run/sim_write_count_structured_q1_grid.R` writes the ordinary Poisson/NB2
  q=1 `spatial()`, `animal()`, and `relmat()` count `mu` smoke artifact set
  with aggregate, replicate-level, manifest, failure-ledger, fixed-effect Wald
  interval, Wald coverage, direct `log_sd_phylo` profile-target, optional
  profile-interval, interval-evidence, interval-diagnostics, and
  interval-failure CSVs. It also reads artifacts back through
  `phase18_audit_count_structured_q1_boundary_gate()`, which applies the
  fitted-replicate boundary rule before a larger pilot is proposed. The
  10-replicate diagnostic pilot from Actions run `26631771105` returned
  `hold_diagnostic`, so `phase18_count_structured_q1_followup_conditions()`
  now separates `stable`, `stable_watch`, and `boundary_stress` cells before
  another diagnostic dispatch. The stable-set run `26638116979` passed the
  boundary gate with decision `propose_next_pilot`, which permits a separate
  formal-pilot design note but not a recovery or coverage claim.
- `run/sim_write_nbinom2_sigma_random_effect_grid.R` writes the ordinary NB2
  log-`sigma` random-intercept smoke artifact set with aggregate,
  replicate-level, manifest, failure-ledger, fixed-effect Wald interval, Wald
  coverage, direct `log_sd_sigma` profile-target, optional profile-interval,
  interval-evidence, interval-diagnostics, and interval-failure CSVs beside
  resumable per-replicate RDS files.
- `run/sim_write_gaussian_mu_random_slope_grid.R`,
  `run/sim_write_gaussian_sigma_random_slope_grid.R`,
  `run/sim_write_phylo_mu_slope_grid.R`,
  `run/sim_write_spatial_mu_slope_grid.R`,
  `run/sim_write_animal_mu_slope_grid.R`, and
  `run/sim_write_relmat_mu_slope_grid.R` write simple aggregate,
  replicate-level, manifest, and failure-ledger artifact sets for the ordinary
  Gaussian `mu` random-slope, independent Gaussian `sigma` random-slope, and
  Gaussian structured `mu` one-slope lanes. The ordinary Gaussian `mu` and
  `sigma` random-slope smoke grids are part of the first-wave summary runner
  and therefore run through `task = "first_wave_summary"` or `task = "all"`.
  The manual `phylo_mu_slope`, `spatial_mu_slope`, `animal_mu_slope`, and
  `relmat_mu_slope` Actions tasks can run the structured one-slope lanes as
  opt-in artifact dispatches. They remain excluded from `task = "all"` and do
  not add mesh/SPDE, sparse large-pedigree speed claims, multiple structured
  slopes, structured slope correlations, residual-scale structured slopes,
  non-Gaussian structured slopes, recovery, coverage, or power claims.
- `run/sim_write_biv_gaussian_mu_slope_grid.R` writes the same simple artifact
  set for the matching bivariate Gaussian `mu1`/`mu2` slope-only lane. The
  manual `biv_gaussian_mu_slope` Actions task can run it. Manual run
  `26689587073` audited a one-replicate, two-cell artifact with clean manifest,
  convergence, `pdHess`, and failure-ledger results, but the lane remains
  opt-in and excluded from `task = "all"` until a deliberately sized recovery
  or coverage grid is designed.
- `run/sim_write_biv_gaussian_q6_location_grid.R` writes the same simple
  artifact set for the matching bivariate Gaussian q=6 `mu1`/`mu2` location
  lane. The manual `biv_gaussian_q6_location` Actions task can run it. The
  lane remains opt-in and excluded from `task = "all"` until a deliberately
  sized recovery or coverage grid is designed.
- `run/sim_write_biv_gaussian_q2_scale_grid.R` writes the same simple artifact
  set for the bivariate Gaussian q=2 residual-scale intercept covariance lane.
  The manual `biv_gaussian_q2_scale` Actions task can run it. The lane remains
  opt-in and excluded from `task = "all"` until a deliberately sized recovery
  or coverage grid is designed.
- `run/sim_write_biv_rho12_grid.R` writes the same artifact set for the
  bivariate Gaussian residual `rho12` grid, with optional profile,
  parametric-bootstrap, combined interval-evidence, interval-diagnostics, and
  interval-failure CSVs. Replicate-runner and bootstrap backends are separate
  so a run can parallelize one layer without nesting parallel work in both;
  the runner errors if both layers would use more than one worker.
