# Setup — all paths relative to the repo root (run this notebook from there).
import json, warnings
from pathlib import Path
import numpy as np, pandas as pd
import matplotlib.pyplot as plt
from matplotlib import colors as mcolors
warnings.filterwarnings("ignore")
plt.rcParams.update({"figure.dpi": 110, "font.size": 10, "axes.grid": False})

CA = Path("cluster_results/ca")
RES = Path("cluster_results")
INDEX = json.load(open(CA / "index.json"))
PROV  = json.load(open(CA / "provenance.json"))

# row_format from the shipped index — the contract every webapp_data_*.json obeys.
COLS = INDEX["row_format"]            # [lat,lon,discover,conservation,env,staleness,urgency,travel_min,n_train]
AXES = ["discover", "conservation", "env", "staleness", "urgency"]
print("row format :", COLS)
print("axes       :", AXES)

def load_group(group="All biodiversity"):
    d = json.load(open(CA / f"webapp_data_{group.replace(' ', '_')}.json"))
    rows = d[list(d)[0]]
    return pd.DataFrame(rows, columns=COLS)

ALL = load_group("All biodiversity")
print(f"loaded {len(ALL):,} cells × {len(COLS)} columns for 'All biodiversity'")

# A reusable Canada scatter-map used throughout. Cells are points on the 0.25° grid;
# small squares read as a surface. ASPECT ~ 1/cos(59°N) keeps Canada from looking stretched.
LON, LAT = ALL["lon"].to_numpy(), ALL["lat"].to_numpy()
ASPECT = 1.7
def canada_map(ax, values, title, cmap="magma", vmax=None):
    ax.scatter(LON, LAT, c=values, s=2.0, marker="s", cmap=cmap, vmin=0, vmax=vmax, linewidths=0)
    ax.set_title(title, fontsize=10)
    ax.set_xticks([]); ax.set_yticks([]); ax.set_aspect(ASPECT)
    for s in ax.spines.values(): s.set_visible(False)

ALL.head(3)

row format : ['lat', 'lon', 'discover', 'conservation', 'env', 'staleness', 'urgency', 'travel_min', 'n_train']
axes       : ['discover', 'conservation', 'env', 'staleness', 'urgency']
loaded 31,804 cells × 9 columns for 'All biodiversity'

bbox = INDEX["bbox"]
print(f"Grid          : {INDEX['n_cells']:,} land cells, {INDEX['res']}° (~25 km)")
print(f"Extent        : lon {bbox[0]}…{bbox[2]}, lat {bbox[1]}…{bbox[3]} (all of Canada + a US land sliver)")
print(f"Life groups   : {len(INDEX['groups'])} → {', '.join(INDEX['groups'])}")
print(f"Per cell      : 5 axes (0–1) + travel_min + n_train")
print(f"Provenance    : {PROV['manifest_hash'][:16]}…  ({len(PROV['files'])} files frozen)")

# Verify the axes are REAL in the actual data (not from the doc, not from stale metadata):
print("\nAxis reality check (from the shipped data itself):")
for ax in AXES:
    col = ALL[ax].to_numpy(float)
    nz = (col != 0).mean()
    print(f"  {ax:13s} range [{col.min():.2f}, {col.max():.2f}]  nonzero in {nz*100:5.1f}% of cells")
stale_corr = np.corrcoef(ALL['discover'], ALL['staleness'])[0,1]
print(f"\n  staleness is NOT a copy of discover  →  corr(discover, staleness) = {stale_corr:+.3f}")
print(f"  conservation is a real sparse layer →  {(ALL['conservation']>0).sum():,} cells carry at-risk richness")

Grid          : 31,804 land cells, 0.25° (~25 km)
Extent        : lon -141.0…-52.0, lat 41.0…84.0 (all of Canada + a US land sliver)
Life groups   : 11 → All biodiversity, Plantae, Insecta, Aves, Fungi, Mammalia, Actinopterygii, Reptilia, Amphibia, Arachnida, Mollusca
Per cell      : 5 axes (0–1) + travel_min + n_train
Provenance    : 93f50ffa87672b40…  (27 files frozen)

