Downscaling mixed systems #5

dlebauer · 2025-09-23T04:39:27Z

Update downscaling to support multiple PFTs including mixed systems, with demonstration of orchard + ground cover. Combine reports into webpage and publish to gh-pages.

New Utilities & Modular Scripts
- Update and rename combine_mixed_crops.R for mixed‐PFT aggregation
- New ggsave_optimized.R to streamline plot saving; use svg and webp to reduce size and increase loading speed
- Generalize helper match_site_ids_by_location.R to match site‐IDs based on lat/lon
Workflow Documentation & Site
- Update README.md with Quarto build/publish instructions
- Add _quarto.yml, index.qmd, and expanded docs (docs/, reports/) for a full Quarto‐powered website
- Rename and streamline docs/workflow_documentation.md for clarity
Configuration & Environment
- Update .Rprofile for consistent repository messaging
- Update 000-config.R, includin using argparse for development vs. production modes
- auto‐source R/ scripts

…e PFTs

…uture.R (set in config); use all outputs in production to aid in development of multi-crop workflow

… downscaling

…add change since start as delta_<variable> to standard output; aggregate output to monthly means

no longer load renv in .Rprofile; hard code path to renv libraries in .Renviron

Revise mixed_system_prototype.qmd for clarity and to fix incorrect labels.

Co-authored-by: Chris Black <chris@ckblack.org>

…cripts

…on density and stock and mask counties that cumulatively contribute <1% of total (consider this below detection limit given uncertainty)

…y and stock, save variable importance results; add logging and memory tracking

…rmat

…qmd` from `downscaling_report.qmd` - Updated `040_downscale.R` to improve naming conventions and ensure training covariate matrices are saved correctly. - Refactor `041_aggregate_to_county.R` to streamline the aggregation process and improve readability. - Enhance `042_downscale_analysis.R` to load predictions and covariates, compute variable importance, and generate plots for partial dependence, ALE, and ICE effects using the `iml` package. - Improved logging.

…n reports

…ove masks, re-organize plots for ease of comparison

…mportance report in variable_importance.qmd

…entation. Misc updates mostly to formatting.

dlebauer · 2025-10-30T17:46:47Z

R/match_site_ids_by_location.R

+#'        For California analyses, use "EPSG:3310" (NAD83 / California Albers).
+#' @param map_all logical. If TRUE, compute nearest distances for all target rows.
+#'        If FALSE, only compute nearest distances for IDs missing from reference;
+#'        matched-by-ID rows are returned with distance 0.


define max_distance in docs

divine7022

Thanks! LGTM overall, the way you are thinking about scaling is excellent.
I have dropped a few inline below; (mostly validation improvements) non-blocking, feel free to take a look and address in this PR or a follow-up

divine7022 · 2025-11-07T08:30:56Z

scripts/040_downscale.R

+#   ca_field_attributes
+# )
+
+# TODO: Need to put a canonical design_points CSV in repository


good that this temporary approach is documented with ### FOR NOW
Training RF on only 10 design points with 7 predictors and predicting on all California agricultural fields without validation is statistically dangerous.
For very small datasets (<100 samples): Random Forest is prone to severe overfitting

divine7022 · 2025-11-07T08:31:16Z

scripts/040_downscale.R

+        } else if (!is.null(mdl$y)) {
+          y_train <- mdl$y
+        }
+        r2_oob <- extract_oob_r2(mdl, y_train)


OOB samples are still from the same 10 sites, doesn't test spatial extrapolation to new locations so it can give falsely optimistic R2 with small samples.
here i would lean towards having spatial cross-validation.
This is the key validation gap regardless of design point count.

divine7022 · 2025-11-07T08:31:23Z

scripts/040_downscale.R

+    dp <- dplyr::bind_rows(dp, mixed_df_all)
+  }
+}
+


is there a plan to validate the "woody + annual" combination against real mixed-crop field data?
I am wondering if we have observational data (e.g, from orchards with cover crops) to check if predicted carbon increments are realistic.
Could we add a sanity check comparing predicted (woody + annual) / woody ratios against the ranges?

