Fire-Fiscal Master Report: value-per-acre analysis, fiscal productivity, citations - #13
Open
SlowSpeedChase wants to merge 29 commits into
Open
Fire-Fiscal Master Report: value-per-acre analysis, fiscal productivity, citations#13SlowSpeedChase wants to merge 29 commits into
SlowSpeedChase wants to merge 29 commits into
Conversation
Standalone, runnable building blocks for the value-per-acre metro map (v2: Travis + Williamson + Hays). All endpoints, field mappings, and PVS ratios live-verified 2026-06-29. - v2_county_sources.py: per-county ArcGIS endpoints + field->canonical mappings (no roll-join needed; every county bundles value+geometry), plus PVS_RATIOS (Travis 1.00 / Williamson 0.96 / Hays 0.97, cross-verified via XHR scrape + headless render). Runnable self-check of parcel counts. - v2_fetch_parcels.py: geopandas-optional paginated reader normalizing all three counties to one schema; stdlib core smoke-tested against live data. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
notebooks/value_per_acre_metro.ipynb: config-driven pipeline that fetches parcels, computes value_per_acre / value_per_acre_adj / tax_per_acre, runs the Urban3 top/bottom sanity check, and renders an extruded pydeck 3D scene (log color ramp, p99-winsorized, tallest parcel -> 2500 m). Default SCOPE is a fast downtown-Travis bbox slice; flip to full counties for the metro. Validated: 21,092 downtown parcels; top by value/acre = downtown hi-rise office/condo/hotel ($336-483M/acre), bottom = large low-value tracts. Output HTML is gitignored. - v2_fetch_parcels.py: add bbox (envelope) support to fetch_county_records for bounded slices; fix fetch_county_gdf duplicate with_geometry kwarg. - requirements.txt: +pydeck, +pyarrow. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
- Acreage clean: MIN_ACRES=0.01 floor drops sub-436-sqft error slivers (the $392M/acre artifacts); validated to remove 9 downtown parcels with no effect on the legitimate top-10 hi-rise parcels. - Metro scaling: H3 hex aggregation (value=sum(market)/sum(acres) per cell) rendered as a 2D folium choropleth (shareable) and a 3D extruded H3HexagonLayer (metro hero) — true-parcel 3D auto-skips above 200k parcels. - Geometry simplification (SIMPLIFY_M, projected to EPSG:2277) for the true-parcel 3D scene; parquet caching via LOAD_METRO_CACHE for the ~638k-parcel metro pull. - requirements.txt: +h3. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The suburban CADs leave land_acres null for ~60% of platted lots, which dropped Williamson/Hays parcels en masse. Fix: derive acreage from each parcel's geometry footprint (EPSG:2277) wherever the CAD attribute is null/<=0. Recovers the suburbs and unifies the three counties. Result: full metro cleans to 576,855 parcels (Travis 225,687 / Williamson 265,506 / Hays 85,662), aggregated to 6,412 H3 res-8 hexes. Validated: downtown Austin tops all hexes ($17-43M/acre); Georgetown Sq ($3.6M), Round Rock ($2.6M), Cedar Park ($2.2M), San Marcos ($1.3M) register as local peaks above their county medians. - notebook: geometry-acreage fallback in clean cell; acre_source flag. - .gitignore: ignore the 118 MB metro parquet. - Metro maps (gitignored outputs): map_value_per_acre_metro.html (2D H3 choropleth) + map_value_per_acre_metro_3d.html (3D extruded hexes). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Color rework for "see what is where": - Quantile-binned (8-class) turbo palette instead of log-continuous ramp, so the full color range is used and small differences stay visible. - branca StepColormap doubles as an on-map legend (value/acre thresholds). - Light labeled basemap (CartoDB positron / pydeck 'road') for orientation; thin hex outlines; fill_opacity 0.7. Highways: fetch metro motorway+trunk from OpenStreetMap (cached geojson, simplified for web), overlaid as white-cased dark lines on the 2D map and a PathLayer on the 3D scenes so the main highways (I-35, MoPac, 183, 130, 290, 71, 360) read clearly over the choropleth. CMAP / N_BINS / BASEMAP are config knobs. .gitignore: +highways geojson. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