Axis reality check (from the shipped data itself):
  discover      range [0.00, 1.00]  nonzero in 100.0% of cells
  conservation  range [0.00, 1.00]  nonzero in  10.4% of cells
  env           range [0.00, 1.00]  nonzero in  99.9% of cells
  staleness     range [0.00, 1.00]  nonzero in  28.1% of cells
  urgency       range [0.00, 1.00]  nonzero in  40.9% of cells

  staleness is NOT a copy of discover  →  corr(discover, staleness) = -0.526
  conservation is a real sparse layer →  3,292 cells carry at-risk richness

us_west = (ALL["lon"] < -95) & (ALL["lat"] < 49.0)         # clearly-US band (W of -95, S of the 49th-parallel border)
n_us = int(us_west.sum()); n_us_rec = int((us_west & (ALL["n_train"] > 0)).sum())

fig, ax = plt.subplots(figsize=(8.5, 4.6))
ax.scatter(ALL.lon[~us_west], ALL.lat[~us_west], s=2, marker="s", color="#cbd5e0", label="Canadian / border cells")
ax.scatter(ALL.lon[us_west],  ALL.lat[us_west],  s=2, marker="s", color="#c0392b", label="clearly-US cells")
ax.axhline(49, color="#2c3e50", lw=0.8, ls="--"); ax.text(-138, 49.4, "49°N border", fontsize=8, color="#2c3e50")
ax.set_aspect(ASPECT); ax.set_xticks([]); ax.set_yticks([])
for s in ax.spines.values(): s.set_visible(False)
ax.legend(loc="lower left", fontsize=9, markerscale=4)
ax.set_title(f"{n_us:,} cells (~{us_west.mean()*100:.0f}%) are clearly US — but only {n_us_rec} carry iNaturalist records")
plt.tight_layout(); plt.show()
print(f"Clearly-US cells: {n_us:,}  ·  of those with real iNat records: {n_us_rec}")
print("→ Most US cells are gaps only because the Canadian density COG stops at the border.\n")

# Transect across the western border (lon -125..-100): the seam at ~49°N is the COG edge.
band = (ALL.lon > -125) & (ALL.lon < -100)
print("Transect across the 49°N border (the visible seam):")
print(f"{'lat':>6} {'n_train':>9} {'%zero-obs':>10} {'discover':>9}")
for lo in [47.5, 48.0, 48.5, 49.0, 49.5]:
    m = band & (ALL.lat >= lo) & (ALL.lat < lo + 0.5)
    side = "US " if lo < 49 else "CA "
    print(f"{side}{lo:5.1f} {ALL.n_train[m].mean():9.0f} {(ALL.n_train[m]==0).mean()*100:9.0f}% {ALL.discover[m].mean():9.2f}")
print("→ density 0→thousands and zero-obs 95%→0% across the line = a data-footprint edge, not ecology.")

Clearly-US cells: 3,154  ·  of those with real iNat records: 1673
→ Most US cells are gaps only because the Canadian density COG stops at the border.

Transect across the 49°N border (the visible seam):
   lat   n_train  %zero-obs  discover
US  47.5        59        45%      0.36
US  48.0       755        45%      0.37
US  48.5      2346        24%      0.28
CA  49.0     12314         0%      0.08
CA  49.5      3261         0%      0.09
→ density 0→thousands and zero-obs 95%→0% across the line = a data-footprint edge, not ecology.

