Dataset Viewer
Auto-converted to Parquet Duplicate
city_id
stringclasses
210 values
year
int64
2.02k
2.03k
metric
stringclasses
4 values
predicted_value
float64
2.99
5.51k
USC00042863
2,019
annual_avg_temp
66.949
USC00042863
2,019
total_precip_in
13.219
USC00042863
2,019
hottest_day_temp
104.058
USC00042863
2,019
cdd
1,719.956
USC00042863
2,020
annual_avg_temp
67.016
USC00042863
2,020
total_precip_in
13.219
USC00042863
2,020
hottest_day_temp
104.058
USC00042863
2,020
cdd
1,734.805
USC00042863
2,021
annual_avg_temp
67.084
USC00042863
2,021
total_precip_in
13.219
USC00042863
2,021
hottest_day_temp
104.058
USC00042863
2,021
cdd
1,749.653
USC00042863
2,022
annual_avg_temp
67.151
USC00042863
2,022
total_precip_in
13.219
USC00042863
2,022
hottest_day_temp
104.058
USC00042863
2,022
cdd
1,764.501
USC00042863
2,023
annual_avg_temp
67.218
USC00042863
2,023
total_precip_in
13.219
USC00042863
2,023
hottest_day_temp
104.058
USC00042863
2,023
cdd
1,779.35
USC00042863
2,024
annual_avg_temp
67.286
USC00042863
2,024
total_precip_in
13.219
USC00042863
2,024
hottest_day_temp
104.058
USC00042863
2,024
cdd
1,794.199
USC00042863
2,025
annual_avg_temp
67.353
USC00042863
2,025
total_precip_in
13.219
USC00042863
2,025
hottest_day_temp
104.058
USC00042863
2,025
cdd
1,809.047
USC00166584
2,019
annual_avg_temp
67.587
USC00166584
2,019
total_precip_in
56.141
USC00166584
2,019
hottest_day_temp
100.945
USC00166584
2,019
cdd
2,851.41
USC00166584
2,020
annual_avg_temp
67.655
USC00166584
2,020
total_precip_in
56.141
USC00166584
2,020
hottest_day_temp
100.945
USC00166584
2,020
cdd
2,874.747
USC00166584
2,021
annual_avg_temp
67.722
USC00166584
2,021
total_precip_in
56.141
USC00166584
2,021
hottest_day_temp
100.945
USC00166584
2,021
cdd
2,898.083
USC00166584
2,022
annual_avg_temp
67.789
USC00166584
2,022
total_precip_in
56.141
USC00166584
2,022
hottest_day_temp
100.945
USC00166584
2,022
cdd
2,921.419
USC00166584
2,023
annual_avg_temp
67.856
USC00166584
2,023
total_precip_in
56.141
USC00166584
2,023
hottest_day_temp
100.945
USC00166584
2,023
cdd
2,944.755
USC00166584
2,024
annual_avg_temp
67.924
USC00166584
2,024
total_precip_in
56.141
USC00166584
2,024
hottest_day_temp
100.945
USC00166584
2,024
cdd
2,968.091
USC00166584
2,025
annual_avg_temp
67.991
USC00166584
2,025
total_precip_in
56.141
USC00166584
2,025
hottest_day_temp
100.945
USC00166584
2,025
cdd
2,991.427
USC00280734
2,019
annual_avg_temp
54.302
USC00280734
2,019
total_precip_in
45.275
USC00280734
2,019
hottest_day_temp
94.75
USC00280734
2,019
cdd
1,086.139
USC00280734
2,020
annual_avg_temp
54.369
USC00280734
2,020
total_precip_in
45.275
USC00280734
2,020
hottest_day_temp
94.75
USC00280734
2,020
cdd
1,093.307
USC00280734
2,021
annual_avg_temp
54.437
USC00280734
2,021
total_precip_in
45.275
USC00280734
2,021
hottest_day_temp
94.75
USC00280734
2,021
cdd
1,100.475
USC00280734
2,022
annual_avg_temp
54.504
USC00280734
2,022
total_precip_in
45.275
USC00280734
2,022
hottest_day_temp
94.75
USC00280734
2,022
cdd
1,107.642
USC00280734
2,023
annual_avg_temp
54.571
USC00280734
2,023
total_precip_in
45.275
USC00280734
2,023
hottest_day_temp
94.75
USC00280734
2,023
cdd
1,114.81
USC00280734
2,024
annual_avg_temp
54.638
USC00280734
2,024
total_precip_in
45.275
USC00280734
2,024
hottest_day_temp
94.75
USC00280734
2,024
cdd
1,121.978
USC00280734
2,025
annual_avg_temp
54.706
USC00280734
2,025
total_precip_in
45.275
USC00280734
2,025
hottest_day_temp
94.75
USC00280734
2,025
cdd
1,129.146
USC00286055
2,019
annual_avg_temp
54.834
USC00286055
2,019
total_precip_in
49.423
USC00286055
2,019
hottest_day_temp
97.411
USC00286055
2,019
cdd
1,192.792
USC00286055
2,020
annual_avg_temp
54.902
USC00286055
2,020
total_precip_in
49.423
USC00286055
2,020
hottest_day_temp
97.411
USC00286055
2,020
cdd
1,202.973
USC00286055
2,021
annual_avg_temp
54.969
USC00286055
2,021
total_precip_in
49.423
USC00286055
2,021
hottest_day_temp
97.411
USC00286055
2,021
cdd
1,213.154
USC00286055
2,022
annual_avg_temp
55.037
USC00286055
2,022
total_precip_in
49.423
USC00286055
2,022
hottest_day_temp
97.411
USC00286055
2,022
cdd
1,223.335
End of preview. Expand in Data Studio

ProtoStar Challenge

Team Placeholders: Protostar Climate Challenge Submission

Climate Risk Intelligence | Team Placeholders

Negative result: no

Forecasts of yearly weather for 210 U.S. cities, 2019–2025, made as of the end of 2018 (no data that reveals 2019–2025 weather), for the Protostar Challenge: Climate Risk Intelligence.

The ProtoStar Challenge, hosted by Carnegie Mellon University, explores how machine learning can improve climate forecasting using historical weather records from 210 U.S. cities. The challenge focuses on predicting four key climate indicators: average annual temperature, total annual precipitation, maximum daily temperature, and cooling degree days, translating historical climate data into actionable insights for urban resilience and adaptation.