Thoughts on whether this validation is in scope for this or future iteration ?

divine7022 · 2025-11-07T09:36:01Z

scripts/041_aggregate_to_county.R

+
+ens_members_by_county <- ens_county_preds |>
+  dplyr::group_by(model_output, pft, county) |>
+  dplyr::summarize(n_vals = dplyr::n_distinct(total_c_Mg), .groups = "drop")


This checks if all counties have the same number of ensemble members, is there a case where values are identical and ends up not getting variability ? i suspect to also have check for values are actually different

divine7022 · 2025-11-07T09:46:37Z

scripts/041_aggregate_to_county.R

+  )
+}
+
+county_summaries <- ens_county_preds |>


if i am understanding this correct, when aggregating ensemble predictions, this calculates sd(total_c_Tg) which is the between-ensemble variability, but why it ignores within-county spatial heterogeneity ?

divine7022 · 2025-11-07T12:56:47Z

scripts/042_downscale_analysis.R

+  par(mar = c(5, 5, 4, 2))
+
+  randomForest::partialPlot(rf,
+    pred.data = design_covariates, x.var = top_predictors[1], ylim = yl,


yl is not defined. Either remove ylim let R auto scale y or calculate ylim from training response value.

y_range <- range(rf$y, na.rm = TRUE) yl <- c(floor(min(y_range)), ceiling(max(y_range)))

divine7022 · 2025-11-07T12:59:30Z

scripts/042_downscale_analysis.R

+  dev.off()
+  PEcAn.logger::logger.info("Saved importance/PDP figure:", importance_partial_plot_fig)
+}
+


The script currently creates PDPs first and ALE plots afterward.
This order is backwards, since the covariates (clay, temp, precip) are highly correlated.
PDPs fail with correlated features because they marginalize over unrealistic or physically impossible feature combinations, while ALE plots remain unbiased and valid in such cases.

The script already generates ALE plots, but they’re treated as secondary.
Suggestion: swap the priority, generate ALE plots first, and use PDPs only if features are uncorrelated.

divine7022 · 2025-11-07T14:57:04Z

R/match_site_ids_by_location.R

+#' @param map_all logical. If TRUE, compute nearest distances for all target rows.
+#'        If FALSE, only compute nearest distances for IDs missing from reference;
+#'        matched-by-ID rows are returned with distance 0.
+#' @return a tibble with mapping and distances (same number of rows as target_df)


Suggested change

#' @return a tibble with mapping and distances (same number of rows as target_df)

#' @return a tibble with one row per target row, containing:

#' \describe{

#' \item{target_site_id}{Original target site ID}

#' \item{matched_site_id}{Matched reference site ID (may differ from target if not found by ID)}

#' \item{target_lat, target_lon}{Original target coordinates}

#' \item{ref_lat, ref_lon}{Matched reference coordinates}

#' \item{distance_m}{Distance between target and matched reference (meters)}

#' \item{close}{Proximity classification: "same location", "very close", "close", "moderate", "far"}

#' }

#' Rows are returned in the same order as target_df.

divine7022 · 2025-11-07T15:02:35Z

R/match_site_ids_by_location.R

+      paste(setdiff(req_ref, colnames(reference_df)), collapse = ", ")
+    )
+  }
+


nit: there is no check for duplicate target site_ids. Could target_df contain duplicate site_ids with different coordinates?

divine7022 · 2025-11-07T15:14:42Z

scripts/021_clustering_diagnostics.R

dplyr::sample_n(min(nrow(sites_clustered), 1000)) |>

consider setting seed here for reproducibility and potentially migrate to slice_sample() since sample_n is suppressed

divine7022 · 2025-11-07T15:15:39Z

ca_fields_pts <- ca_fields |> 
sf::st_as_sf(coords = c("lon", "lat"), crs = 4326)

is ca_fields already an sf object (from line 5)? If so, this conversion will fail.

ca_fields_pts <- ca_fields |>
 sf::st_centroid() |>
 sf::st_transform(crs = 4326)

this makes an intent explicit.