notebooks/fiscal_productivity.ipynb tests the Strong Towns hypothesis with a land/acre break-even model: revenue = market_value * 2.1%; break-even = total levy / total acres (uniform cost-per-acre); net/acre = tax/acre - break-even. Two lenses (the headline is that the answer depends on which): - All land (break-even $9,566/ac): Austin core +surplus; the big deficit is rural/unincorporated sprawl, not incorporated suburbs. - Developed only, <=5 ac (break-even $34,624/ac): Austin strongly positive (+$27,771/ac); most measured suburbs fall below break-even (Georgetown, Leander, San Marcos, Hutto net-negative) -> supports the hypothesis among developed land. City labels via spatial join to Census incorporated places. Outputs: a diverging "does this land pay for itself" H3 map (red=drain/green=surplus, highways overlaid) + per-city and density-gradient charts. Coverage gap documented with a one-line PROP_ID join hook: Travis independent suburbs (Lakeway, Pflugerville, Bee Cave, West Lake Hills) are excluded because TCAD per-parcel values are gated; geometry+acreage+city are already free from the TCAD GIS layer, only value needs a roll. .gitignore: +tx_places.zip cache. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…eline Replaces the Austin-only Travis source with the COMPLETE county: TCAD public GIS geometry (386,682 parcels) joined by PROP_ID to the TCAD 2025 Certified Appraisal Roll (parsed from the published PACS PROP.TXT export). Every Travis city is now covered with current values incl. Lakeway, Pflugerville, Bee Cave, West Lake Hills. Metro = 724,639 parcels at ~2025 vintage (Williamson/Hays from current CAD feeds), so the 2020 vintage caveat is resolved. - 12_parse_tcad_roll.py: streaming fixed-width parser for the certified roll PROP.TXT -> prop_id/market_value/appraised/legal_acreage (validated: Lakeway 438295 = $888k 2025). Documents the field layout + mid-July refresh. - 13_build_metro_parcels.py: join Travis geometry+roll values + Williamson/Hays -> metro parquet with metrics. One command to rebuild for a newer roll year. - fiscal_productivity.ipynb: re-run on 2025 metro; swap the old data-gap section for a Data Provenance & Vintage section (sources, vintages, refresh steps, why StratMap Travis values are unusable / use the certified roll). Finding sharpens: Lakeway and Pflugerville are net CONTRIBUTORS per developed acre; the drains are low-value exurban/rural land (unincorporated -$2.0B), not affluent close-in suburbs. It's value-per-acre, not city-vs-suburb, that decides. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Adds a second, more defensible cost-to-serve model alongside the uniform land/acre one: cost proportional to LOCAL road-miles (TIGER roads, excluding state-maintained S1100 interstates/primary), calibrated to break even metro-wide -> Strong Towns' "value per road-mile." Result sharpens and changes 11 city verdicts: road-heavy growth suburbs flip to net drains (Kyle, Buda, Georgetown, San Marcos), while big-lot high-value enclaves strengthen (West Lake Hills, Bee Cave, Lakeway). The ~-$2B/yr unincorporated deficit persists under the road model -> not a uniform-cost artifact. Deciding variable becomes value-per-infrastructure, not city-vs-suburb. - notebook: TIGER road load (download-if-absent), per-city land-vs-road comparison chart, road-cost fiscal map; Findings + caveats updated. - Also regenerated value-per-acre hero maps on the 2025 full-Travis data. - Outputs: map_fiscal_productivity_roadcost.html, fiscal_land_vs_road.png. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
notebooks/fire_use_vs_pays.ipynb: does each Austin response area fund the fire
service it consumes? pays-in = area's share of citywide property value (fire-
funding proxy) x AFD budget; uses measured under two lenses:
- Demand: cost-weighted calls (structure fire 10x trash fire, etc.)
- Coverage: fixed standby cost per zone (each first-due area needs a staffed
company) -- the realistic lens since ~90% of fire cost is 24/7 staffing.
Finding: the lenses disagree by design. Demand lens -> busy inner suburbs look
subsidized (-$22M), outer suburbs contribute (+$21M). Coverage lens flips it ->
inner suburbs contribute (+$45M), low-density OUTER suburbs are the net drain
(-$51M): each spread-out zone still needs a station but holds less taxable
value. Coverage echoes the road result -- low-density development is expensive
to cover per dollar it funds.