- `run/sim_write_correlation_block_status.R` writes read-only status CSVs for
  the correlation-block workflow plan, dispatchable rows, remaining wrapper
  targets, and correlation-block registry counts. The manual
  `correlation_block_status` Actions task can run it without fitting models or
  promoting q=4 derived correlations to interval-ready status.
- `run/sim_write_student_shape_grid.R` writes the same artifact set for the
  Student-t fixed-effect shape `nu` grid, with optional profile,
  parametric-bootstrap, combined interval-evidence, interval-diagnostics, and
  interval-failure CSVs. Replicate-runner and bootstrap backends are separate
  for the same reason as the bivariate `rho12` grid, with the same nested-
  parallel guard. Replicate rows also carry the `check_drm()` `student_nu`
  status, value, and message so finite-variance-boundary warning rates can be
  audited in future guard-sensitivity pilots.
- `run/sim_write_skew_normal_fixed_effect_grid.R` writes the same artifact set
  for the fixed-effect skew-normal `nu` grid, with optional profile,
  parametric-bootstrap, combined interval-evidence, interval-diagnostics, and
  interval-failure CSVs. The default grid uses `n = 720` and `1440`; larger
  formal operating-characteristic runs remain a separate evidence step.
- `run/sim_write_model_selection_smoke.R` writes model-selection candidate,
  selection-summary, manifest, and failure CSVs. The 200-replicate
  article-support artifact lives at
  `docs/dev-log/simulation-artifacts/2026-06-09-model-selection-n200/`, and the
  compact vignette table is copied to
  `inst/sim/reports/model-selection-article-summary.csv`.
- `run/sim_write_first_wave_artifact_status.R` writes bound
  artifact-manifest and surface-status CSVs from multiple grid-writer outputs,
  giving report templates a small preflight table before they read individual
  simulation artifacts.
- `run/sim_write_first_wave_table_bundle.R` writes selected first-wave artifact
  tables combined across grid-writer outputs, preserving source surface and
  artifact names while filling missing columns, and writes an artifact-grain
  preflight table so aggregate-only or missing-grain inputs cannot be mistaken
  for replicate-ready displays.
- `run/sim_render_first_wave_summary_report.R` writes first-wave status and
  table-bundle artifacts, then optionally renders the first-wave summary report
  HTML from those staged CSVs.
- `run/sim_run_first_wave_summary_smoke.R` executes the current Gaussian
  location-scale, `meta_V(V = V)`, paired Poisson/NB2 `mu` random-effect,
  zero-truncated NB2 `mu` random-intercept, fixed-effect proportion,
  bounded-response `mu` random-intercept, fixed-effect positive-continuous,
  fixed-effect Tweedie, positive-continuous `mu` random-intercept, Student-t
  `mu` random-intercept, fixed-effect ordinal, fixed-effect zero-one beta,
  ordinary Gaussian `mu` and `sigma` random-slope, and coordinate-spatial
  Gaussian `mu` slope first-wave smoke surfaces, stages the combined
  first-wave summary report, and records requested versus actual worker
  counts.
- `run/sim_run_interval_heavy_summary_smoke.R` executes the Student-t shape and
  bivariate residual `rho12` smoke surfaces as a separate interval-heavy report
  lane, keeping their Wald/profile/bootstrap artifacts out of the baseline
  first-wave runner.
- `run/sim_run_actions_cell.R` is the GitHub Actions entrypoint for manual
  long-run Phase 18 dispatch. It can run the first-wave summary task, the
  interval-heavy task, standalone zero-truncated NB2 `mu` random-intercept,
  fixed-effect proportion, bounded-response `mu` random-intercept,
  positive-continuous fixed-effect, fixed-effect Tweedie, fixed-effect
  skew-normal, count structured q=1, positive-continuous `mu` random-intercept,
  Student-t `mu`
  random-intercept, ordinal, zero-one beta, and bivariate Gaussian ordinary
  covariance tasks for `mu1`/`mu2` and `sigma1`/`sigma2`, or the opt-in Poisson
  and NB2 phylogenetic q=1 formal-grid tasks. It writes an RDS result beside the
  task artifact tables and caps
  requested replicate or bootstrap workers at 10 before dispatch. The workflow
  never uses both replicate-layer multicore and bootstrap-layer multicore at
  the same time. Standalone random-slope tasks outside the first-wave summary,
  phylogenetic formal tasks, and standalone family tasks are manual-only and
  are excluded from `task = "all"` by the workflow matrix. Formal tasks also
  accept one-based `condition_shard` and `condition_shards` inputs so the
  288-cell formal tables can be split across multiple Actions runs.
