🪲 BugOut Index

Methodology — v1.1.1

The BugOut Index is a directional stress and stability reading. Six published metrics are placed between fixed endpoints and combined into one weighted 0–100 score. Higher means those inputs sit closer to the stable end of the ranges. The number shows how hard current published conditions are pressing. It is not a forecast, and it is not an instruction to bug out. This page describes how the current week's number was produced. Methodology version is v1.1.1.

Raw weights are relative coefficients that sum to 0.72. The “share of score” column is weight / 0.72 — the fraction of the finished index each metric actually carries after the publisher divides by the sum of weights. Tile labels that say “Raw weight 15%” are the raw weight × 100, not the post-division share.

Metric	Raw value	Unit	Normalization range	Normalized	Raw weight	Share of score	Source
Inflation Rate	3.72	%	−10 to 15	45.13	15.0%	20.8%	FRED (CPIAUCSL)
Violent + Property Crime	2,443.27	per 100k	500 to 8,000	74.09	12.0%	16.7%	Real-Time Crime Index
Unemployment Rate	4.10	%	0 to 25	83.6	12.0%	16.7%	FRED (UNRATE)
Debt-to-GDP Ratio	121.19	%	0 to 200	39.41	12.0%	16.7%	FRED (GFDEGDQ188S)
Homelessness Rate	0.22	% of pop.	0 to 0.5	56.0	9.0%	12.5%	HUD AHAR
Trust in Government	39.00	% confidence	0 to 80 (inverted)	48.75	12.0%	16.7%	Edelman Trust Barometer

Publications through 2026-10-02 are methodology 1.0.0. Crime in those weeks is the locked input 2,723.0, an old unweighted print of the Real-Time Crime Index file. Those history rows are not rewritten.

Methodology 1.1.0 scores crime from RTCI’s cleaned file (AH-Datalytics/rtci, docs/app_data/final_sample.csv on main). The rate is violent plus property incidents per 100,000 people for the latest calendar month in that file. The weekly job uses RTCI’s Nationwide Full Sample row when that row appears once, its counts are usable, and its rate matches the population-weighted total of the other usable rows (within 0.02). That is the national figure RTCI publishes. On the current file the two figures match, including the agency crime and population totals. If the row is missing, duplicated, or outside that tolerance, the job scores the population-weighted total instead and records which one it used.

The crime observation date is the last day of that month, taken from the file (for example 2026-04-30). It is not the download time, and it is not filled in for weeks that were published under 1.0.0. The normalization range stays 500 to 8,000, inverted, and the raw weight stays 0.12. If the RTCI download fails or that rate cannot be computed, the weekly job refuses to publish, the same way it refuses when another core input fails. It does not carry the previous crime value forward.

This week is methodology 1.1.0. Crime in the table is the RTCI population-weighted national rate for the latest month in the file. Weeks through 2026-10-02 remain methodology 1.0.0, with crime locked at 2,723.0.

Each metric is normalized to 0–100 using predefined thresholds:

normalized = (1 − (raw − min) / (max − min)) × 100

Trust in Government is inverted (higher trust → higher stability) by applying 100 − normalized after the above transform — equivalently a linear map of the trust percent on 0–80. An Edelman reading of 41 normalizes to 51.25. Inflation at 4% on −10…15 normalizes to 44.

The index is the weight-normalized sum of the per-metric scores (divide by the sum of weights; with all six present that denominator is 0.72). A perfect week on every metric therefore scores 100:

BOI = Σ (normalized_i × weight_i) / Σ weight_i

These four bands and cutoffs are methodology v1.1.1. They locate where today’s score sits. They do not forecast a crisis, and they are not a siren.

Range	Band	Interpretation
70 – 100	High Stability	Low risk — focus on long-term planning.
55 – 69.99	Moderate Stability	Warning signs — monitor trends; prepare contingencies.
40 – 54.99	Low Stability	Heightened risk — take this reading seriously. Conditions are pressing harder than in Moderate.
Below 40	Critical Instability	Collapse Likely — severe pressure on the published inputs. Not an instruction to bug out.