Scope: AFD/Austin, fire calls only (no EMS), 2022-24; 285 served response areas,
$263B value, 20,920 calls. AFD_BUDGET ($264M) and WEIGHTS are config knobs that
scale $ only. Outputs: map_fire_balance_{coverage,demand}.html,
fire_by_urban_class.png, fire_equity_scatter.png.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Two refinements to the coverage lens, from the City of Austin LOCATION_fire_stations layer (64 AFD stations, RESOURCES apparatus field): - Apparatus-weighted coverage: each zone's standby cost scaled by its first-due station's apparatus (ladder/quint > single engine). Moderates but confirms the flat result -- outer suburbs -$51M->-$33M (mostly single engines), inner +$45M->+$32M (carry the costly ladder/quint companies). Direction robust. - Distance to nearest station: outer-suburban zones median 1.2mi vs 0.8mi inner (worst >5mi) -- sprawl is both costlier to cover AND slower-served (double penalty), reinforcing the road-infrastructure finding. Adds station fetch (download-if-absent), a 3-cost-model + distance chart, and updated findings. Output: fire_apparatus_distance.png. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…vity docs/FISCAL_PRODUCTIVITY_REPORT.md — methodology + results + evidence synthesis across the three analyses (value-per-acre, road-cost fiscal, fire use-vs-pays). Core argument: fiscal productivity is governed by value-per-unit-of-infrastructure, not city-vs-suburb; three independent datasets/methods converge, so the finding is robust to any one method's weaknesses. 14_build_report_pdf.py renders it to a styled PDF via Chrome headless (markdown -> HTML with base64-inlined figures -> --print-to-pdf; no LaTeX/pandoc needed). requirements.txt: +markdown. PDF output is gitignored (regenerable). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
… and interactive map Builds the large educational report, Austin civic color system, and four-layer deck.gl map on the data present in this container (response areas, incidents, census). Parcel-, station-, and highway-dependent steps are written and guarded to activate automatically once parcels_value_per_acre_metro.parquet, fire_stations.geojson, and metro_highways.geojson are added to processed_data/ (network policy blocks regenerating them here).
…harts The cloud delivery was parquet-free (gitignored 360MB of data absent), so its parcel/fiscal/net-balance checks were PENDING and the interactive map fell back to placeholders. Run everything locally against the real data: - validation.ipynb: all 18 checks PASS against the full 724,639-parcel roll (market value $597.9B, revenue $12.56B, break-even $8,421, net-balance conservation 1.1e-16, 285 served areas, 64 stations). Exports fire_net_balance.geojson for the map. - interactive_map.ipynb: all four deck.gl layers populate (8,182 value/acre hexes, 64 stations, 12 landmarks, true coverage-lens net balance). - 15_build_report_figures.py: add fig_interactive_map_preview (static Austin-palette choropleth + landmark pins) for the PDF. - Report: embed the three validated analysis charts (§6.2-6.3), drop the stale "parquet absent / PENDING" caveat, mark all 18 checks green, scope the color-consistency claim honestly, and cite the web-verified budget sources ($5.9B all funds / $1.4B GF / AFD ~$263-264M). - gitignore the generated fire_net_balance.geojson. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01P683RYjieUENCwnNz3wyfe
- fig_colloquial_inventory: compute per-response-area value/acre (parcel sjoin) + merge validated net balance, write outputs/colloquial_inventory.csv (285 areas), render Austin-palette top-contributors / biggest-drains table. Embed in report §7 with the floor explanation (drains cluster at -1/285 x budget). - fig_palette_swatch: add hspace so the DIV_NET title no longer overlaps the SEQ_VPA axis labels. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01P683RYjieUENCwnNz3wyfe
- 14_build_report_pdf.py --scribe: page sized to the Scribe's native 6.2x8.27in (1860x2480 @ 300ppi) panel with larger ~12pt body type and slim margins, so text renders 1:1 and large on e-ink. Outputs *_scribe.pdf. - The diverging burnt-orange<->bluebonnet net-balance scale collapses in grayscale (both signs go dark); add a luminance-monotonic cividis variant (fig_interactive_map_preview_gray, dark=drain -> light=contributor) and swap it (and its caption) into the Scribe build only. Verified legible via grayscale render. Other figures survive grayscale (numeric labels / position / light-vs-dark series), so only the map needed the variant. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01P683RYjieUENCwnNz3wyfe
Address review feedback ("sounds too much like AI"). Strip the flagged tells
and self-praise, keep the analysis neutral (no advocacy framing per request):
- Cut "Color is used as a signal, not decoration", "the honest qualifier",
"Concepts primer — every idea, in plain English", the repeated "In plain
English:" boxes, "the defensible model", "the realistic one", "exactly the
textbook result"; plainer headings ("Key concepts", "Calculations and
methodology"); lead the summary with the plain question.