titles = {
    "discover":     "Discover\n(under-sampled)",
    "conservation": "Find rare species\n(at-risk richness)",
    "env":          "Cover every habitat\n(climate surprisal)",
    "staleness":    "Freshest gaps\n(quiet lately)",
    "urgency":      "Sample before it's lost\n(forest loss)",
}
fig, axs = plt.subplots(1, 5, figsize=(15, 3.6))
for ax, k in zip(axs, AXES):
    canada_map(ax, ALL[k], titles[k], cmap="magma")
fig.suptitle("The five priority axes — 'All biodiversity', 31,804 real cells (darker = higher priority)", y=1.02)
plt.tight_layout(); plt.show()

# n_train ≈ density × cell-area, so it's a faithful stand-in for the raw density the
# naive formula would have used. Compare the naive saturating transform to the shipped rank.
dens = ALL["n_train"].to_numpy(float)
naive = 1.0 / (dens + 1e-3)
naive = (naive - naive.min()) / (naive.max() - naive.min())   # min–max norm(1/density)
shipped = ALL["discover"].to_numpy(float)                      # the de-saturated rank

fig, axs = plt.subplots(1, 3, figsize=(14, 3.8))
axs[0].hist(naive, bins=60, color="#c0392b"); axs[0].set_title("Naive  norm(1/density)\nbimodal: a binary mask")
axs[0].set_xlabel("axis value"); axs[0].set_ylabel("cells")
axs[1].hist(shipped, bins=60, color="#2980b9"); axs[1].set_title("Shipped  under-sampling rank\nsmooth: a plannable gradient")
axs[1].set_xlabel("axis value")
canada_map(axs[2], shipped, "Shipped discover over Canada", cmap="magma")
frac_pinned = (naive > 0.99).mean()
fig.suptitle(f"Why a rank, not 1/density:  {frac_pinned*100:.0f}% of cells pin at the max under the naive transform", y=1.03)
plt.tight_layout(); plt.show()

cons = pd.read_csv(CA / "ca_atrisk_richness.csv")
top = cons.nlargest(8, "n_species").reset_index(drop=True)

def name_place(lat, lon):
    spots = [("Point Pelee / Carolinian SW Ontario", 42.0, -82.5),
             ("Southern Vancouver Island / Garry Oak", 48.5, -123.4),
             ("Okanagan", 49.9, -119.5), ("Lower Mainland BC", 49.2, -122.5),
             ("Niagara / Golden Horseshoe", 43.1, -79.4), ("Montréal / St. Lawrence", 45.5, -73.6)]
    return min(spots, key=lambda s: (s[1]-lat)**2 + (s[2]-lon)**2)[0]
top["likely region"] = [name_place(r.lat, r.lon) for r in top.itertuples()]
# Region first, with explicit widths, so its long labels get the room and never clip the right edge.
top = top[["likely region", "n_species", "lat", "lon"]]

fig, axs = plt.subplots(1, 2, figsize=(12, 4.2))
canada_map(axs[0], ALL["conservation"], "Conservation axis (25 km, all-taxa pooled)", cmap="magma")
axs[1].axis("off")
axs[1].set_title("Top at-risk-richness cells = Canada's real hotspots", fontsize=10)
tbl = axs[1].table(cellText=top.round(2).values, colLabels=top.columns, loc="center",
                   cellLoc="left", colWidths=[0.52, 0.16, 0.16, 0.16])
tbl.auto_set_font_size(False); tbl.set_fontsize(8); tbl.scale(1, 1.5)
fig.suptitle("Find rare species — concentrated, validated against known hotspots, deliberately coarse", y=1.0)
plt.tight_layout(); plt.show()
print("Caveat: reflects ASSESSED species only (CAN-SAR ~2021); IUCN/COSEWIC under-assess inverts, plants, fungi.")