As Team Placeholders, we are developing an iterative, data-driven forecasting pipeline that combines statistical baselines with machine learning models to improve prediction accuracy. Our workflow is built around Python and uv for reproducible dependency management, Google Colab for collaborative experimentation, and MLflow for tracking model performance, parameters, and artifacts. Through continuous evaluation and refinement, we aim to develop reliable city-level forecasts and translate our findings into a practical climate adaptation recommendation.

Team

Name Hugging Face username University, Program
Meet Bhatt Meet2304 Carnegie Mellon University, MSAIE-ETIM
Kohav Dey DragonDare Carnegie Mellon University, MISM
Ananya Gahalaut Vyaghra Carnegie Mellon University, MSAIE-IS
Achintya Gahalaut agahalau Carnegie Mellon University, MSAIE-IS

Method

Current version: a 50/50 blend of Placeholder_v9-S and combined model v3 (code/make_blend_submission.py). Each of the 5,880 forecasts is the average of the two models' forecasts, both made as of 2018-12-31.

  • Why a blend: the two models make different mistakes. v9-S (below) is a richer neighbour-graph model with a strong warming term; combined model v3 (code/combo_v3.py) is simple: for average temperature, each city's 2009–2018 mean carried forward at the median warming rate across all cities; for CDD, the same level growing at the median percent rate across cities (so cities with little cooling need get a proportionally small rise); for precipitation and hottest day, the 30-year average.
  • How we chose it (rule fixed before running): code/blend_test.py trains each model only on years up to an origin and forecasts the next 7 years, for origins 1986, 1994, 2002 and 2010, scored with the leaderboard formula on the rows the challenge would score. The blend would be submitted only if its mean Score beat both models; it did (reports/blend_test.csv). No 2019–2025 observation or leaderboard score was used.
Score (leaderboard formula) Origin 1986 1994 2002 2010 Mean
Placeholder_v9-S 0.00 1.70 4.11 12.23 4.51
Combined model v3 2.68 2.14 3.81 9.38 4.50
50/50 blend (submitted) 0.91 2.59 4.32 11.07 4.72
  • Honest limits: the margin is modest (+0.2), v9-S alone is best on the most recent origin (2010), and v9-S was itself selected on rolling origins 1990–2011, so some of these windows are not new to it. We kept the pre-declared rule.
  • Reproduce: python code/make_blend_submission.py runs code/make_submission.py (v9-S) and code/combo_v3.py (v3), averages them and writes submission.csv (MD5 c6b204fa935d53fc5cca8d88436a0f97); reports/submission_parts.csv keeps both components.

Component 1: Placeholder_v9-S (code/make_submission.py → code/placeholder_v9s.py). It is Placeholder_v3.8 (below) with one change, in CDD: a better estimate of how many °F-days CDD gains per °F of warming.

  • v9-S's CDD response: r = 0.75·k + 0.25·d. It replaces v3.8's response in both CDD terms below.
    • d (physical): the city's average number of days a year with a daily mean above 65 °F, over 1989–2018, valid CDD years only, from the challenge's daily data. A uniform +1 °F adds 1 °F-day on each such day.
    • k (statistical): v3.8's CDD-on-temperature regression, averaged over the city and the 15 cities nearest in both distance and d.
    • How 0.25 was picked: on origins ≤ 2004, leads 4–7.
    • Fallback: cities with fewer than 10 valid years keep v3.8's response.
    • Origin of the idea: it came from the planning council (plan v9, docs/plans/plan-v9-lineup.md), and two training stations built it independently.

Placeholder_v3.8, which v9-S keeps for everything except that response:

Metric Forecast for 2019–2025
Average temperature The city's mean of its last 15 years (2004–2018), carried forward at 1.75 × the warming rate shared by all cities (the median of the cities' 1969–2018 least-squares slopes). v3.6 and v4 used 1 ×. The 50-year median slope lags the recent warming rate; the multiplier was chosen on origins ≤ 2004 only
Cooling degree days v3.6's CDD, plus the city's CDD response to annual temperature (fitted on the city and its 15 nearest neighbours) times the extra temperature warming above. No new tuned number. v3.6's CDD is v4's forecast (last 10 years, shared relative 30-year rate) plus half of any shortfall against a warming trend: clip(v4 + 0.5 · max(trend − v4, 0), 0), where the trend is the city's 1989–2018 mean plus the same CDD response times the national 30-year warming rate
Total precipitation The city's 1989–2018 average plus Placeholder_v1.4's neighbour-graph term: a centred shift that moves a city relative to its neighbours, never the national level. Its two coefficients are refitted on pseudo-origins up to 2014, leads 4–7, with data up to 2018 only
Hottest day Same as precipitation: 1989–2018 average plus the graph term

Precipitation and CDD are clipped at 0.

Inputs:

  • the challenge annual_indices table up to 2018-12-31 (data/annual_indices_train_to_2018.parquet, official 10-day validity rule);
  • city coordinates from data/city_info.csv;
  • per city-year counts of days with both Tmax and Tmin and of days with a mean above 65 °F (data/daily_warm_days.csv), built by code/prepare_warm_days.py from the challenge daily subset up to 2018-12-31.

v9-S uses no external data, no pretrained model and no randomness (seed 42). code/make_submission.py alone rebuilds v9-S's forecasts byte for byte (SHA-256 d1873055…7e5f); the submitted file is the blend above.

Compared with v3.8: only CDD changes, by +2.4 °F-days on average (+1.9 in 2019 to +2.9 in 2025; per city −90 to +41). v3.8 itself was higher than v3.6 by 0.30 °F (2019) to 0.51 °F (2025) for temperature and by 34 to 56 °F-days for CDD.