- `run/sim_summary_gaussian_mu_random_slope_smoke.R` runs a tiny ordinary
  Gaussian `mu` q=3 random-slope summary smoke grid and returns grouped bias,
  RMSE, MCSE, manifest, and warning/error ledger outputs.
- `run/sim_summary_gaussian_sigma_random_slope_smoke.R` runs a tiny Gaussian
  `sigma` independent one-slope summary smoke grid and returns grouped bias,
  RMSE, MCSE, manifest, and warning/error ledger outputs.
- `run/sim_summary_spatial_mu_slope_smoke.R` runs a tiny coordinate spatial
  Gaussian `mu` one-slope summary smoke grid and returns grouped bias, RMSE,
  MCSE, manifest, and warning/error ledger outputs.
- `run/sim_summary_phylo_mu_slope_smoke.R` runs a tiny phylogenetic Gaussian
  `mu` one-slope summary smoke grid and returns grouped bias, RMSE, MCSE,
  manifest, and warning/error ledger outputs.
- `run/sim_summary_animal_mu_slope_smoke.R` runs a tiny dense-pedigree
  `animal()` Gaussian `mu` one-slope summary smoke grid and returns grouped
  bias, RMSE, MCSE, manifest, and warning/error ledger outputs.
- `run/sim_summary_relmat_mu_slope_smoke.R` runs a tiny known-matrix
  `relmat()` Gaussian `mu` one-slope summary smoke grid and returns grouped
  bias, RMSE, MCSE, manifest, and warning/error ledger outputs.
- `run/sim_summary_poisson_mu_random_effect_smoke.R` runs a tiny
  non-zero-inflated Poisson `mu` random-effect summary smoke grid and returns
  grouped bias, RMSE, MCSE, manifest, and warning/error ledger outputs.
- `run/sim_summary_poisson_phylo_q1_smoke.R` runs a tiny non-zero-inflated
  Poisson phylogenetic q=1 `mu` summary smoke grid and returns aggregate,
  replicate, manifest, failure-ledger, fixed-effect Wald interval, Wald
  coverage, direct profile-target status, optional direct profile interval,
  interval-evidence, interval-diagnostics, and interval-failure outputs.
- `run/sim_summary_nbinom2_phylo_q1_smoke.R` runs a tiny non-zero-inflated NB2
  phylogenetic q=1 `mu` summary smoke grid with fixed-effect `sigma` and an
  ordinary grouped species-intercept comparator. It returns aggregate,
  replicate, manifest, failure-ledger, fixed-effect Wald interval, Wald
  coverage, direct profile-target status, optional direct profile interval,
  interval-evidence, interval-diagnostics, and interval-failure outputs.
- `run/sim_summary_count_structured_q1_smoke.R` runs a tiny ordinary
  Poisson/NB2 q=1 `spatial()`, `animal()`, and `relmat()` count `mu` summary
  smoke grid. It returns aggregate, replicate, manifest, failure-ledger,
  fixed-effect Wald interval, Wald coverage, direct profile-target status,
  optional direct profile interval, interval-evidence, interval-diagnostics,
  and interval-failure outputs. Replicate rows carry `sd_structured` plus
  fit-level Hessian and SD-boundary diagnostics so the boundary-gate audit can
  collapse them to fitted-replicate units.
- `run/sim_summary_nbinom2_mu_random_effect_smoke.R` runs a tiny
  non-zero-inflated NB2 `mu` random-effect summary smoke grid and returns
  grouped bias, RMSE, MCSE, manifest, warning/error ledger, formula-coefficient
  Wald interval, Wald coverage, direct random-effect SD profile interval, and
  profile coverage outputs.