Caveat: reflects ASSESSED species only (CAN-SAR ~2021); IUCN/COSEWIC under-assess inverts, plants, fungi.

fig, axs = plt.subplots(1, 2, figsize=(12, 3.8))
canada_map(axs[0], ALL["env"], "How rare each square's climate is", cmap="magma")
axs[1].hist(ALL["env"], bins=60, color="#16a085")
axs[1].set_title("Percentile-ranked → uses the full 0–1 range"); axs[1].set_xlabel("env"); axs[1].set_ylabel("cells")
plt.tight_layout(); plt.show()

met = pd.read_csv(CA / "ca_inat_metrics.csv")
# Show staleness is a genuinely distinct signal, and surface a concrete "quiet now" cell.
fig, axs = plt.subplots(1, 2, figsize=(12, 4.0))
s = ALL.sample(6000, random_state=0)
axs[0].scatter(s["discover"], s["staleness"], s=4, alpha=0.25, color="#8e44ad")
axs[0].set_xlabel("discover (under-sampled)"); axs[0].set_ylabel("staleness (quiet lately)")
r = np.corrcoef(ALL["discover"], ALL["staleness"])[0,1]
axs[0].set_title(f"Distinct, not a mirror — corr = {r:+.2f}\n(well-recorded ≠ recently-recorded)")
busy_then_quiet = met[(met.n_all >= 200)].assign(recent_frac=lambda d: d.n_recent/d.n_all).nsmallest(8, "recent_frac")
axs[1].axis("off"); axs[1].set_title("Recorded a lot all-time, almost nothing recently", fontsize=10)
show = busy_then_quiet[["lat","lon","n_all","n_recent"]].round(2) if "lat" in busy_then_quiet else busy_then_quiet[["n_all","n_recent"]]
t = axs[1].table(cellText=show.values, colLabels=show.columns, loc="center"); t.auto_set_font_size(False); t.set_fontsize(8); t.scale(1,1.5)
plt.tight_layout(); plt.show()

fig, ax = plt.subplots(figsize=(7.5, 4.2))
canada_map(ax, ALL["urgency"], "Urgency — Hansen forest-loss fraction", cmap="magma")
plt.tight_layout(); plt.show()
fm = ALL.iloc[((ALL.lat-56.7)**2 + (ALL.lon+111.4)**2).idxmin()]
print(f"Fort McMurray-area cell urgency = {fm.urgency:.2f}  (vs national median {ALL.urgency.median():.3f})")

Fort McMurray-area cell urgency = 0.75  (vs national median 0.000)

DEFAULT_W = {"discover": 0.8, "conservation": 0.0, "env": 0.7, "staleness": 0.0, "urgency": 0.3}  # 'Biodiversity impact' preset
raw = sum(DEFAULT_W[a] * ALL[a].to_numpy(float) for a in AXES)
minmax = (raw - raw.min()) / (raw.max() - raw.min()) * 100
order = raw.argsort(); pct = np.empty_like(raw); pct[order] = np.linspace(0, 100, len(raw))

fig, axs = plt.subplots(2, 2, figsize=(12, 7.4))
canada_map(axs[0,0], minmax, "If we coloured by the RAW total", cmap="magma", vmax=100)
canada_map(axs[0,1], pct,    "What the app shows: a RANKING", cmap="magma", vmax=100)
axs[1,0].hist(minmax, bins=60, color="#c0392b"); axs[1,0].set_title("Raw total: a few Arctic gaps crush everything to ~0"); axs[1,0].set_xlabel("score 0–100"); axs[1,0].set_ylabel("squares")
axs[1,1].hist(pct, bins=60, color="#27ae60"); axs[1,1].set_title("Ranking: full 0–100 range, usable for planning"); axs[1,1].set_xlabel("score 0–100")
crushed = (minmax < 5).mean()
fig.suptitle(f"Same blended score, two ways to colour it — the raw total pins {crushed*100:.0f}% of squares below 5/100", y=1.0)
plt.tight_layout(); plt.show()