How v9-S was chosen (November 9–10, plans v8 and v9):

  • Promotion rule (plan v8 §2): the final ranking scores 2022–2025 only, which is leads 4–7 from 2018. So a candidate must:
    • beat v3.8 at leads 4–7 on origins 2004–2011 and on origin 2011;
    • beat it on the late window (origins 2012–2014, leads 4–6) with a 90% city-bootstrap interval excluding 0;
    • lose at most 0.10 on 1990–2003;
    • have every tuned number chosen on origins ≤ 2004.
  • Plan v8 (rejected): six candidates, all re-run unchanged in the submission session (docs/plans/v8-analysis/v8_compare.txt). One retuned the warming rate, one computed CDD from daily temperatures, and the others changed the level window, used regional slopes or used homogenised input. None passed except a homogenised input that moved the 2018 forecast by only 0.009 °F.
  • Plan v9: an independent planning council of two separate reviewers (Codex gpt-6.1-sol and Grok grok-4.7) screened precipitation, hottest-day, new model-family and CDD ideas. Only the CDD response survived.
  • The two builds: the K station (Meet/train) and the 04_ii station (Meet/train-dl) built it independently. Both were re-run here unchanged, and they give identical forecasts. Comparison in docs/plans/v9-analysis/v9_compare.txt.
Leaderboard-style score (0–100) 2004–2011, leads 4–7 Late 2012–2014, leads 4–6 Origin 2011, leads 4–7 1990–2003, leads 4–7
Placeholder_v3.8 (previous submission) 9.675 18.754 20.060 4.772
Placeholder_v9-S (submitted) 9.863 18.924 20.264 4.793
Difference (90% city interval) +0.189 (+0.104 to +0.295) +0.170 (+0.094 to +0.254) +0.204 +0.021

Robustness and limits:

  • v9-S wins at all 8 origins in 2004–2011.
  • Its late-window interval stays above 0 under a 5° spatial-block bootstrap.
  • Leads 1–3 also gain about +0.10.
  • Known limits: the gain is small, about +0.2 points. The late window had been read in earlier plans. Both council members estimate only a 15–25% chance of a final score above 20, because the 2022–2025 weather matters far more than this change.

Placeholder_v3.8 (previous submission, rebuilt by code/make_submission_v3_8.py, SHA-256 7b799ac9…72e5), as described in the table above.

How v3.8 was chosen (November 9, plan v7, docs/plans/plan-v7-warming-rate.md). All earlier candidates shared v4's temperature forecast. Three training stations each built one change of plan v7: Placeholder_v3.7 (temperature rate × 1.75), Placeholder_v3.8 (v3.7 plus CDD following temperature), and two principled replacements for the multiplier (a recency-weighted rate and a quadratic trend). Every candidate was rerun unchanged through code/bench_adapters.py, on origins 1990–2011 and on a new late window, origins 2012–2017 (every lead ≤ 2018). Paired city bootstrap, 1,000 draws; full output in reports/bench_v38_selection.txt.

Leaderboard-style score (0–100) Origins 2004–2011 Origins 1990–2003 Origin 2011 Late 2012–2017 Leads 4–7, 2004–2011
Placeholder_v3.6 (previous submission) 7.41 4.00 10.68 12.92 8.50
Placeholder_v3.7 (temperature rate × 1.75) 7.54 4.09 11.88 14.22 8.95
v3.7 with a quadratic trend 7.56 4.10 12.20 14.21 9.06
v3.8 with rate × 1.5 7.97 4.05 12.04 14.34 9.41
Placeholder_v3.8 (rate × 1.75, submitted) 8.08 4.04 12.54 14.82 9.68
  • Promotion rule (plan v7 §4): better on 2004–2011, on origin 2011 and on the late window with a 90% interval excluding 0; at most 0.10 worse on 1990–2003; every tuned number chosen on origins ≤ 2004.
  • v3.8 against v3.7: +0.54 on 2004–2011, +0.60 on 2012–2017 (90% interval +0.37 to +0.81), +0.66 on 2011, −0.05 on 1990–2003. It passes. Against v3.6 the late-window gain is +1.90 (+1.45 to +2.34).
  • The multiplier: 1.75 maximises temperature skill on origins ≤ 2004 (v3.7's declared rule). The K station's rule (full points on origins ≤ 2004, ties within 0.01 to the smaller value) picked 1.5, which led 1.75 by 0.002. We used the plan's definition, v3.8 = v3.7 + CDD follow-through. The late window had already been seen when this was decided, so it is not a blind test of that choice.
  • Known weaknesses: v3.8 is much worse than v3.6 at origin 2007 (0.43 against 2.66 points) and slightly worse at origins 2002–2003; its CDD skill in 1990–2003 drops from 1.80% to 1.66%. It is a bet that warming continues at about the 2005–2018 pace.

Placeholder_v3.6 (previous submission, rebuilt by code/make_submission_v3_6.py, SHA-256 b2a69a69…1494): v4 with the half-warmth CDD term above and the neighbour-graph term on precipitation and hottest day; temperature as v4.

How v3.6 was chosen (November 9). Three training stations independently built "v4 plus the graph term on precipitation and hottest day" (Placeholder_tf1, plan I's K1 and Placeholder_v3.5); all three give identical scores. Every candidate was rerun, unchanged, through code/bench_adapters.py (origins 1990–2011, data ≤ 2018, paired city bootstrap with 1,000 draws):

Leaderboard-style score (0–100) Origins 2004–2011 Leads 1–3 Leads 4–7 Origins 1990–2003 Origin 2011
Combined model v4 (previous submission) 6.53 4.92 7.58 3.50 10.44
tf1 = K1 = v3.5 (v4 + graph term) 6.83 5.26 7.82 3.89 10.51
tf3 (tf1 + regional warming rates) 6.83 5.24 7.87 3.80 10.55
tf10 (tf1 + boosted hottest-day correction) 7.02 5.28 8.22 3.81 10.88
K3 (CDD half v4, half Placeholder_v0.3)* 7.23 4.25 9.75
Placeholder_v3.6 (submitted) 7.41 5.73 8.50 4.00 10.68

* Plan I's own run.

  • v3.6 against tf1/K1/v3.5: +0.58 on 2004–2011 (90% interval +0.45 to +0.72), +0.11 on 1990–2003, and +0.17 on 2011 (+0.06 to +0.30). It is the only candidate that passes the three-part rule against them.

  • v3.6 against tf10: +0.39 on 2004–2011 (+0.25 to +0.53) and +0.19 on 1990–2003 (+0.07 to +0.33). It loses on 2011 by 0.21.

  • tf10 fails the rule against tf1 on 1990–2003. Its gain comes from a boosted learner tuned on this period.

  • Nested check: picking on origins ≤ 2004 alone also gives v3.6.

  • Mean skill over origins 2004–2011:

    Metric v4 v3.6
    Temperature 12.1% 12.1%
    Precipitation 0.0% 0.5%
    Hottest day 0.0% 0.7%
    CDD 14.0% 16.4%
  • Comparison files: reports/bench_v36_selection_table.csv, reports/bench_v36_selection_pairs.csv and reports/bench_v36_selection_per_origin.csv.