- `run/sim_summary_nbinom2_sigma_random_effect_smoke.R` runs a tiny
  non-zero-inflated NB2 log-`sigma` random-intercept summary smoke grid and
  returns grouped bias, RMSE, MCSE, manifest, failure-ledger,
  formula-coefficient Wald interval, Wald coverage, direct `log_sd_sigma`
  profile-target rows, optional profile interval rows, interval diagnostics,
  and interval-failure outputs.
- `run/sim_summary_truncated_nbinom2_mu_random_intercept_smoke.R` runs a tiny
  zero-truncated NB2 ordinary `mu` random-intercept summary smoke grid and
  returns aggregate, replicate, manifest, failure-ledger, fixed-effect Wald
  interval, Wald coverage, direct-SD profile interval, and profile coverage
  outputs.
- `run/sim_summary_proportion_fixed_effect_smoke.R` runs a tiny fixed-effect
  `beta()` and `beta_binomial()` summary smoke grid and returns aggregate,
  replicate, manifest, failure-ledger, fixed-effect Wald interval, and Wald
  coverage outputs.
- `run/sim_summary_bounded_response_mu_random_intercept_smoke.R` runs a tiny
  `beta()` and `beta_binomial()` ordinary `mu` random-intercept summary smoke
  grid and returns aggregate, replicate, manifest, failure-ledger,
  fixed-effect Wald interval, Wald coverage, direct-SD profile interval, and
  profile coverage outputs.
- `run/sim_summary_positive_continuous_fixed_effect_smoke.R` runs a tiny
  fixed-effect `lognormal()` and `Gamma(link = "log")` summary smoke grid and
  returns aggregate, replicate, manifest, failure-ledger, fixed-effect Wald
  interval, and Wald coverage outputs.
- `run/sim_summary_tweedie_fixed_effect_smoke.R` runs a tiny fixed-effect
  `tweedie()` summary smoke grid and returns aggregate, replicate, manifest,
  failure-ledger, formula-coefficient Wald interval, and Wald coverage outputs.
- `run/sim_summary_positive_continuous_mu_random_intercept_smoke.R` runs a
  tiny `lognormal()` and `Gamma(link = "log")` ordinary `mu`
  random-intercept summary smoke grid and returns aggregate, replicate,
  manifest, failure-ledger, fixed-effect Wald interval, Wald coverage,
  direct-SD profile interval, and profile coverage outputs.
- `run/sim_summary_ordinal_fixed_effect_smoke.R` runs a tiny fixed-effect
  `cumulative_logit()` ordinal summary smoke grid and returns aggregate,
  replicate, manifest, failure-ledger, fixed-effect Wald interval, Wald
  coverage, cutpoint, and cutpoint-ordering outputs.
- `run/sim_summary_zero_one_beta_fixed_effect_smoke.R` runs a tiny
  fixed-effect `zero_one_beta()` summary smoke grid and returns aggregate,
  replicate, manifest, failure-ledger, fixed-effect Wald interval, and Wald
  coverage outputs.
- `run/sim_summary_count_mu_random_effect_pilot.R` runs the first paired
  Poisson/NB2 `mu` random-effect pilot, returning combined aggregate, manifest,
  failure-ledger, Wald interval, Wald coverage, profile interval, and profile
  coverage tables for count-family comparisons.
- `run/sim_render_count_mu_gallery_smoke.R` runs a tiny paired count pilot and
  renders the Florence-facing count-pilot gallery into a local output folder.
- `run/sim_summary_meta_v_smoke.R` does the same for vector and dense
  `meta_V(V = V)` smoke cells. For `sigma`, it retains the actual public Wald
  interval status and reports finite-and-covering intervals over all attempts;
  a separate conditional-on-finite set-coverage table is diagnostic only.
- `run/sim_summary_biv_rho12_smoke.R` does the same for the bivariate
  Gaussian residual `rho12` smoke grid, including formula-coefficient Wald
  intervals, optional profile and parametric-bootstrap interval evidence, and
  coverage outputs.
- `run/sim_summary_biv_gaussian_mu_slope_smoke.R` reduces the matching
  bivariate Gaussian `mu1`/`mu2` slope-only smoke run into aggregate,
  replicate, manifest, and failure-ledger tables.
- `run/sim_summary_biv_gaussian_q6_location_smoke.R` reduces the matching
  bivariate Gaussian q=6 `mu1`/`mu2` location smoke run into aggregate,
  replicate, manifest, and failure-ledger tables.