PRESETS = [
    ("Biodiversity impact", [0.8,0,0.7,0,0.3], "blitz-the-gap-2026-general"),
    ("The Other 99%",       [1,0,0.3,0,0],     "blitz-the-gap-the-other-99"),
    ("Most Wanted",         [0.1,1,0,0,0.3],   "blitz-the-gap-canada-s-most-wanted"),
    ("Too Hot to Handle",   [0.2,0,0.2,0,1],   "blitz-the-gap-too-hot-to-handle"),
    ("Climate Gap",         [0.2,0,1,0,0],     "blitz-the-gap-closing-the-climate-gap"),
    ("Revisit the Past",    [0.3,0,0,1,0],     "blitz-the-gap-revisiting-the-past"),
]
pd.set_option("display.max_colwidth", 60)
pdf = pd.DataFrame([(n, *w, proj) for n,w,proj in PRESETS], columns=["preset",*AXES,"iNat project"])
display(pdf)

# Two presets, two genuinely different maps — same geometry, different weighting.
def impact_pct(weights):
    raw = sum(weights[i]*ALL[AXES[i]].to_numpy(float) for i in range(5))
    o = raw.argsort(); p = np.empty_like(raw); p[o] = np.linspace(0,100,len(raw)); return p
fig, axs = plt.subplots(1, 3, figsize=(14, 3.8))
for ax,(name,w,_) in zip(axs, [PRESETS[1], PRESETS[2], PRESETS[3]]):
    canada_map(ax, impact_pct(w), name, cmap="magma", vmax=100)
fig.suptitle("Same cells, different values → different trips (The Other 99% · Most Wanted · Too Hot to Handle)", y=1.03)
plt.tight_layout(); plt.show()

rec = ALL[ALL["n_train"] > 0]
M = rec[AXES].corr(method="spearman")
fig, ax = plt.subplots(figsize=(5.8, 5))
im = ax.imshow(M, cmap="RdBu_r", vmin=-1, vmax=1)
ax.set_xticks(range(5)); ax.set_yticks(range(5))
ax.set_xticklabels(AXES, rotation=40, ha="right"); ax.set_yticklabels(AXES)
for i in range(5):
    for j in range(5):
        ax.text(j, i, f"{M.iloc[i,j]:+.2f}", ha="center", va="center",
                color="white" if abs(M.iloc[i,j])>0.5 else "black", fontsize=9)
ax.set_title(f"How much each pair of goals agrees on where to go\n(+1 same places · 0 unrelated · −1 opposite; over {len(rec):,} recorded squares)")
fig.colorbar(im, fraction=0.046, pad=0.04); plt.tight_layout(); plt.show()

app = json.load(open(RES / "voi_appscore_results.json"))
rows = []
for r in app:
    s = r["scores"]
    rows.append({"taxon": r["taxon"],
                 "app (gap-filling)": s["app_leakfree"]["spearman"],
                 "discover only":     s["discover_leakfree"]["spearman"],
                 "go-where-busy":     s["opportunistic_density"]["spearman"]})
bt = pd.DataFrame(rows).set_index("taxon")

fig, ax = plt.subplots(figsize=(9.5, 4.4))
bt.plot.bar(ax=ax, color=["#27ae60", "#2980b9", "#c0392b"], width=0.8)
ax.axhline(0, color="k", lw=0.8)
ax.set_ylabel("Agreement with what was discovered next\n(+1 nailed it · 0 chance · −1 exactly wrong)")
ax.set_title("Sending people to gaps finds more new species; sending them where it's busy finds fewer (mirror image)")
ax.legend(loc="lower right", fontsize=9); ax.set_xlabel("")
plt.xticks(rotation=0); plt.tight_layout(); plt.show()

m_app  = bt["app (gap-filling)"].mean()
m_busy = bt["go-where-busy"].mean()
print(f"Average agreement — app's gap-filling map : {m_app:+.2f}  (correct on {bt['app (gap-filling)'].gt(0).sum()}/{len(bt)} animal groups)")
print(f"Average agreement — 'go where it's busy'  : {m_busy:+.2f}  (the near-exact opposite)")

Average agreement — app's gap-filling map : +0.34  (correct on 5/5 animal groups)
Average agreement — 'go where it's busy'  : -0.58  (the near-exact opposite)