divine7022 · 2025-11-07T15:16:26Z

sites_clustered <- readRDS(file.path(cache_dir, "sites_clustered.rds"))

no validation that clustering was successful or that file exists

divine7022 · 2025-11-07T15:17:07Z

scripts/031_aggregate_sipnet_output.R

ens_combined <- ens_wide |> 
dplyr::rowwise() |> 
dplyr::mutate( 
value_combined = combine_value(...) 
)

rowwise() is slow for large datasets, consider using purrr::pmap here

divine7022 · 2025-11-07T15:17:56Z

annual_init = dplyr::case_when( 
# TODO: think through this 
# we want to add AGB to the ecosystem level value 
# for SOC, we only want the diff 
# this probably isn't the best place to store this logic 
# also, 
variable == "AGB" ~ 0, 
variable == "TotSoilCarb" ~ value 
)

Why treating AGB and SOC differently?

divine7022 · 2025-11-07T15:18:41Z

dplyr::filter(mix_description == "100% woody + 50% annual")

why is 50% ground cover the "canonical" scenario for EFI output?

divine7022 · 2025-11-07T15:19:30Z

C_combined = C_woody + f × C_annual

current approach assumes linear additivity, but mixed systems have non-linear interactions ?

infotroph

Just realized I wrote these comments ages ago and they've been sitting unposted -- sorry for the wait.

000-config.R

infotroph · 2025-11-04T17:11:32Z

000-config.R

+  "- model_outdir.  : {model_outdir}\n\n",
+  "### Other Settings ###\n",
+  "- will extract variables: {paste(outputs_to_extract, collapse = ', ')}\n",
+  "- ca_albers_crs  : {ca_albers_crs}{if(ca_albers_crs == 3310) ', which is NAD83 / California Albers' else ''}\n"


Nits:

This seems like a lot of custom logic for an output that's only helpful in one case. Would terra::crs(ca_albers_crs, describe = TRUE)$name help?

Between the variable name and the special handling of 3310 in the summary, I'm hearing an implicit assumption that the projection will be Albers of at least some flavor. Does the code rely on that? If yes consider enforcing more vigorously than just notes in the printed summary, if no then probably clearer to drop "albers" from the variable name.

infotroph · 2025-11-04T17:16:00Z

R/combine_mixed_crops.R

+#' @param woody_value numeric. Pool size for the woody PFT (kg/m2).
+#' @param annual_value numeric. Pool size for the annual PFT (kg/m2).


Does this work on truly any pool, or only select ones? I think I knew this at one point but can't remember on a cold reread, so probably worth documenting.

R/combine_mixed_crops.R

tests/testthat/helper.R

infotroph · 2025-11-05T00:08:06Z

R/match_site_ids_by_location.R

+#' @param reference_lat_col character. Latitude column in reference_df (default "lat")
+#' @param reference_lon_col character. Longitude column in reference_df (default "lon")
+#' @param crs character. Coordinate reference system of input points (default "EPSG:4326").
+#'        For California analyses, use "EPSG:3310" (NAD83 / California Albers).


I read this as recommending the user pass crs = "EPSG:3310" for any input located in CA, but that will be incorrect any time the locations are given in degrees lat/lon.

As ?terra::crs says (their emphasis):

set the CRS of a dataset (if it does not come with one) to what it is, not to what you would like it to be. See project() to transform an object from one CRS to another.

So when the input units are angular degrees the CRS should be geographic (as 4326 is), not projected (as 3310 is). Only use 3310 if both targ and ref really do give their lat and lon in units of meters from the CA Albers origin point.

If the CRS is set wrong, the distance calculations will be (probably very) incorrect, and could even make it match the wrong site:

> targ <- data.frame(lat = 40, lon = -120, site_id = 100) > ref <- data.frame(lat = c(40.0, 40.001), lon = c(-120.0011, -120.0), site_id = 1:2) > match_site_ids_by_location(targ, ref, crs = "epsg:4326") 2025-11-05 13:40:48.787131 WARN [match_site_ids_by_location.R#65: PEcAn.logger::logger.warn] : 1 target sites not in reference by ID; matching by nearest location. # A tibble: 1 × 8 target_site_id matched_site_id target_lat target_lon ref_lat ref_lon distance_m <dbl> <int> <dbl> <dbl> <dbl> <dbl> <dbl> 1 100 1 40 -120 40 -120. 93.9 # ℹ 1 more variable: close <chr> > match_site_ids_by_location(targ, ref, crs = "epsg:3310") 2025-11-05 13:40:54.977141 WARN [match_site_ids_by_location.R#65: PEcAn.logger::logger.warn] : 1 target sites not in reference by ID; matching by nearest location. # A tibble: 1 × 8 target_site_id matched_site_id target_lat target_lon ref_lat ref_lon distance_m <dbl> <int> <dbl> <dbl> <dbl> <dbl> <dbl> 1 100 2 40 -120 40.0 -120 0.00100 # ℹ 1 more variable: close <chr>

infotroph · 2025-11-05T00:29:28Z

R/match_site_ids_by_location.R

@@ -0,0 +1,209 @@
+#' Match site IDs by geographic location
+#'
+#' Matches site IDs from a target data.frame to the nearest reference site IDs based on latitude/longitude.


High level impression: If I'm following the code right, this function tries to match first by ID without considering location and then by location for sites whose IDs didn't match -- sort of a fuzzy join between two datasets assumed to be drawn from the same set of IDs and almost the same set of points. But this documentation seems to present it as a match by location first (or only?), and expecting that left me confused for my first few test runs. I think in general I'd want to do an id-then-location join as an explicitly two-step process.

My immediate use case for this function would be to generate mappings between datasets that are already known to have different reference and target site IDs. In those cases I want every target point to match the closest reference location, even if there's a further reference point whose ID collides with the target's ID. As far as I can tell that's not available with the current design: Even with map_all = TRUE, the returned match will be to the matching ID not the nearest location, yeah?

infotroph · 2025-11-06T08:41:35Z

R/match_site_ids_by_location.R

+#'
+#' @param target_df data.frame
+#' @param reference_df data.frame
+#' @inheritParams match_site_ids_by_location


Nit: Roxygen will probably complain about conflicting definitions of target_df and reference_df (the others come in from inheritParams) and generate a confusing Rd file from them

infotroph · 2025-11-30T09:40:14Z

tests/testthat/test-match_site_ids.R

+  reference <- data.frame(
+    site_id = c("A", "B"),
+    lat = c(34.0, 35.0),
+    lon = c(-118.0, -119.0)
+  )


Nit: If these are supposed to really be identical, copy one so readers don't have to play spot the difference

Suggested change

reference <- data.frame(

site_id = c("A", "B"),

lat = c(34.0, 35.0),

lon = c(-118.0, -119.0)

)

reference <- target

Co-authored-by: Chris Black <chris@ckblack.org>

R/ggsave_optimized.R

infotroph · 2025-12-01T04:12:08Z

000-config.R

+  type = "logical", default = TRUE,
+  help = "Set to true for production mode, false for faster development (default: FALSE)"


default/doc mismatch?

Suggested change

type = "logical", default = TRUE,

help = "Set to true for production mode, false for faster development (default: FALSE)"

type = "logical", default = TRUE,

help = "Set to true for production mode, false for faster development (default: TRUE)"

dlebauer

Consider that canopy_cover != fraction of impact. This is model-specific assumption that should be integrated into write.configs.SIPNET

dlebauer · 2026-01-14T23:53:08Z

R/combine_mixed_crops.R

+  out_of_range_woody <- (woody_cover < 0 - tol) | (woody_cover > 1 + tol)
+  if (any(out_of_range_annual | out_of_range_woody, na.rm = TRUE)) {
+    n_bad <- sum(out_of_range_annual | out_of_range_woody, na.rm = TRUE)
+    PEcAn.logger::logger.severe(paste0(n_bad, " rows have cover fractions outside [0,1] (<U+00B1>tol)."))


Suggested change

PEcAn.logger::logger.severe(paste0(n_bad, " rows have cover fractions outside [0,1] (<U+00B1>tol)."))