  • Development history: the DL station's screens P–U on Meet/train-dl (docs/models/placeholder-dl.md).

Known weakness of v3.6: the DL station rated its robustness "thin".

  • Its 1990–2003 margin depends on origins 2000–01.
  • The 0.5 weight matters: at weight 1, 1990–2003 drops to 3.62.
  • All of 1990–2011 has now been used for development, so no untouched pre-2019 test period remains.

Combined model v4 (previous submission, rebuilt by code/make_submission_v4.py):

  • Temperature and CDD: as above, without the CDD trend term.
  • Precipitation and hottest day: the city's 1989–2018 average.

Why v4 replaced v1.4, and why Placeholder_v1.4 was a mistake. We chose Placeholder_v0.1–v2.7 on three historical forecasts made as of 1983, 1990 and 1997.

  • The test windows favoured the average. Warming was weak in those windows, because the 30-year windows include mid-century cooling. "Stay close to the 30-year average" won there, so every Placeholder version stayed anchored to it (v1.4 adds at most +0.5 °F).
  • The real forecast is different. It is made from 2018, in a period of strong warming.
  • The new test bench (code/tournament.py) scores every model the leaderboard's way from every origin 1990–2011:
    • each origin forecasts the next 7 years: leads 1–3 resemble 2019–2021, leads 4–7 resemble 2022–2025;
    • every origin and test year is inside the training data;
    • the origins nearest 2018 (2004–2011, test years 2005–2018) get the most weight.

Selection rule. Everything uses data up to 2018. No 2019–2025 observation and no leaderboard score was used to choose.

  • A metric keeps the choice from Kohav's combined model v3 unless a candidate beats it on all three checks: mean skill over origins 2004–2011, mean skill over origins 1990–2003, and skill on origin 2011 alone.
  • Only average temperature changed: a 15-year level instead of 10 years, and a 50-year shared slope instead of 30 years.
  • Candidates tried for each metric:
    • levels: the 30-year mean, the mean of the last 5, 10 or 15 years, a level built from the national anomaly plus half the city's own, and medians over 10–30 years (MAE is minimised by the median);
    • slopes: none, the city's own 30-year slope, shared 30-, 50- and 70-year slopes, and a shared relative slope;
    • each level and slope pair alone and mixed 50/50 with the 30-year mean.
Leaderboard-style score (0–100, mean over origins) Origins 2004–2011 Leads 1–3 Leads 4–7 Origins 1990–2003 Origin 2011
30-year average 0.00 0.00 0.00 0.00 0.00
30-year linear trend 2.14 3.06 2.02 0.46 2.31
10-year mean (Ocarina of Ten) 3.67 3.16 4.36 1.81 5.48
Placeholder_v1.4 (previous submission) 3.56 3.31 3.65 4.01 3.17
Combined model v1 5.05 4.47 5.77 3.27 7.59
Combined model v2 5.61 4.77 6.62 3.58 9.33
Combined model v3 6.03 4.93 7.27 3.58 10.10
Combined model v4 6.53 4.92 7.58 3.50 10.44

Mean MAE over origins 2004–2011:

  • Average temperature: v4 1.281 °F; v3 1.316; v1.4 1.382; 30-year average 1.463.
  • CDD: v4 and v3 170.1; v1.4 183.1; 30-year average 197.9.
  • Precipitation and hottest day: v4 equals the 30-year average (6.018 in and 2.685 °F).

The full tables are reports/tournament_leaderboard.csv, reports/tournament_mae_recent.csv and reports/tournament_metric_summary.csv. Every candidate, origin and metric is in reports/tournament_scores.csv.

Limits.

  • Overlapping origins. Each origin tests 7 years, so origins 1990–2011 hold only about three independent periods.
  • Why the rule is conservative. Picking each metric's best candidate on origins 2004–2011 alone would have scored 7.03 there. But the same procedure, run on origins up to 2004, scored worse than v3 on origin 2011 (7.3 vs 10.1). Maximising each metric over many candidates overfits.
  • Precipitation and hottest day stayed at the 30-year average in v4 because no candidate in that family beat it consistently. v3.6 adds the graph term instead.
  • Not used as forecast inputs: station-change information, ENSO and other outside data (tested; see Outside data we tested below).

Earlier versions.

  • Placeholder_v0.1–v2.7: neighbour-graph and shrinkage models. See code/placeholder_v1_4.py, code/make_submission_v1_4.py, docs/models/placeholder.md and submissions/Placeholder_v1.4/.
  • Combined models v1–v3: code/combo_model.py; v2 and v3 are on Kohav's kohav branch.
  • The challenge's starter random forest: code/starter_model.py; worse than the 30-year average on three of the four metrics.

Outside data we tested

El Niño adjustment, tested and left out (USE_ENSO = False in code/combo_v3.py; the 38-origin test script code/enso_test.py is on Kohav's kohav branch, commit 4cbbf9e; results in reports/enso_rolling_test.csv). We joined NOAA's Oceanic Niño Index (ONI) to the city records by year and shifted each city's first forecast year by its El Niño sensitivity (fitted on earlier years only, averaged over the city and its 8 nearest cities). Over 38 one-year-ahead forecasts (1981–2018, reports/enso_rolling_test.csv) it cut first-year errors by about 1.2% for temperature and 1.5% for CDD, but it was better in only 21 and 23 of the 38 years, and it lowered the 4-window backtest Score from 5.4 to 5.1 (big misses after strong La Niña years, e.g. the forecast for 2011). By our selection rule it stays out. Details: Data and Feedback to the Ecosystem.

Fixed geographic traits (vegetation, soil, distance from the sea, relief, latitude): cities that warmed fastest in 1959–1988 were not the fastest in 1989–2018 (correlation of city warming rates −0.08; latitude vs 1989–2018 temperature trend −0.03), so a trait that does not change cannot predict which cities warm faster. We did not use them as forecast inputs; tree canopy and cool surfaces appear in the policy idea instead.

