BonQuery

Data for the greater good

BonQuery turns humanitarian open data into clear, useful analysis — starting with Toronto’s shelter system capacity, demand, and flow.

Author

Miriam Marling

daily_occ = FileAttachment("data/daily_occupancy.json").json()
shelter_flow = FileAttachment("data/shelter_flow.json").json()
bayesian_gender = FileAttachment("data/bayesian_gender.json").json()

Welcome

BonQuery breathes life into humanitarian-related data — turning numbers that sit unread on government open-data pages into clear, useful analysis for the people who can do something about the issues behind them.

We use publicly available data to investigate urgent humanitarian issues. The work is open, reproducible, and built for journalists, researchers, advocates, policymakers, and anyone paying attention.

First project — Inside Toronto’s Shelter System

Toronto publishes monthly data on its taxed shelter system, but numbers alone don’t tell the story. This project closes that gap with headache-free takeaways: what the numbers mean, why they matter, and what the data leaves out.

Explore the project →

Monthly Shelter System Flow

Toronto publishes monthly data on its taxed shelter system, and the City’s own dashboards do a good job presenting the headline numbers. This section goes a step further: interactive filters on every chart, a clear note on what the data captures and what it leaves out, and findings that usually stay buried in technical documentation.

The data covers every month from January 2018 to the most recent published month, sourced directly from the City of Toronto’s Shelter System Flow dataset. New data loads automatically each month, usually around the 15th, after the City publishes the prior month’s figures.

The dashboards

Open each dashboard directly in your browser — they’re built for interactive use and work best full-screen on a phone.

The three tabs below show different views of the data, with interactive filters on every chart.

A point-in-time view of the shelter system for any month from January 2018 forward. Use the date filter at the top to change months.

Open in a new tab →

Compares the current year’s cumulative figures against the same period the prior year. Useful for seeing whether the system is on track to exceed, match, or fall below last year.

Open in a new tab →

Multi-year time series for the headline metrics, with filters for population group, year, and month. Useful for spotting long-term patterns or seasonal trends.

Open in a new tab →

How to read this data

These numbers reflect the shelter system, not all homelessness. The data captures people who used a City-funded overnight service at least once in the past three months. It doesn’t include people sleeping outside, in shelters that aren’t City-funded, or staying temporarily with friends or family. The City estimates that roughly 18% of people experiencing absolute homelessness in Toronto aren’t reflected in these numbers. Trends here reflect changes in the shelter system specifically — not in total homelessness in the city.

Key terms

The dashboards use a few terms that have specific meanings worth knowing before you read the charts.

Actively homeless. A person who has used City-funded shelter services at least once in the past three months and hasn’t been recorded as moving to permanent housing. This is the headline number on every dashboard. Because it looks back three months, the count includes people who may not be in shelter on any given night.

Chronic homelessness. The federal definition: someone who has spent at least 180 nights in shelter over the past year, or at least 546 nights over the past three years. A person can meet this definition while still actively in the shelter system — they don’t need to have left and returned.

Inflow categories (people entering the shelter system this month):

Newly Identified. People entering the shelter system for the first time. One exception for the “Chronic” group: in that row, this column counts people who became chronically homeless during the reporting month, regardless of how long they’d already been using the shelter system.
Returned from Permanent Housing. People who previously moved to permanent housing and have come back to the shelter system.
Returned to Shelter. People who were in the system, didn’t use it for at least three months, and have now returned.

Outflow categories (people leaving the shelter system this month):

Moved to Permanent Housing. People who left the shelter system for permanent housing.
Became Inactive. People who haven’t used shelter services in the past three months, including the reporting month.

Definitions are based on the City of Toronto’s Shelter System Flow Data page.

Read all data findings →

Data source

All data comes from the City of Toronto’s Shelter System Flow dataset, published monthly on the City’s open data portal.

Contains information licensed under the Open Government Licence – Toronto.