- `run/sim_summary_biv_gaussian_q2_scale_smoke.R` reduces the matching
  bivariate Gaussian q=2 residual-scale intercept covariance smoke run into
  aggregate, replicate, manifest, and failure-ledger tables.
- `run/sim_summary_student_shape_smoke.R` does the same for the Student-t
  fixed-effect shape `nu` smoke grid, including formula-coefficient Wald
  intervals, optional profile and parametric-bootstrap interval evidence,
  coverage outputs, interval-failure ledgers, and `student_nu` diagnostic
  status from `check_drm()`.
- `run/sim_summary_skew_normal_fixed_effect_smoke.R` does the same for the
  fixed-effect skew-normal residual-slant smoke grid, including
  formula-coefficient Wald intervals, optional profile and parametric-bootstrap
  interval evidence, coverage outputs, and interval-failure ledgers.
- `run/sim_interval_coverage_smoke.R` adds synthetic interval columns to
  parameter summaries so coverage-table plumbing can be tested before real
  interval methods are attached.
- `reports/phase18-smoke-report-template.Rmd` is the first reader-facing report
  template for smoke aggregate, manifest, and warning/error ledger outputs.
- `reports/phase18-first-wave-status-report.Rmd` is the first report-staging
  template for checking bound artifact-manifest and surface-status CSVs before
  larger first-wave reports read individual simulation artifacts.
- `reports/phase18-first-wave-summary-report.Rmd` is the first table-first
  summary skeleton for artifact status, aggregate operating-characteristic
  rows, a compact aggregate-bias overview, interval coverage, interval
  diagnostics, interval failures, run-manifest summaries, manifests, compact
  warning/error summaries, and full warning/error ledgers.
- `reports/phase18-count-mu-gallery.Rmd` is the first Florence-facing figure
  gallery template for paired Poisson/NB2 `mu` random-effect pilot outputs.
  It sources `R/sim_gallery_grain.R` before drawing replicate-error clouds.

## Power simulations

`R/sim_power.R` adds a power layer on top of the recovery harness. It is for a
maintainer who wants to estimate, for an admitted surface, the probability that
the public 95% interval excludes the null as a function of effect size and
sample size. The design contract is
`docs/design/154-phase-18-power-simulation-plan.md`.

The pieces, all surface-agnostic:

- `phase18_power_grid_conditions()` crosses a base condition grid with an
  effect-size sweep, tagging each cell with `effect_size`, `null_value`, and
  `is_null`. The `is_null` cell (effect equals the null) is the Type I error
  cell; the rest are power cells.
- `phase18_summarise_power()` reports per-cell `power = mean(ci_excludes_null)`
  with binomial MCSE and an `inference` label (`power`, `type_i_error`, or
  `mixed`).
- `phase18_assemble_power_table()` turns a recovery summary into that power
  table, adding Wald intervals if absent and joining condition metadata on
  `cell_id`.
- `phase18_power_curve_data()` adds a Monte Carlo band and orders rows by sample
  size; `phase18_power_target_sample_size()` interpolates the simulated curve to
  a target power; `phase18_summarise_type_i_error()` audits the null cells
  against a nominal alpha before any power number is trusted.

The runnable orchestration lives in `run/`:

- `run/sim_run_power_grid.R` defines `phase18_run_power_grid()`, the engine that
  drives any DGP/fit/summarise adapter set through the sweep and returns
  `$power`, `$curve`, and `$sample_size`.
- `run/sim_run_gaussian_ls_power_smoke.R`, `run/sim_run_meta_v_power_smoke.R`,
  and `run/sim_run_poisson_mu_re_power_smoke.R` are thin per-surface wrappers.
- `run/sim_write_power_grid.R` persists a result to CSV artifacts plus a
  manifest, and the `phase18_write_*_power_grid_outputs()` wrappers run and
  persist in one call.

To launch a grid, dispatch one of the Actions tasks `gaussian_ls_power`,
`meta_v_power`, or `poisson_mu_re_power` from
`.github/workflows/phase18-simulation-grid.yaml`, choosing `n_reps` from the
MCSE budget in doc 154. Artifacts land under
`inst/sim/results/actions/<task>/tables/`. Join the per-cell power table back to
the condition registry on `cell_id` to draw power curves.