AI Tools and Agents

  • Tools and models: Claude Code (Anthropic).
  • What they did: read the challenge rules and submission template; set up the repository layout, code/AGENTS.md / CLAUDE.md, the data-rule guard, prepare_data.py, starter_model.py (from the challenge's starter code) and the backtest; drafted this README and the data card. Downloaded only the challenge training dataset (data up to 2018).
  • Combined model: Claude (Anthropic, via the Claude app) wrote code/combo_model.py, ran the four-window backtest and drafted this Method section, working only from data/annual_valid.csv (data up to 2018).
  • Exploration (notebooks/Meet/01-dataset-structure.ipynb): Claude Code (Claude Opus 5.5) wrote the dataset-structure and data-quality notebook, working only from the challenge training data (up to 2018). A person should review the cleaning rules in its section 10 before they are used for a model.
  • Cleaning pipeline (notebooks/Meet/02-cleaning-daily-annual.ipynb, notebooks/Meet/meet_cleaning.py): Claude Code (Claude Opus 5.5) designed and wrote the cleaning of the daily and annual_indices subsets: deleting impossible and stuck readings, checking outliers against neighbouring cities, rebuilding the yearly table from cleaned days, filling missing days and empty years for inputs only, adjusting for station changes, and a four-origin backtest (all years up to 2018) to choose which inputs to use. It used only the challenge training data (up to 2018); submission.csv was not changed. A person should review the isolated outliers in notebooks/Meet/qc/daily_outlier_decisions.csv, the thresholds at the top of meet_cleaning.py, and the station-change adjustment before it feeds a model.
  • Cleaning of all four subsets (November 8): Claude Code (Claude Opus 5.5, through T3 Code) planned the cleaning of daily, annual_indices, city_info and coverage_and_stations; had the plan reviewed over four rounds by Sol (GPT-6.1-sol) and Grok (Grok 4.7) until both rated it at least 90 (final 95 and 97; notebooks/Meet/plan/); and implemented it as scripts/02_clean_data.py (logic in src/protostar_challenge/cleaning.py, 17 tests in tests/test_cleaning.py), writing data/processed/. Documented step by step in documentation/. Only the challenge training data (up to 2018) was used; every backtest fold was rebuilt from data up to its origin; official labels are unchanged (= data/annual_valid.csv); submission.csv was not changed. A person should review the isolated outliers in data/qc/daily_outlier_decisions.csv, the adopted station steps in data/processed/qc/homog_steps.csv, and the input choice (only Avg Temp ships a cleaned input; documentation/06-input-selection.md) before a model uses it.
  • Final blend and outside data (November 7–10): Claude (Anthropic, via the Claude app) wrote code/prepare_ocean_indices.py (NOAA/JISAO indices from Internet Archive copies captured on or before 2018-12-31), code/enso_test.py (on Kohav's branch), code/prepare_population.py (2020 Census via the Census Geocoder), code/policy_cdd.py, code/blend_test.py and code/make_blend_submission.py; ran the El Niño and blend tests; and drafted the outside-data, blend and policy-number parts of this README. Kohav Dey ran every script on his machine, checked the outputs and the MD5 of submission.csv, and decided the pre-declared blend rule.
  • What a person reviewed: Meet Bhatt directed every agent task, set the data rules and stopping points, chose the decision rules (the reward ladder), reviewed the head-to-head results that picked Placeholder_v1.4, and approved generating, uploading and submitting this forecast on 2026-11-08. The model code and the agent-drafted Policy Idea have not yet had a line-by-line human review.
  • External-data research (November 7): Codex (GPT-6.1-sol, through T3 Code) reviewed the official challenge rules and rubric, the current model and stored backtests, and primary documentation for adjacent climate datasets; wrote reports/adjacent-datasets-research.md and a judging-weight visualization. No external weather observations were downloaded, no new forecasting experiment was run, and no submission values were changed. A person should review source eligibility, licenses, and the proposed experiments before implementation.
  • Internal evaluation platform (November 7): Codex (GPT-6.1-sol, through T3 Code) built the historical evaluation CLI, model adapters, run registry, local HTML report, rubric self-review and contract tests. It used the already-downloaded pinned training data through 2018, verified the published scoring source and FAQ, and ran the reference methods and current combo model on five seven-year historical tasks. No post-2018 observations were downloaded, no model was selected from official-year scores, and submission.csv was not regenerated. A person should review model adapters for origin-specific leakage, the score-mask interpretation and the limits of reused historical holdouts. See evaluation/README.md.
  • Opus-directed evaluator fixes: Sol (GPT-6.1-sol through T3 Code, including a Sol helper for the nested selector) implemented the Opus audit's origin-aware guards, ranking on unclipped skill, nested method selection, separate development/assessment roles with committed freezes and access logs, period-level comparisons, explicit parameter/timeout failures, external-prediction provenance and run integrity checks. The new default suite remains labeled as reused historical evidence. Tests and real development runs are recorded in reports/evaluation-opus-fixes.md. No model was promoted to the official submission and submission.csv was not rewritten. A person should review the selected procedure, commit intended source before freezing assessment, and retain the exact artifacts with any performance claim.
  • Control baselines, scorecard and MLflow tracking (November 8): Claude Code (Claude Opus 5.5) wrote a plan, had it reviewed over five rounds by Sol (GPT-6.1-sol) and Grok (Grok 4.7) until both rated it above 90 (docs/plans/baselines-and-tracking-plan.md), and implemented it: seven fixed control baselines (code/evaluation/controls.py), a per-run characteristic scorecard (scorecard.py), a verified, reusable control panel (panel.py), an MLflow mirror of verified runs (tracking.py), the one-command loop scripts/eval_model.py, and 26 tests. The evaluator's owner (Sol, through T3 Code) integrated and committed it and built the first panel. Only the challenge training data (up to 2018) and development origins were used; no assessment scores were opened and submission.csv was not changed. A person should review the control definitions, the gates and the first results in docs/evaluation-baselines.md.
  • Placeholder_v0.1 model training (November 8): Claude Code (Claude Opus 5.5, through T3 Code) planned and wrote the first Placeholder model (notebooks/Meet/placeholder_v0_1.py, its tests, and notebooks/Meet/03-placeholder-v0.1-training.ipynb) and the model card docs/models/placeholder.md. Grok 4.7 and Cursor Composer 2.5 reviewed the plan (docs/plans/placeholder-training-plan.md) over four rounds, ending at 93/100 each. It used only the suite's history up to each development origin (1983, 1990, 1997) and city coordinates; no post-2018 data was used, and submission.csv was not changed. A person should review the shrinkage rule, the pre-run inner-fold look described in the plan, and whether any later version earns a freeze and assessment.
