Close out backlog

[malcolmbarrett]

• pin the solver tolerance defaults alongside the iteration caps at the savvy boundary
• document the resolved solver defaults in every bw_*() constructor
• assert both alerts in the aliasing full-set constraint test
• state the aliasing QR tolerance explicitly and restore the stacked-bread comment
• sweep the pooled ipw class for stray S3 registrations
• pin the focal standardization of the categorical mean rows
• pin the print block and the near-zero scrub for a dropped factor level
• scrub near-zero balance values to a placeholder in cli snapshots
• pin a classed error for non-finite weights on the continuous entropy path
• guard the continuous entropy rescale against non-finite weights
• pin a classed refusal for an aliased outcome coefficient in ipw()
• refuse an aliased outcome coefficient in ipw() before stacking
• test that the balance-exceeded warning shows a small imbalance
• report the largest imbalance to significant digits
• test that the msm sandwich returns its coefficient surface
• return the coefficient surface from ipw_deli_msm_sandwich() instead of recomputing it
• pin the balance table against a prebuilt constraint matrix
• hand the balance table its prebuilt constraint matrix and vectorize its column statistics
• pin the stacked psi assembly against rbind
• fill a preallocated matrix for the stacked psi assembly
• pin the weighted standardizing sd against a date-time offset
• compute the weighted standardizing variance in two passes
• say what the stacked psi fill actually saves
• add failing tests for a difftime covariate in balance()
• coerce difftime covariates to numeric
• add failing tests for a three-significant-digit largest imbalance line
• print the largest imbalance to three significant digits
• require the development version of propensity in Suggests
• pin the OSQP wrapper's own refusal of a spec with no decision variables
• refuse a spec with no decision variables in the OSQP wrapper
• rename IptOptions to ScoreOptions in the savvy crate
• say stdout, not stderr, for the OSQP validation line
• pin the energy tolerance default, the quadratic-program convergence advice, and a usable fallback iterate
• give energy balancing a reachable solver tolerance, quadratic-program convergence advice, and a fallback solve
• add worked .by and joint exposure examples to the inference vignette
• pin the stacked sandwich against a fully differenced reference before the contrast block goes analytic
• correct the joint exposure comment and the subgroup covariance sentence in the inference vignette
• handle the deterministic contrast rows without differencing them
• qualify the quadratic-program tolerance warning as default-backend behavior
• cross-reference the two rank tolerances in constraints.R and ipw-deli.R
• scope the energy fallback wording to the fits it describes
• refuse a psi block whose column count is not the sample size
• state the near-zero snapshot cutoff its exponent rule actually reaches
• scrub an exact zero on the largest-imbalance line like a near-zero one
• exempt a reported tolerance from the near-zero snapshot rule
• name the estimand methods in the borrowed ipw accessor sweep
• match the propensity skips in test-weights.R to the Suggests floor
• move the by and joint ipw fixtures into helper-dgp.R
• avoid a word the spell check does not carry in the energy iteration note
• read the physical core count once per session
• standardize a constant column to zero under sampling weights
• translate deli's non-finite estimating functions during stacking
• read Date and POSIXct covariates as the numbers they store
• collapse balance-table alignment in the snapshot scrub
• read the weighted solver box as column arithmetic
• center the scaled Euclidean transform at its weighted column means
• pin the weighted moments and the transforms across thread counts above the reduction chunk
• add a criterion bench for the covariate distance transforms
• rename the balance() focal_level argument to .focal_level
• correct the iterations, energy backend, and scaled_euclidean documentation
• coerce a POSIXlt covariate the way a POSIXct one is coerced
• name the sx accumulation error alongside the representation floor in the offset-column comment
• read a constant column from its values in the entropy solver box
• cut the stacked psi assembly's allocation by writing its rows in place
• poll R's interrupt flags instead of pushing a toplevel context on Unix
• record what the Unix interrupt poll gives up from R_ProcessEvents
• assert the column exists before asserting its values are finite
• say that RAYON_NUM_THREADS has no bearing on the resolved thread count
• pin the constant-column tolerance box on the cfd and sbw paths
• read a column with a missing value as varying in column_is_constant()
• pin the t kernel's accuracy on a date-time column at its epoch offset
• read R's interrupt flags as relaxed atomics rather than volatile
• document that a fit is interruptible between solver iterations
• close the empty-column hole in the suite's sibling column comparisons
• read every column of a zero-row matrix as varying in column_is_constant()
• give balance_terms(moments) its correlation meaning on the continuous energy path
• cover the continuous energy re-solve at a reachable tolerance
• close the empty-column and missing-answer holes in the suite's column expectations
• route the suite's bare all() assertions through a non-empty expectation
• say what a continuous energy fit does with an ignored tolerance
• build the continuous energy marginal rows from the covariates alone
• cut ipw()'s bread allocation by summing the psi blocks instead of stacking them
• word the psi block width refusal from the entry and report it at the assembly
• refuse a psi block that does not hold doubles
• build the continuous entropy marginal rows from the covariates alone
• pin the table length in two within_tolerance assertions
• honor a positive tolerance on the continuous energy correlation rows
• state the refinement cost in solves rather than in wall time
• report the tolerance a bare-tolerance cfd fit enforced
• drop the stack_psi_blocks width test the snapshot already covers
• fall back to the last converged iterate in the energy correlation refinement
• pin that a continuous entropy interaction column gets no marginal row
• cover the correlation refinement restore branches for energy and stable balancing
• repair a spliced test comment and pin the cfd relaxation tolerance column
• sum every correlation refinement pass in the reported stable balancing iterations
• pin the energy restore spec at the tolerance that keeps the re-solve fallback out