• Freshness mismatch. Inflation, unemployment, and debt-to-GDP update monthly or quarterly. Crime updates when RTCI adds a month to the cleaned file; the value is that month, not the week of the BugOut publish. The Edelman trust score updates annually. Homelessness is annual (HUD PIT). The BOI is recomputed weekly but will only move when underlying data publishes.
• National scope. From methodology 1.1.0, crime is the RTCI sample’s population-weighted national rate (the Nationwide Full Sample), not a census of every U.S. agency. Homelessness is a national point-in-time rate. Local conditions vary substantially.
• Not a forecast, and not a bug-out order. The index is a directional reading of published statistics relative to fixed endpoints. It shows how hard those conditions are pressing now. It does not say what happens next, and a lower band is not an instruction to leave.
• Not a historical crisis siren. Do not judge the index mainly by whether it would have flipped in every past recession. A mechanical FRED replay of methodology v1.1.1 — inflation, unemployment, and debt-to-GDP only, with crime, homelessness, and trust excluded or held constant and labeled — is in the source repository (results at a glance). It is not a full six-metric historical BugOut Index. In that replay, a Great Recession–shaped path stays in Moderate Stability, which is consistent with “2008 should not have been bug-out level.” A COVID-scale unemployment spike moves toward Low Stability, which is closer to “take this seriously.”
• Labor utilization is not in the score. The homepage charts prime-age employment-population ratio (FRED LNS12300060) and prime-age participation (FRED LNS11300060) as a shadow series. They are not in the BugOut Index and they have no weight. Replacing headline unemployment would be a versioned change after a backtest.
• Food prices are not in the score. The homepage charts the 12-month percent change in food CPI (FRED CPIUFDNS, BLS CUUR0000SAF1, not seasonally adjusted) as a shadow series. It is not in the BugOut Index and it has no weight. A draft 0–10% normalization in the incubating note is not applied.
• The NYC DHS shelter census is not in the score. The homepage shows the daily count of individuals in the New York City Department of Homeless Services shelter system (NYC Open Data k46n-sa2m, field total_individuals_in_shelter) as a shadow series. It is New York City only. It is not a U.S. figure, it is not the HUD AHAR homelessness rate, and it has no weight. HUD AHAR remains the annual national homelessness input.
• Other local shelter companions are not in the score. The homepage shows a U.S. regional panel: Nashville–Davidson (HMIS monthly people experiencing homelessness), Austin/Travis County (ECHO sheltered card, with publication lag), Ramsey County, Minnesota (open data 9mck-bcqu), Denver (HOST pilot occupancy rate, not a headcount), and San Francisco (DataSF kc49-udxn measure 279, a monthly occupancy rate, not a headcount). Nashville, Austin, and Denver are a manual checklist. Each one names its own place. None of them is a U.S. figure, none has a weight, and none replaces HUD AHAR. Chicago has no current public citywide census on this panel. Issue #81.
• Pew public trust is not in the score. The homepage shows the Pew Research Center share who trust the government in Washington to do what is right just about always or most of the time. It is a shadow series with no weight. It is not interchangeable with the Edelman Trust Barometer US Government input. The weekly job reads a short checklist excerpt. It does not download Pew’s full series.
• Gallup confidence in institutions is not in the score. The homepage shows the current annual “great deal” plus “quite a lot” readings for Congress, the presidency, the U.S. Supreme Court, and the 14-institution average Gallup publishes for institutions measured each year since 1993. They are a shadow series with no weight. They are not interchangeable with the Edelman trust input. The checklist stores that current year only. It is not a republication of Gallup’s historical series.
• Threshold sensitivity. Because the index is built on fixed normalization ranges, slow drift inside a range will not move the score — the thresholds are calibrated to reflect historically extreme conditions at the endpoints.

Every weekly run commits the raw inputs, the computed normalized scores, and the final index. Those committed files are the record of that week. The code that produced this page is open source. Rerunning runtime/publish/weekly_run.py fetches the latest observations, so it does not rebuild a past week’s score. Methodology version is v1.1.1.