  • Placeholder_v0.2–v2.7 and the submitted forecast (November 8): Claude Code (Claude Opus 5.5, through T3 Code) wrote every later Placeholder version (plans A and B, reviewed by Sol (GPT-6.1-sol) and Grok (Grok 4.7): 92/94 and 93/93) and scored each on development origins 1983/1990/1997 only; v1.4 became the champion and v2.7 was worse. A separate Claude Code session (same model) re-scored v1.4 and v2.7 on suite-v3-r5 as labelled diagnostics (same ranking: the two suites hold the same data), moved v1.4 into code/placeholder_v1_4.py (byte-identical to the scored adapter), wrote code/make_submission.py, generated submission.csv from it (trained on the challenge data up to 2018-12-31; no 2019+ data), ran both validators, and drafted this Method section, the Policy Idea and the Feedback items. No value was written by hand, no model was chosen on 2019+ data or the leaderboard, and the one-shot assessment origins (2004, 2011) were not opened. A person should review the Policy Idea (agent draft), code/make_submission.py against evaluation.core.build_fold, and the evidence in submissions/Placeholder_v1.4/DISCLOSURE.md.
  • Combined model v4, the new test bench and the v4 submission (November 9): Placeholder_v1.4's live score fell below the linear-trend baseline. Meet then asked Claude Code (Claude Opus 5.5, through T3 Code) to find the best model and submit it overnight.
    • Use of the leaderboard: it only prompted a re-examination of the test design. It was never used to choose between models.
    • What Claude did: wrote code/tournament.py (rolling origins 1990–2011, all inside the training data); scored the baselines, Kohav's combined models v1–v3, Placeholder_v1.4 and a family of simple candidates; fixed a conservative selection rule; wrote code/make_submission.py (now code/make_submission_v4.py) for the winner (combined model v4); generated and validated submission.csv; rewrote this Method section; uploaded the dataset; and pressed Submit in the Space, as Meet had authorised.
    • Data: no 2019+ data was read, and no value was written by hand.
    • For a person to review: the selection rule, the tournament's candidate family, the decision to change only the temperature forecast, and whether the team agrees with replacing v1.4 (v4 is built on Kohav's v3).
  • Placeholder_v3.6 and this submission (November 9): the training stations, each running Claude Code (Claude Opus 5.5, through T3 Code), built candidates on the new bench. Sol (GPT-6.1-sol) and Grok (Grok 4.7) critiqued them, and each screen was declared before it was run.
    • The stations' candidates: the TF series (Meet/train-static-v1), plan I's K series (Meet/train), and Placeholder_v3.5/v3.6 (Meet/train-dl; v3.6's CDD rule was proposed by Grok).
    • What the submission session did: reran every candidate unchanged through code/bench_adapters.py and compared them with a paired city bootstrap (reports/bench_v36_selection_*.csv). It found that tf1, K1 and v3.5 are the same model, and that v3.6 is the only candidate passing the three-part rule against them. It then copied the adapter into code/placeholder_v3_6.py, wrote code/make_submission.py, and verified submission.csv against the bench harness's own 2018 export and the DL station's file (SHA-256 b2a69a69…, identical). Finally it updated this README, uploaded the dataset and pressed Submit in the Space, at Meet's request.
    • Who chose: Meet chose v3.6 over tf10 after seeing the comparison.
    • Data and selection: no 2019+ data was read, no value was written by hand, and the leaderboard was not used to choose.
    • For a person to review:
      • v3.6's "thin" robustness label: the 0.5 weight and origins 2000–01;
      • that all of 1990–2011 is now reused development data;
      • the CDD trend term in code/placeholder_v3_6.py.
  • Placeholder_v9-S and this submission (November 9–10, plans v8 and v9): at Meet's request, the submission session (Claude Code, Claude Opus 5.5, through T3 Code) did the following:
    • wrote plan v8 (docs/plans/plan-v8-final-leads.md), which judges candidates on leads 4–7 because the final score covers only 2022–2025;
    • re-ran the stations' plan v8 candidates unchanged and found that none qualified;
    • started a planning council with no access to the repository: Codex gpt-6.1-sol and Grok grok-4.7, briefed separately, read-only, data ≤ 2018. Their records are in docs/plans/v9-council/.
    • What the council found: Sol proposed the CDD response. Grok tested new model families (analogues, a direct ridge, regime-chosen rates) and showed that they under-warm.
    • The builds: the K station and the 04_ii station (Claude Code, Claude Opus 5.5) built v9-S independently, each committing its selection rule before running.
    • What the submission session then did:
      • re-ran both builds, which gave identical forecasts;
      • copied 04_ii's adapter into code/placeholder_v9s.py, changing only the input path;
      • added code/prepare_warm_days.py and data/daily_warm_days.csv;
      • verified that submission.csv equals the bench harness's 2018 export;
      • updated this README, uploaded the dataset and pressed Submit, at Meet's request.
    • Data and selection: no 2019+ data was read, no value was written by hand, and the leaderboard was not used to choose.
    • For a person to review:
      • the warm-day count as a CDD response;
      • the fallback for cities with fewer than 10 valid years;
      • that the late window had been read in earlier plans.
  • Placeholder_v3.8 (November 9, plan v7): after v3.6's live score, Meet asked the submission session (Claude Code, Claude Opus 5.5, through T3 Code) to find out why and how to improve. It ran checks on data ≤ 2018 only, found that every candidate shared v4's temperature forecast, whose warming rate lags, wrote plan v7 (docs/plans/plan-v7-warming-rate.md, with a new late test window, origins 2012–2017) and sent one task to each training station.
    • The stations' candidates: Placeholder_v3.7 (Meet/train-dl), Placeholder_v3.8 (Meet/train), and the principled-slope variants (Meet/train-static-v1), each built by Claude Code (Claude Opus 5.5) with its multiplier or weight declared and chosen on origins ≤ 2004.
    • What the submission session did: reran all of them unchanged (docs/plans/v7-analysis/v7_compare.py, reports/bench_v38_selection.txt); found that the two v3.7 builds are identical and that v3.8 at 1.75 × passes the rule against v3.7; copied the K station's adapter into code/placeholder_v3_8.py with the multiplier set to 1.75; pointed code/make_submission.py at it; and verified submission.csv against the bench harness's own 2018 export of the station's adapter (maximum difference 0). It then updated this README, uploaded the dataset and pressed Submit in the Space, at Meet's request (2026-11-09 20:06 UTC).
    • Who chose: Meet asked for v3.8 to be submitted after seeing the comparison.
    • Data and selection: no 2019+ data was read, no value was written by hand, and the leaderboard was not used to choose.
    • For a person to review: the choice of 1.75 over the K station's 1.5 (explained under Method); that the late window 2012–2017 was used to screen ideas before the candidates were built; the CDD follow-through in code/placeholder_v3_8.py.
  • Agent instructions file: code/AGENTS.md (the single agent instructions file; Claude Code loads it through CLAUDE.md at the repository root).