PEcAn.logger::logger.severe("weighted: cover fractions outside must be in the range [0,1] (+/- tol).", n_bad, "rows violate.")

dlebauer · 2026-01-14T23:54:16Z

R/combine_mixed_crops.R

+    not_one <- abs(woody_cover - 1) > tol
+    if (any(not_one, na.rm = TRUE)) {
+      n_bad <- sum(not_one, na.rm = TRUE)
+      PEcAn.logger::logger.severe(paste0("incremental: woody_cover must be 1 (+/- ", tol, "); ", n_bad, " rows violate."))


Why are errors phrased differently?

I've rearranged the sentences. For incremental, woody must be 100%.

Suggested change

PEcAn.logger::logger.severe(paste0("incremental: woody_cover must be 1 (+/- ", tol, "); ", n_bad, " rows violate."))

PEcAn.logger::logger.severe("incremental: woody_cover must be 1 (+/- ", tol, "); ", n_bad, " rows violate.")

dlebauer · 2026-01-14T23:58:20Z

R/combine_mixed_crops.R

+      PEcAn.logger::logger.severe(paste0("incremental: woody_cover must be 1 (+/- ", tol, "); ", n_bad, " rows violate."))
+    }
+    res <- woody_value + annual_cover * (annual_value - annual_init)
+    return(as.numeric(res))


Suggested change

return(as.numeric(res))

return(res)

dlebauer and others added 29 commits September 23, 2025 01:00

added exploration demo of different approaches to aggregating multipl…

e07e3a8

…e PFTs

add points to plot

5c8d46e

update config - change location of model outputs, deprecate use of .f…

1560630

…uture.R (set in config); use all outputs in production to aid in development of multi-crop workflow

Add nearest‐polygon matching for unmatched sites

ddbde37

add note in docs that workflows will be split into site selection and…

696a933

… downscaling

removed duplicate chunk from 009

ff23d5e

Refactor SIPNET output extraction script for clarity and efficiency; …

15e6316

…add change since start as delta_<variable> to standard output; aggregate output to monthly means

clarify documentation; disable renv auto activation in .Rprofile

12f4a43

major refactoring of mixed_system_prototype.qmd

d0cd427

no longer load renv in .Rprofile; hard code path to renv libraries in .Renviron

Added 031_agggregate_sipnet_output.R to handle PFT mixture logic

68ef946

Revise mixed_system_prototype.qmd for clarity and to fix incorrect labels.

Update docs/mixed_system_prototype.qmd

258ef31

Co-authored-by: Chris Black <chris@ckblack.org>

add warning that we need new method for flux variables

07ea167

addressing pr suggestions

2758e83

small wording changes to explain figs

7e4c034

few more minor changes

da406d5

few more tweaks

1101916

add mixed aggregation helper function

c517598

standardize naming of ensemble_data object

36a3054

in the middle of refactoring to handle multiple PFTs in the downscaling

412ffd5

Rename downscaling scripts, split downscale+aggregate into separate s…

08e1532

…cripts

Set up quarto publish gh-pages: add _quarto.yml, update README, etc

295ec2e

Add ggsave_optimized function and update plot saving in scripts

7de1418

rename, vectorize, and test combine_value --> combine_mixed_crops.

61afd26

Refactor county-level plotting to include delta calculations for carb…

b576d1e

…on density and stock and mask counties that cumulatively contribute <1% of total (consider this below detection limit given uncertainty)

downscaling script: add mixed scenario calculations for carbon densit…

176e200

…y and stock, save variable importance results; add logging and memory tracking

Add testthat helper for logging configuration and sourcing R functions

f5a8bf2

Add references section and update workflow documentation title and fo…

7e4c5e6

…rmat

Update document formats to non-self-contained and adjust references i…

a2e81d8

…n reports

dlebauer force-pushed the mixed_downscaling branch from 62d0577 to a2e81d8 Compare September 23, 2025 05:35

dlebauer changed the title ~~Mixed downscaling~~ Downscaling mixed systems Sep 23, 2025

dlebauer added 2 commits September 24, 2025 19:37

Remove unused masking threshold variable from configuration

d89297e

Add county comparison sections and update workflow documentation, rem…

57c2f90

…ove masks, re-organize plots for ease of comparison

dlebauer force-pushed the mixed_downscaling branch from 673106e to 57c2f90 Compare September 25, 2025 00:01

dlebauer added 5 commits September 25, 2025 01:02

Update file paths in mixed_system_prototype.qmd and refine variable i…

47c0aa1

…mportance report in variable_importance.qmd

Change default mode production=true, refine README and workflow docum…

9e3bc0b

…entation. Misc updates mostly to formatting.

