tidycreel

• estimate_effort_aerial_glmm(boot = TRUE) now returns NA confidence interval bounds when a multiplier's uncertainty was declared unknown, instead of an interval that quietly left it out. The total carries uncertainty from three places: the GLMM fit, the visibility correction, and the angler-to-people ratio. Declaring one unknown — visibility_correction = "none", or an NA angler_ratio_se — makes the bootstrap skip that component, because there is nothing to draw from. The standard error already went NA for precisely that reason. The interval did not: it was built independently from the bootstrap quantiles and kept a value. What that value described was an interval conditional on the unknown multiplier being exact. Measured, with the same data and the same seed: declaring the visibility correction unknown and declaring it known with zero uncertainty produced bit-for-bit identical bounds of [322.920, 449.732], distinguished only by se being NA in one and 33.98 in the other. "Never studied" was reported exactly as "studied, and found no uncertainty" — the one equivalence this package must never assert. The interval now goes NA alongside the standard error. The delta path already behaved this way, because its interval is derived from the SE and inherited the NA for free; the bootstrap path built quantiles independently and had to say so explicitly. The two paths now agree. The point estimate is untouched — this is a reporting change, not an estimation one. To get an interval back, state the correction's uncertainty (visibility_se, angler_ratio_se). Declaring it known and zero is a claim you can make deliberately; it is just no longer what silence means.

Documentation

• Examples no longer contain commented-out code. The estimate_total_catch(), estimate_total_harvest() and estimate_harvest_rate() examples now run the lines they previously only showed, and get_site_contributions(), which had no runnable example, now has one. CRAN's review of 7.0.0 asked for this.
• The validate_incomplete_trips() example and the incomplete-trips vignette restore the equivalence-threshold option they change.
• Four vignettes described behaviour the package does not have. Each was found by running the vignette and reading its own output, not by reading the source. incomplete-trips.Rmd told a reader whose validation had just **failed** to "stick with use_trips = "complete" (default)". The default is NULL, and on a roving design — which is what that vignette is about — an unspecified use_trips selects the all-trip mean-of-ratios estimator of Hoenig et al. (1997), not complete trips. Following the parenthetical produced the estimator the line above it warned against. Measured on the example data, catch rate differs by 30% between the two (mean-of-ratios-cpue at 1.257 against ratio-of-means-cpue at 0.967). The same claim appeared twice more, once in a code comment. All three are corrected, and a new section demonstrates the routing with a runnable example rather than asserting it. aerial-glmm.Rmd compared a one-day GLMM total against a twelve-day total in a table headed "side-by-side comparison", and attributed the resulting 13.5x gap to the diurnal correction. The table now states each row's basis, and the like-for-like rows differ by 10.8%, which is the diurnal correction the prose actually describes. Its "Downstream Estimation" section also claimed that GLMM effort "feeds directly into the standard downstream estimators"; it does not — estimate_total_catch() takes no effort argument and derives effort itself, so the section now says what the code does. aerial-surveys.Rmd explained that ci_lower / ci_upper "give the 95% confidence interval" directly beneath output where both were NA, because the design declares visibility_correction = "none". Both aerial vignettes now explain why an undeclared correction yields NA rather than a number, and where to supply visibility_se to get an interval. camera-surveys.Rmd said negative-duration rows are "set to NA and excluded from the daily sum". They are excluded, but the day reports a smaller number rather than NA — on the package fixture one flipped ingress/egress pair takes a day from 11 hours to 7.75 — so the warning is the only place the exclusion is visible.
• The PDF reference manual builds again. Six characters across four help pages had no LaTeX definition under pdflatex — tau, kappa, pi, a combining macron, U+2212 minus and U+1D62 subscript i — and stopped R CMD check from producing the manual at all. They are now \eqn{} math.
• The README names vignettes rather than linking to the published article pages. Those vignettes ship inside the tarball, so vignette("bus-route-surveys") works from an installed package where a link to the website only works with a network. It also means the README no longer depends on a checking machine being able to reach the documentation host.
• Every exported function now has a runnable example, and nothing is wrapped in \dontrun{}. Seventeen exported functions had no example at all, and fourteen help pages wrapped theirs in \dontrun{}. Most of those wrappers were not protecting anything: six ran as written once unwrapped, and the rest referenced objects that were never defined — design_a, my_effort, counts_df — so the block was a sketch rather than an example. One demonstrated writing a CSV into the user's working directory. Examples that genuinely need a suggested package are guarded with @examplesIf rlang::is_installed(...) rather than hidden: estimate_effort_aerial_glmm() needs lme4 and summarize_by_county() needs zipcodeR. All 125 example topics run, each in under five seconds.
• DESCRIPTION cites the methods the estimators implement — Hoenig, Jones, Pollock, Robson and Wade (1997) <doi:10.2307/2533116> for roving catch rates and Kinloch, McGlennon, Nicoll and Pike (1997) <doi:10.1016/s0165-7836(97)00068-4> for bus-route designs — and records the maintainer as copyright holder. The README shows install.packages().

Bug fixes

• Printing a validate_incomplete_trips() result no longer leaves the plot margins changed for the rest of the session. print() widened par(mar) to draw its scatter plot and never put it back; it now restores the caller's setting on exit.
• simulate_creel_data() and simulate_creel_catch() no longer leave the caller's random number stream reset. Both called set.seed(seed) and never put back what was there. seed is a convenience for reproducing one simulation, not a licence to take over the session — but a script that seeded its own analysis and then called either function part-way through silently continued from *our* seed. Every draw after that point was determined by an argument passed for one function's benefit, with nothing to say so. Measured: after set.seed(999), the next runif(1) was 0.389071 on its own and 0.685170 with a simulate_creel_catch(seed = 42) call in between. Both now give 0.389071. The stream is captured before set.seed() and restored on exit. A session that had never drawn at all is left without a .Random.seed again, rather than inheriting ours. Reproducibility is unchanged: the same seed still produces the same data, and omitting seed still consumes randomness normally.