- §6.2: list the flipping cities with median home value (computed from the
parcel roll) — West Lake Hills ~$1.69M strengthens ... Kyle ~$315k flips.
- §6.3: define urban-core / inner / outer suburban; replace "the realistic one"
with the why (≈90% of a fire budget is fixed standby).
- §4: add "what we use it for" to each data-source group.
- Color note: acknowledge the diverging scale is hard to distinguish; numbers
carry the same info. Did NOT add the wealth-distribution framing (per request).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01P683RYjieUENCwnNz3wyfe
- §5.3 / §13: drop remaining "more defensible" self-praise; glossary "realistic cost driver" -> plain. - Simplify the staged-run runbook prose (was jargony: "flip green", "guarded branches"). - Plainer headings: "The data inputs", "The data sources, in detail", "The color key". - Fix the intro's color-key link to the renamed anchor (#the-color-key). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01P683RYjieUENCwnNz3wyfe
Prose corrections (all caught by independent recomputation): - §6.2: San Marcos does not flip (drain under both models); Bee Cave flips to contributor; the 11 verdict-changers are now enumerated and exported to outputs/city_fiscal_verdicts.csv by fiscal_productivity.ipynb - §6.3: 20,136 served-area calls (20,920 is the citywide total) - §5.1: geometry fallback fills 13.8% of Travis parcels, not ~60%; legal_acreage step documented as inert - §6.2 medians now computed/exported by committed code (WLH ~$1.74M) - §7 top-contributor claim scoped to top four; San Marcos landmark row fixed - §9: disclose Travis-only pays-in denominator - §5.2: note revenue model is market-basis, not a collections forecast - §2.5: cite the ~90% standby claim (FireRescue1 / Fitch & Associates) Validation gate: +6 recomputed checks covering §6.2/§6.3 (verdict flips, class table, distances, served calls, WLH median) — 24 checks, all green; external-constant rows relabeled as citation log; KUT source aligned. Presentation: notebook charts restyled with viz_palette (flip cities highlighted), concept-figure title overlap fixed, tighter table figures, map preview extent/ticks cleaned, code blocks wrap, page numbers stamped (pypdf+reportlab), TOC excludes title/subtitle. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…docs Layout (git mv, nothing deleted): - incident_pipeline/ 01-06 + census_variables, run_all, quick_start (run_all now anchors raw_data/outputs/requirements to the repo root) - report_pipeline/ 12-15, v2_*, viz_palette (REPO anchors -> parent.parent; report md footnotes, README, and notebook sys.path updated to match) - legacy/ 03b, 04b, 07-10 orphaned scripts + README explaining provenance (10_townhome kept for open follow-up tasks 5/7) Tracking hygiene: - Untracked ~150MB of generated artifacts that .gitignore already claimed (5 folium maps, 2 response-area geojsons, tracts_incidents_map.html, REPORT_Austin_Fire_Analysis.html); files remain on disk - .gitignore restructured: outputs/* with negations so the small csv/png/txt/md artifacts are tracked intentionally instead of grandfathered; new report figures and verdict CSVs now tracked for fresh clones - Removed stray __pycache__/ and .DS_Store Docs: PROJECT_GUIDE and DATA_DICTIONARY updated for the fiscal pipeline (new artifact section, all paths fixed); FISCAL_PRODUCTIVITY_REPORT marked superseded by FIRE_FISCAL_FULL_REPORT; README rewritten for the new layout. requirements.txt: add scikit-learn (incident_type_eda.ipynb). Verified: all scripts compile; validation gate re-run green (24/24); fiscal/fire notebooks re-executed; figures + both PDFs rebuild from report_pipeline/; run_all.py --check-only works from repo root. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Every measure and data source now cited at first mention, web-verified 2026-07-01. Merges source citations into the existing code-pointer footnotes (unified stream) and turns §14 into a References section with a 25-entry alphabetized bibliography. Corrections caught during verification: AFD open-data title rolled to 2023-2025 (same dataset ID), approved AFD budget is $262.2M (the $264M constant matches the proposed figure), Tax Code §26.01 caption fixed, ACS span stated as 2018-2022, H3 attributed to Uber (UCF defunct). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