Policy Idea

  1. What risk did we notice? Cities plan cooling (cooling centres, heat-alert staffing, school and transit air conditioning, utility peak capacity) against 30-year "normals". Our backtests show that a 30-year average lags a warming climate: adding half of the gap between the last 10 years and the 30-year mean was one of the two changes that made our model clearly beat it, and our submitted forecast puts 2019–2025 above the 1989–2018 normal in every one of the 208 scored cities for both average temperature (median +1.1 °F) and cooling degree days (median +185 °F-days a year, +16%). Budgets sized to the old normal will be short in exactly the hot years they exist for, and the shortfall is largest where neighbouring cities show the same drift.
  2. What would we do about it, and who would do it? Each city's sustainability or emergency-management office would publish, every January, a one-page "cooling outlook": its CDD and average-temperature forecast for the next seven years (from public station records, as in this challenge), the gap to its 30-year normal, and the same for its neighbours within about 600 km. Cooling-centre budgets, heat-plan staffing and utility demand planning would be sized to the outlook instead of the normal. A regional council of governments or the state climatologist would run one shared script for all its cities, so small cities get it without staff time.
  3. Why is it sustainable? It needs only free, already-published daily station records and a small open-source script that runs in seconds; no new sensors or paid data. It updates itself each year as a new year of records arrives, and it is checked against reality every year: last January's outlook is compared with the observed year, so cities can see whether it beats the old normal before relying on it more.

Where it matters most (our CDD forecasts × 2020 Census, code/policy_cdd.py, reports/policy_cdd_by_area.csv): across the metro areas (or counties) of the 210 stations, 216 million people, the population-weighted rise is +266 °F-days a year. The 10 areas with the largest total extra cooling need (population × rise) hold 85.8 million people (40%): New York, Houston, Dallas–Fort Worth, Miami, Phoenix, Washington DC, Los Angeles, Atlanta, Riverside–San Bernardino and Chicago, so the outlook should start there. The fastest relative rises are in cooler cities that have planned least for heat (Reno +35%, Colorado Springs +34%, San Diego +32%, Muskegon and Burlington +29%), where the outlook matters most for cooling centres and building rules. Pairing it with heat-island measures (cool roofs, cool pavements, tree canopy; EPA Heat Island Reduction) lowers the cooling demand itself rather than only budgeting for it.

Feedback to the Ecosystem

  • What got in the way: the challenge dataset is served from a CDN host (us.aws.cdn.hf.co) separate from huggingface.co, so sandboxed agent environments with a host allowlist failed to read the parquet files until that host was added. scikit-learn's RandomForestRegressor with n_jobs=-1 gives slightly different predictions between runs (summing trees in thread order), which breaks byte-for-byte reproducibility unless prediction runs on a single thread.
  • What would fix it: list every host the dataset is served from on the dataset card, for teams running agents behind an allowlist; mention the parallel-prediction issue next to the starter code.
  • Getting other agencies' data as of 2018: the climate indices we needed (NOAA CPC's ONI, JISAO's PDO, NOAA PSL's AMO) are published only as files that run to the present, with no way to request data up to a date, so a team that follows the data rule strictly cannot call them. We used Internet Archive copies captured before 2019, which worked, but the archive coverage is uneven: NOAA PSL's own ONI and PDO files were last captured in early 2017, so we had to find other publishers (CPC, JISAO) with captures from December 2018. The indices also come in different layouts (PSL year-by-12-months tables with a missing-value line, CPC 3-month seasons, JISAO with ** markers and partial years), so each needs its own parser.
  • Joining people to stations: the stations are identified only by GHCN ID, name and coordinates, with no state or Census code, and most sit at airports outside the city they are named for (55 of 210), so a city-name join fails; we had to geocode the coordinates. The Census Bureau's data API now requires a key, while its Geocoder (no key) returns 2020 populations directly, which is what made this quick.
  • Joining people to stations: the stations are identified only by GHCN ID, name and coordinates, with no state or Census code, and most sit at airports outside the city they are named for (55 of 210), so a city-name join fails; we had to geocode the coordinates. The Census Bureau's data API now requires a key, while its Geocoder (no key) returns 2020 populations directly, which is what made this quick.
  • What would fix it: a snapshot of common climate indices (ONI, PDO, AMO, NAO) cut at 2018-12-31 next to the challenge dataset, in one tidy layout (index, year, month, value); and each city's state and Census place/metro code in city_info, so teams can join social and economic data without geocoding.
  • Reused history is the main evaluation risk: with only 1851–2018 to learn from, every team tests on the same few historical periods many times, so backtest gains are optimistic. Publishing a few fixed, pre-2019 held-out periods (with the answers hidden and a small daily scoring budget) would let teams check a model without touching 2019–2025.
  • Hard-to-beat metrics: on total precipitation and hottest day, almost nothing beats the 30-year average, so those two metrics mostly add noise to the score; saying so in the rules (as the hosts partly do) helps teams spend time where skill is possible.

Files

File Content
submission.csv Our forecasts: city_id,year,metric,predicted_value
code/ Scripts that rebuild submission.csv (see code/README.md) and the agent instructions
data/ The curated data our code uses; described in its data card, data/README.md
reports/ Backtest results written by the code: bench_v9s_selection.txt (the comparison that chose the submitted model), bench_v38_selection.txt and bench_v36_selection_*.csv (earlier ones), tournament_*.csv (v4's tournament), placeholder_v1_4_selection.json and older backtests
submissions/Placeholder_v1.4/ Evidence for the previous model (Placeholder_v1.4): run ids, per-origin and per-metric scores, comparisons, and the full disclosure
reports/blend_test.csv, reports/submission_parts.csv The pre-declared blend test and both components of the submitted forecasts
reports/policy_cdd_by_area.csv, reports/policy_cdd_by_city.csv, reports/enso_rolling_test.csv Policy analysis (forecast CDD × 2020 Census) and the El Niño test
notebooks/ Each team member's exploration notebooks (not needed to reproduce submission.csv)
documentation/ Data cleaning documentation: index of every flaw found in the four training subsets and the strategy applied to it (daily, annual_indices, city_info, coverage_and_stations, station adjustment, input-selection backtest, outputs, review history)
scripts/ Data pipeline: 01_download_data.py (raw data → data/raw/), 02_clean_data.py (cleaning → data/processed/, see documentation/)

Metric codes: annual_avg_temp (°F), total_precip_in (inches), hottest_day_temp (°F), cdd (°F-days). City IDs: the station ID from the city_info table of the training dataset (e.g., Pittsburgh = USW00094823).

