texanshootR

Mission. To contribute to dubious research and questionable p-values.

In this day and age where publish or perish reigns king, a lone shooter helps you out in your predicament.

A shooter starts composed. Drunk, confident, and determined to convince everyone of his amazing aim, he fires at the side of a barn through most of the night. In the morning he walks over, studies the damage, and paints a target around the densest cluster of bullet holes.

Then he points at the wall and says, “See?”

texanshootR is that loop, and every shoot() call is one of his runs. He starts composed. By attempt fifty he is uncertain. By two hundred he is worried. By the time the budget is almost out he is desperate enough to escalate to a derived metric. Either p <= 0.05 lands somewhere on the wall or it does not.

Installation

install.packages("pak")
pak::pak("gcol33/texanshootR")

The package compiles a small C++ backend (penalised least squares for gam, profile-likelihood mixed model for glmm) on first install.

Quick start

library(texanshootR)

run <- shoot(mtcars)
print(run)
summary(run)

shoot() flails opportunistically (not systematically) across predictor subsets, transformations, interactions, outlier exclusions, subgroup splits, and (at higher career tiers) model families, biasing toward whatever is almost significant and abandoning specs that go cold. The print method leads with the shooter’s face and the arc he followed (composed -> resolved, panicked -> resolved (last-minute), escalated to derived metrics, etc.) before the formula and the p-value.

Every run is deterministic given a seed. The seed, R version, package version, and a hash of the search trace are recorded on the returned object so the full search can be replayed from any saved run.

Unpublishable data is now the shooter’s problem, not yours.

What’s in the box

texanshootR is a state-of-the-art specification-search engine, generously bundled with a career-progression system to recognise your dedication to the craft. Progression is not decoration: your tier governs which model families enter the search pool and which output generators you may deploy. Mastery, after all, is earned.

The search engine

• Seven model families, each with a native fitter (no soft dependencies on mgcv, lme4, or lavaan — the heavy ones are implemented in C++ via Rcpp, by hand, for your shooting needs):
  • lm — ordinary least squares via .lm.fit()
  • cor — bivariate Pearson / Spearman / Kendall
  • glm — Gaussian / binomial / Poisson / Gamma with link choices
  • wls — two-stage feasible weighted least squares
  • gam — penalised B-spline regression with GCV smoothing-parameter selection
  • glmm — Gaussian random-intercept LMM with profile-likelihood estimation
  • sem — single-mediator path model with Sobel test for the indirect effect
• A family selector that decides which fitter to attach to each spec based on career tier, escalation state, and outcome shape. At low tiers it sticks to lm. As tiers rise it leans on glm. Under desperation it occasionally forces a coercion (binomial on a continuous outcome, Poisson on a rounded one) — the shooter chasing the next-decimal p-value.
• Search moves: predictor-subset enumeration, transformation grids (log, sqrt, ^2, centring), interaction screening, outlier-exclusion seeds, subgroup splits, and a derived-metrics escalation arc when the budget is running out.
• Single result contract: every fitter returns the same shape so the selector, the highlight chooser, and the run record never branch on family.

The publication chain

A finished run is a tx_run. When shoot() lands a result that clears p <= 0.05, it opens a publication chain — six ordered output stages, each gated by its own wall-clock window:

abstract(run)            # one-paragraph deadpan summary
manuscript(run)          # IMRaD draft; Methods match the *winning* spec
presentation(run)        # 8-slide deck; residual plot on slide 7
reviewer_response(run)   # opens "we thank the reviewer for their thoughtful comments"
graphical_abstract(run)  # the figure your PI will retweet
funding(run)             # the next grant, citing the just-shipped finding

Stages must be redeemed in order. Finish the chain through your currently-unlocked prefix and you collect a length-bonus on top of the per-stage XP.

The chain breaks — bonus forfeit, partial XP kept — if you fire a fresh shoot() before finishing. Calling the wrong stage (or the wrong run) signals a tx_chain_error but leaves the chain open so you can retry.

