Clinical-reporting vignettes should use an explicit rounding policy

Problem

The executable clinical-reporting vignettes use base round() and formatC(..., format = "f") directly for percentages and summary statistics. Under a SAS-compatible, ties-away-from-zero reporting policy, these calls can display the wrong value at ties.

This is not a request to change RTF layout/geometry rounding. It is limited to examples that calculate or render clinical-style report statistics.

Reproduced on current master

• Commit: 8d7b4adc69db750a639b9129290be4818916717e
• R: 4.5.3

formatC(1.25, format = "f", digits = 1) # "1.2"
round(1.25, 1)                            # 1.2

For the intended ties-away policy, both values should be 1.3; likewise -1.25 should become -1.3. Decimal examples also depend on binary representation, so a solution must be tested with exact and inexact decimal ties rather than assuming that formatC() has a single tie mode.

Affected reporting paths

• vignettes/example-basechar.Rmd:56,91-93,103-105 — percentage, mean, SD, and median display with formatC().
• vignettes/example-efficacy.Rmd:40,42,55-56,69,156 — fmt_est, fmt_ci, and fmt_pval display estimates, confidence intervals, p-values, and residual SD with formatC().
• vignettes/example-ae-summary.Rmd:39 — AE percentage calculation uses round(..., 2).
• vignettes/rtf-row.Rmd:143 — executable table example uses formatC().

The package's inch-to-twip, font-size, pagination, and cell-geometry calculations are intentionally out of scope.

Requested change

Make the reporting vignettes use one explicit, documented rounding policy before fixed-width display. Keep calculations unrounded until the final display step; then apply a ties-away-from-zero helper at the required digits and render with a fixed-width formatter such as formatC(..., format = "f", digits = ...).

The implementation may use a versioned dependency or a small documented helper, but it should not add a hidden policy change to generic RTF layout code. If the project does not intend these vignettes to promise SAS-compatible rounding, document that limitation prominently instead of silently relying on base R behavior.

Acceptance criteria

• Reporting examples use an explicit and documented rounding policy at their display boundaries; generic RTF layout/encoding code remains unchanged.
• At one decimal, positive and negative exact ties use ties-away behavior: 1.25 -> "1.3" and -1.25 -> "-1.3".
• Tests cover representative decimal ties that expose binary-representation effects, not only x.5 at zero decimals.
• Values that round to zero never render as "-0" or "-0.0".
• Fixed-width display is retained where the vignette requests it (for example, 76 -> "76.00" at two decimals).
• Summary statistics are calculated from unrounded inputs and rounded only when prepared for display.
• Affected vignettes render successfully after the change.

Audit evidence

The repository-root audit scanned 57 supported reporting files (45 .R, 12 .Rmd) and inventoried 421 candidates. The Rounding Skill probe completed 7/7 invariants in the audit environment. The detailed audit record is available at https://github.com/RConsortium/pharma-skills/issues/174#issuecomment-5577387523.

Read the affected executable vignettes: vignettes/example-basechar.Rmd, example-efficacy.Rmd, example-ae-summary.Rmd, and rtf-row.Rmd. First reproduce the supplied R tie examples, then inspect each reporting display boundary and verify the acceptance cases, including negative ties, decimal representation effects, fixed-width output, unrounded summaries, and successful vignette rendering.