How to Reproduce

For rapid internal model testing, start with the evaluation platform guide. uv run python code/evaluate.py prepare builds version-2 historical tasks with declared phases; uv run python code/evaluate.py run --name my-model evaluates nested method selection on development origins; uv run python code/evaluate.py report writes the local experiment dashboard. Teammates can instead score their own CSV predictions or plug in a Python adapter. Imported predictions require a provenance manifest. Assessment runs require a committed model freeze; all current historical periods remain reused evidence. The official FAQ prohibits choosing models by scoring variants against public 2019–2025 observations, so this platform's internal labels end by 2018.

  1. Install Python 3.13 and the packages: pip install -r code/requirements.txt (or uv sync and prefix the commands below with uv run).
  2. python code/make_blend_submission.py builds the submitted forecasts (the 50/50 blend; it runs both scripts below) and writes submission.csv (MD5 c6b204fa935d53fc5cca8d88436a0f97). python code/blend_test.py reruns the test that chose it. Outside data: python code/prepare_ocean_indices.py, python code/prepare_population.py, then python code/policy_cdd.py. python code/make_submission.py builds component 1 (Placeholder_v9-S, code/placeholder_v9s.py) from data/annual_indices_train_to_2018.parquet, data/city_info.csv and data/daily_warm_days.csv and writes submission.csv byte for byte (SHA-256 d1873055…7e5f). python code/prepare_warm_days.py rebuilds data/daily_warm_days.csv from the challenge daily subset. It takes about 10 seconds and needs no network. python code/bench_adapters.py run --name X --adapter FILE:FUNC and summary rerun the bench; python code/tournament.py --with-v14 reruns v4's tournament.
  3. python code/check_submission.py: checks submission.csv like the challenge Space.

Earlier models: python code/make_submission_v3_8.py (Placeholder_v3.8, the previous submission); python code/make_submission_v3_6.py (Placeholder_v3.6); python code/make_submission_v4.py (combined model v4); python code/make_submission_v1_4.py (Placeholder_v1.4); python code/prepare_data.py rebuilds data/annual_valid.csv and data/city_info.csv; python code/combo_model.py (previous submission) and python code/starter_model.py write their backtests to reports/ and overwrite submission.csv.

Data

Analysis only: a preregistered pre-2019 station benchmark supplements the reused city backtests; see selection, provenance and limits. It does not feed the submitted forecasts.

  • External data (all capped at 2018; details and provenance in data/README.md and data/ocean_indices_sources.csv):
    • Oceanic Niño Index (ONI), NOAA Climate Prediction Center (cpc.ncep.noaa.gov/data/indices/oni.ascii.txt). Forecast input candidate, tested and not used in the submitted forecasts (see Outside data we tested).
    • Pacific Decadal Oscillation (PDO), University of Washington JISAO (research.jisao.washington.edu/pdo/PDO.latest; Mantua et al., 1997). Analysis only.
    • Atlantic Multidecadal Oscillation (AMO, unsmoothed), NOAA Physical Sciences Laboratory (esrl.noaa.gov/psd/data/correlation/amon.us.data, now psl.noaa.gov; Enfield et al., 2001). Analysis only.
    • 2020 Census population around each station (county, metro area and city), U.S. Census Bureau Geocoder (geocoding.geo.census.gov, vintage Census2020_Current), via code/prepare_population.py → data/city_population.csv. Analysis only (policy idea). Joined to the challenge data by each station's latitude/longitude: all 210 stations matched; 155 lie inside city limits, the rest (mostly airports) just outside, so we use the metro area (165 stations) or, outside metro areas, the county (45). Ten areas hold more than one station; area totals count each area once.
    • How we kept them inside the data rule: the live files run to the present and cannot be cut at 2018, so code/prepare_ocean_indices.py never calls them. It downloads the Internet Archive's newest copy of each file captured on or before 2018-12-31, so no 2019+ value can be in the download, and checks every row with assert_no_post_cutoff(). This also means we use the values as published at the time, before later revisions.
    • How they join the challenge data: by year. The indices are one monthly series for a whole ocean, the challenge data one value per city and year, so we summarise each index at the end of a year (ONI: mean of its Aug–Oct and Sep–Nov seasons) and relate it to each city's next year.
    • What we found (1951–2015, next-year departure of each city from its previous 10 years): the year-end ONI relates to the next year's temperature in northern cities (median correlation about +0.27; El Niño winters are mild in the northern U.S.) and weakly to precipitation in southern cities (about +0.14); the PDO relates weakly (about +0.18 for northern temperature, +0.13 for southern precipitation), and the AMO hardly at all (|median r| ≤ 0.11). Index values are the year-end mean of months 9–10 (for ONI, the Aug–Oct and Sep–Nov seasons).
  • Forecast input: data/annual_indices_train_to_2018.parquet, the challenge's annual_indices subset, byte-identical to the pinned revision below (SHA-256 0b5f5fa9…e411). data/city_info.csv gives the list of 210 cities.
  • Curated data: see the data card in data/README.md (each file, its source, license and preparation).

Training data: CMU-OSPO/protostar-climate-train-to-2018, from:

Submitting (team checklist)

  1. The final submission's public Hugging Face dataset: DragonDare/protostar-placeholders-final (Meet's Meet2304/protostar-placeholders plus the blend and outside data).
  2. Upload this repository's submission files: hf upload Meet2304/protostar-placeholders . --repo-type dataset --include "README.md" "submission.csv" "code/*" "data/*" "reports/*" "assets/*" "submissions/Placeholder_v1.4/*" "docs/models/placeholder.md" "evaluation/README.md".
  3. A registered team member pastes the dataset link in the Submission tab of the challenge Space (up to 10 per day; the latest counts).
  4. Record the dataset commit that was submitted, and keep submission.csv here in sync with it.

Made on Earth, by Humans.

Downloads last month
83