# Replication on a disjoint region (Eastern Canada), same harness.
east = json.load(open(RES / "voi_appscore_east_results.json"))
er = pd.DataFrame([{"taxon": r["taxon"],
                    "app (gap-filling)": r["scores"]["app_leakfree"]["spearman"],
                    "go-where-busy":     r["scores"]["opportunistic_density"]["spearman"]}
                   for r in east]).set_index("taxon")
print("Eastern-Canada replication (disjoint from BC):")
display(er.round(2))
print(f"\nDirected stays positive on {er['app (gap-filling)'].gt(0).sum()}/{len(er)} eastern taxa → not a regional artifact.")

Eastern-Canada replication (disjoint from BC):

Directed stays positive on 3/3 eastern taxa → not a regional artifact.

# 1. environment (no credentials needed; all sources public)
uv venv --python 3.11 .venv && uv pip install --python .venv/bin/python -r requirements.txt

# 2. regenerate the national build from cluster_results/ca/ (deterministic, byte-identical)
.venv/bin/python build_webapp.py          # → index.html

# 3. rebuild + execute THIS walkthrough, then render the HTML the in-app link points to
.venv/bin/python _build_walkthrough.py
.venv/bin/python -m nbconvert --to notebook --execute --inplace where-to-blitz-walkthrough.ipynb
.venv/bin/python render_walkthrough_html.py    # → where-to-blitz-walkthrough.html (adds figure alt text)

# OPTIONAL — live iNaturalist "fill the gap" near a chosen cell. Guarded: skips cleanly offline.
import urllib.request
def fill_the_gap(lat, lon, radius_km=25):
    url = ("https://api.inaturalist.org/v1/observations/species_counts"
           f"?lat={lat}&lng={lon}&radius={radius_km}&quality_grade=research"
           "&taxon_geoprivacy=open&threatened=false&per_page=5")
    with urllib.request.urlopen(url, timeout=8) as r:
        return json.load(r)
try:
    # pick the highest-priority reachable-ish CANADIAN cell under the default preset
    # (exclude the clearly-US band from §2.1 so the demo stays on-mission)
    cand = ALL.assign(imp=pct, usw=us_west.to_numpy()).query("travel_min < 600 and not usw")
    cell = cand.nlargest(1, "imp").iloc[0]
    print(f"Top reachable gap under default preset: ({cell.lat:.2f}, {cell.lon:.2f})  impact≈{cell.imp:.0f}/100")
    res = fill_the_gap(cell.lat, cell.lon)
    print(f"iNaturalist shows {res['total_results']:,} research-grade species in the surrounding 25 km. Sample:")
    for t in res["results"][:5]:
        print(f"  • {t['taxon'].get('preferred_common_name', t['taxon']['name'])}  ({t['count']} obs)")
except Exception as e:
    print(f"(live iNaturalist call skipped — offline or rate-limited: {type(e).__name__})")

Top reachable gap under default preset: (58.38, -117.38)  impact≈69/100

iNaturalist shows 90 research-grade species in the surrounding 25 km. Sample:
  • Pine Grosbeak  (6 obs)
  • Four-spotted Ghost Moth  (6 obs)
  • Black-billed Magpie  (5 obs)
  • Redpoll  (5 obs)
  • White-spotted Sawyer  (4 obs)

Module	Responsible for
`build_fullgrid_ca.py`	The four raster goals per square — discover, habitat coverage, urgency, travel — plus `n_train`. Defines the grid + land mask.
`build_atrisk_layer.py` + `join_conservation.py`	The rare-species goal: CAN-SAR (COSEWIC/SARA) species × their GBIF occurrences → per-square status-weighted at-risk richness, aggregated to 25 km.
cluster DuckDB → `ca_inat_metrics.csv` → `build_staleness_layer.py`	The freshness goal: all-time vs last-5-years iNaturalist density per square.
`build_webapp.py`	The single source of truth for the app. Combines the five goals into impact (a percentile rank), holds every UI string (EN/FR), the trip planner, and emits `index.html`.
`build_provenance.py`	Freezes the national build to a hashed manifest (`provenance.json`).
`voi_backtest.py`, `backtest_appscore.py`	The validation layer: do gap-filling priorities actually discover more than going where it's already busy?