fix link

f3a5a9d

fix link

3b0b24e

Move sidebar to left and add gh link; add alert callout to README

04065d9

dlebauer requested review from divine7022 and mdietze October 24, 2025 18:27

dlebauer commented Oct 30, 2025

View reviewed changes

divine7022 reviewed Nov 7, 2025

View reviewed changes

infotroph reviewed Nov 30, 2025

View reviewed changes

dlebauer and others added 4 commits November 30, 2025 19:02

Apply suggestion from @infotroph

a44d5a0

Co-authored-by: Chris Black <chris@ckblack.org>

Apply suggestion from @dlebauer

ec9e30b

Apply suggestion from @infotroph

0627842

Co-authored-by: Chris Black <chris@ckblack.org>

Apply suggestion from @infotroph

169a229

Co-authored-by: Chris Black <chris@ckblack.org>

dlebauer commented Dec 1, 2025

View reviewed changes

R/ggsave_optimized.R Outdated Show resolved Hide resolved

Apply suggestion from @dlebauer

422d3f0

infotroph reviewed Dec 1, 2025

View reviewed changes

dlebauer mentioned this pull request Jan 13, 2026

Downscale management scenarios to field scale ccmmf/organization#167

Open

dlebauer commented Jan 15, 2026

View reviewed changes

-#' @return a tibble with mapping and distances (same number of rows as target_df)
+#' @return a tibble with one row per target row, containing:
+#'   \describe{
+#'     \item{target_site_id}{Original target site ID}
+#'     \item{matched_site_id}{Matched reference site ID (may differ from target if not found by ID)}
+#'     \item{target_lat, target_lon}{Original target coordinates}
+#'     \item{ref_lat, ref_lon}{Matched reference coordinates}
+#'     \item{distance_m}{Distance between target and matched reference (meters)}
+#'     \item{close}{Proximity classification: "same location", "very close", "close", "moderate", "far"}
+#'   }
+#'   Rows are returned in the same order as target_df.

		#' @param woody_value numeric. Pool size for the woody PFT (kg/m2).
		#' @param annual_value numeric. Pool size for the annual PFT (kg/m2).

		type = "logical", default = TRUE,
		help = "Set to true for production mode, false for faster development (default: FALSE)"

	PEcAn.logger::logger.severe(paste0(n_bad, " rows have cover fractions outside [0,1] (<U+00B1>tol)."))
	PEcAn.logger::logger.severe("weighted: cover fractions outside must be in the range [0,1] (+/- tol).", n_bad, "rows violate.")

	PEcAn.logger::logger.severe(paste0("incremental: woody_cover must be 1 (+/- ", tol, "); ", n_bad, " rows violate."))
	PEcAn.logger::logger.severe("incremental: woody_cover must be 1 (+/- ", tol, "); ", n_bad, " rows violate.")

Downscaling mixed systems #5

Are you sure you want to change the base?

Downscaling mixed systems #5

Uh oh!

Conversation

dlebauer commented Sep 23, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

dlebauer Oct 30, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

divine7022 left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

divine7022 Nov 7, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

divine7022 Nov 7, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

divine7022 commented Nov 7, 2025

Uh oh!

divine7022 commented Nov 7, 2025

Uh oh!

divine7022 commented Nov 7, 2025

Uh oh!

divine7022 commented Nov 7, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

divine7022 commented Nov 7, 2025

Uh oh!

divine7022 commented Nov 7, 2025

Uh oh!

divine7022 commented Nov 7, 2025

Uh oh!

infotroph left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

dlebauer left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

dlebauer commented Sep 23, 2025 •

edited

Loading

dlebauer Oct 30, 2025 •

edited

Loading

divine7022 left a comment •

edited

Loading

divine7022 Nov 7, 2025 •

edited

Loading

divine7022 Nov 7, 2025 •

edited

Loading

divine7022 commented Nov 7, 2025 •

edited

Loading