New §15 lists the three unresolved source-vs-constant tensions from the citation verification (2.1% blended rate vs ~1.87-1.98% adopted, $264M proposed vs $262.2M approved AFD budget, rolling open-data title) behind a big UNFINISHED banner on its own page. The PDF build stamps a translucent diagonal UNFINISHED watermark on every page via the existing reportlab pass (grayscale-safe on the Scribe build); the stamp keys off the presence of the §15 banner, so deleting the section when the items are resolved also removes the watermark. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…in the Fire-Fiscal Master Report
…-share framing (#14) - §2.6: define "apparatus"/"company" + add Firefighting apparatus wiki link in [^appw] - §5.3: add missing Herriges/Strong Towns URL to [^vprm] (rendered citation [23]) - §5.4: define N_served_areas (served AFD response areas, ~285 of 765); correct the false "each zone = one staffed company" claim in prose, report code excerpt, and the notebook comment. 64 stations != 285 areas; flat lens is an even standby floor, not one-company-per-area. Response areas are AFD dispatch boxes, not census tracts. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01UKeDxHqo5rpwjwa6CzjdaA
Neutral, data-backed guide to the four open review questions (A: vehicle/trash fires, B: road-mile model, C: revenue vs value, D: §2.4). Markdown renders on GitHub; the self-contained HTML is for printing / giving out. Figures computed from incidents_enriched.csv and outputs/city_fiscal_verdicts.csv. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01UKeDxHqo5rpwjwa6CzjdaA
- docs/METHODS.md: the standard (source/formula/reproduce for every claim) plus a ledger of the handout/report figures (weighted fire demand, verdict agreement, non-structure by area type, revenue definition), each with a reproduction snippet. - CLAUDE.md: "no number ships without provenance" as a development instruction. - Handout (md + html): Methods & provenance section mapping each figure to its ledger entry. - Report: state the provenance standard in §11 and link docs/METHODS.md as companion artifact. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01UKeDxHqo5rpwjwa6CzjdaA
Separate process/working docs from the final product. New docs/working/ folder (mirrors docs/plans/) with per-thread subfolders; pr13/ holds the decision handout (md + print html) and the session handoff. Adds docs/working/README explaining the convention. METHODS.md stays in docs/ (it documents the report's numbers). Repointed METHODS ledger references and the Issue #14 comment link to the new paths. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01UKeDxHqo5rpwjwa6CzjdaA
) C (§2.1-2.3): declare the convention — because revenue = value × a uniform rate, value-per-acre and revenue-per-acre rank identically; the report uses value/acre as its single lens and revenue$ only where an absolute amount is needed. Removes the commingling ambiguity without a full revenue conversion (the exemption-sensitive version needs Travis-only taxable value). D (§2.4): revise, not cut. Retitled "How to read this report: the break-even line" and compressed to one framing paragraph + the diagram. A (vehicle/trash fires) and B (road-mile model) deferred pending Eric's input. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01UKeDxHqo5rpwjwa6CzjdaA
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Consolidates all the fire-fiscal report generation work into a single PR. This branch is a superset of the
productivity_mapandrepo-review-cleanupbranches (both fully contained here), plus the citation/bibliography pass on top.What's included
Value-per-acre & fiscal productivity analysis
v2_county_sources.py,v2_fetch_parcels.py)Fire-Fiscal Master Report
docs/FIRE_FISCAL_FULL_REPORT.md,docs/FISCAL_PRODUCTIVITY_REPORT.md)14_build_report_pdf.py,15_build_report_figures.py)Repo cleanup
Note
Marked UNFINISHED (watermark + §15 worklist) — citation/modeling follow-ups still open.
🤖 Generated with Claude Code