Each generator writes to tempdir() by default and returns the file path invisibly. Override with output_dir = or options(texanshootR.output_dir = ...).

Career, achievements, cosmetics

Every redeemed stage awards XP. Cumulative XP grows the chain — same thing as unlocking deeper stages of the publication lifecycle. Your career tier is a label derived from your unlocked chain length, and it governs which model families enter the search pool:

As pressure in the research career increases, so does the need for correlation = causation…

Locked stages, expired windows, the wrong run, and out-of-order calls all signal a structured tx_chain_error so your test suite can branch on reason. Locked families simply never appear in the search trace. Your earned chain length is permanent; you just can’t shortcut into a stage you haven’t earned.

career()        # tier, runs, favourite method, opaque scores
achievements()  # 20 unlockable badges; hidden ones show as ???
wardrobe()      # equipped cosmetic slots (hat, badge, cloak, poncho, lanyard)
progress()      # HUD: chain length, XP, next unlock, live chain window
run_log()       # tibble of every run on this profile

Persistent state and saves

Your researcher profile persists under tools::R_user_dir("texanshootR", "data") as flat YAML: human-readable, version-controllable, and yours to migrate between institutions. The first interactive save prompts before writing anything to disk. Opt out entirely with options(texanshootR.save_enabled = FALSE) and the package becomes pure-stateless: every call is independent, but progression remains inert at Junior Researcher — and there is no path forward from there.

The message pack

The TUI is driven by a YAML-backed message registry under inst/messages/ with 1,257 entries across phases (blip, loading, promotion, reviewer, derived_escalation, state_transition, banner, event/event_consequence). Every message carries:

• a fallacy tag from vocab_tags (e.g. p_hacking, harking, subgroup_fishing, causal_overreach),
• a rarity weight,
• an optional mascot_state_affinity so the right line lands at the right emotional register,
• an optional model_family_affinity so the GAM-specific blip only fires when the run actually picked a GAM. Every family — lm, cor, glm, wls, gam, glmm, sem — has dedicated coverage.

You can audit the pack from R:

validate_messages()    # schema + tag-vocabulary check
vocab_tags             # canonical fallacy + thematic tags
vocab_phases           # canonical trigger phases
vocab_mascot_states    # composed / uncertain / anxious / desperate / resolved
vocab_careers          # tier ladder

The schema lives at MESSAGE_SCHEMA.md. Adding a new message is a YAML edit and a re-run of validate_messages().

Reset

Should circumstances call for a fresh start — a new institution, a co-author dispute, an opportune hard-drive failure, or because some of you just want to see the world burn — your progression can be retired:

reset_career(force = TRUE)
reset_achievements(force = TRUE)
reset_wardrobe(force = TRUE)
reset_all(force = TRUE)

Documentation

• Getting Started
• Full Reference
• Message Pack Schema
• Contributing

Further reading

Brodeur, A., Cook, N. and Heyes, A. (2020). Methods Matter: P-Hacking and Publication Bias in Causal Analysis in Economics. American Economic Review 110(11): 3634–60. https://doi.org/10.1257/aer.20190687

texanshootR is our humble contribution to a thriving field.

Support

“Where is the money, Lebowski?”

— The Big Lebowski

I’m a PhD student who builds R packages in my free time, on the principle that the tools of dubious research should be free and open. I started these for my own questionable p-values, and figured the field deserved the same head start.

If this package saved you some time (or pre-empted a fit of overfitting), buying me a coffee is a nice way to say thanks.

License

MIT © Gilles Colling. See LICENSE.

Citation

@software{texanshootR,
  author = {Colling, Gilles},
  title  = {texanshootR: Reproducible Audit Trails for Indefensible Research},
  year   = {2026},
  url    = {https://github.com/gcol33/texanshootR}
}