	lat	lon	discover	env	staleness	urgency	travel_min	n_train
0	41.125	-124.125	0.0505	0.625	0.31232	0.403	159.0	894
1	41.125	-123.875	0.2055	0.500	0.22263	0.058	176.0	29
2	41.125	-123.625	0.5688	0.393	0.22998	0.083	158.0	0

Axis	Source	What it is	Resolution	Licence / access
discover	iNaturalist observation density (Biodiversité Québec STAC)	per-group "light up the map" record density — the same surface the official project uses	1 km COG	public COG on Arbutus
conservation	CAN-SAR (COSEWIC/SARA) × GBIF	521 at-risk species × their Canadian occurrences	point → 25 km	CC-BY (OSF `10.17605/OSF.IO/E4A58`) + public GBIF
env	CHELSA bioclimate	temperature, seasonality, precipitation	~1 km	public (`/vsicurl`)
staleness	iNaturalist open-data dump	per-record dates → all-time vs recent	per-record	public AWS dump
urgency	Hansen Global Forest Change	forest-loss fraction (logging / fire / dieback)	0.05°	public
travel	Weiss et al. 2018 accessibility	minutes to the nearest city	~1 km	public (MalariaAtlas)

Axis (code name)	The goal, in plain English	A high score means
`discover`	Go where few people have looked	Hardly any iNaturalist records nearby
`conservation`	Go where at-risk species live	Many threatened species recorded around the square
`env`	Cover every habitat	The square's climate type is barely sampled anywhere
`staleness`	Revisit places that have gone quiet	Lots of old records, very few recent ones
`urgency`	Record it before it's lost	Recent forest-cover loss (logging, fire, dieback)

Where to Blitz the Gap — a walkthrough¶

0 · Zoom out — the whole tool in one picture¶

1 · What's in it — contents at a glance¶

2 · The data¶

2.1 · Why the map spills into the United States¶

3 · The five goals — what each one looks like over Canada¶

3.1 · Discover — and the de-saturation choice¶

3.2 · Find rare species — and the dual-use safety choice¶

3.3 · Cover every habitat — sampling Canada's rarest climates¶

3.4 · Freshest gaps — and the iNat-only data-integrity choice¶

3.5 · Sample before it's lost — recent habitat change¶

4 · The score on the map — "impact 0–100"¶

5 · Presets — value choices, made explicit and linked to real challenges¶

6 · Do the goals actually point to different places?¶

7 · From map to trip — how navigation works¶

8 · Does the core idea actually work? — we tested it like a forecast¶

9 · Responsible use — guardrails as design, not disclaimer¶

10 · Reproduce everything¶

	preset	discover	conservation	env	staleness	urgency	iNat project
0	Biodiversity impact	0.8	0	0.7	0	0.3	blitz-the-gap-2026-general
1	The Other 99%	1.0	0	0.3	0	0.0	blitz-the-gap-the-other-99
2	Most Wanted	0.1	1	0.0	0	0.3	blitz-the-gap-canada-s-most-wanted
3	Too Hot to Handle	0.2	0	0.2	0	1.0	blitz-the-gap-too-hot-to-handle
4	Climate Gap	0.2	0	1.0	0	0.0	blitz-the-gap-closing-the-climate-gap
5	Revisit the Past	0.3	0	0.0	1	0.0	blitz-the-gap-revisiting-the-past

	app (gap-filling)	go-where-busy
taxon
Aves	0.37	-0.58
Insecta	0.30	-0.47
Mammalia	0.52	-0.69