Daily Occupancy & Capacity

function fmtDate(s) {
  const [y, m, d] = s.split("-").map(Number)
  return new Date(y, m - 1, d).toLocaleDateString("en-US", {
    year: "numeric", month: "long", day: "numeric"
  })
}

allDates = [...new Set(daily_occ.map(r => r.date))].sort().reverse()
latestDate = allDates[0]
cityValidation = FileAttachment("data/city_validation.json").json().catch(() => null)

BonQuery’s version of this table adds one feature the City’s page doesn’t have: use the date selector below to browse any date from January 2021 to the present. The City’s Daily Shelter & Overnight Service Usage page shows only its most recently published date, which doesn’t always update on schedule.

Daily Occupancy & Capacity for

latestRows = daily_occ.filter(r => r.date === latestDate)

function getVal(rows, key) {
  const row = rows.find(r => r.key === key)
  return row ? row.ind : null
}

validationChecks = {
  const rows = latestRows
  const checks = []

  // Check 1: all expected keys present
  const expectedKeys = [
    "all_shelter","room_based","singles_sector","singles_shelter","allied_summ",
    "temp_summ","iso_summ","fam_total","fam_emerg","fam_trans_r","fam_hotel",
    "sng_hotel","singles_total","emerg_total","mix_emerg","men_emerg","wom_emerg",
    "yth_emerg","trans_total","mix_trans","fam_trans_b","men_trans","wom_trans",
    "yth_trans","allied_total","respites","dropin","temp_resp","hotels","iso"
  ]
  const presentKeys = new Set(rows.map(r => r.key))
  const missingKeys = expectedKeys.filter(k => !presentKeys.has(k))
  checks.push({
    name: "All expected rows present",
    pass: missingKeys.length === 0,
    detail: missingKeys.length > 0 ? `Missing: ${missingKeys.join(", ")}` : null
  })

  // Check 2: All Shelter Programs total = sum of components
  const allShelter = getVal(rows, "all_shelter")
  const compSum = ["fam_total","sng_hotel","singles_total","allied_total","temp_resp","hotels","iso"]
    .reduce((s, k) => s + (getVal(rows, k) ?? 0), 0)
  checks.push({
    name: "All Shelter Programs rollup matches components",
    pass: allShelter === compSum,
    detail: allShelter !== compSum
      ? `all_shelter.ind=${allShelter}, component sum=${compSum}, diff=${allShelter - compSum}`
      : null
  })

  // Check 3: Singles sectors total = emergency + transitional
  const singlesTotal = getVal(rows, "singles_total")
  const emergTotal = getVal(rows, "emerg_total")
  const transTotal = getVal(rows, "trans_total")
  checks.push({
    name: "Singles sectors rollup: emergency + transitional = singles total",
    pass: singlesTotal === (emergTotal + transTotal),
    detail: singlesTotal !== (emergTotal + transTotal)
      ? `singles=${singlesTotal}, emerg+trans=${emergTotal + transTotal}`
      : null
  })

  // Check 4: For every row with non-null cap, occ + unocc = cap (within 1, for rounding)
  const capRows = rows.filter(r => r.cap !== null && r.cap > 0)
  const arithFails = capRows.filter(r => Math.abs(r.occ + r.unocc - r.cap) > 0.5)
  checks.push({
    name: "Internal arithmetic: occupied + unoccupied = actual capacity",
    pass: arithFails.length === 0,
    detail: arithFails.length > 0
      ? arithFails.map(r => `${r.key}: ${r.occ}+${r.unocc}≠${r.cap}`).join("; ")
      : null
  })

  return checks
}

allPass = validationChecks.every(c => c.pass)

// ── Validation banner ─────────────────────────────────────────────────────
validationBanner = {
  const CITY_CENSUS_URL = "https://www.toronto.ca/city-government/data-research-maps/research-reports/housing-and-homelessness-research-and-reports/shelter-census/"

  if (!allPass) {
    const failedChecks = validationChecks.filter(c => !c.pass)
    return html`<div style="
      background:#fff3cd;border:1px solid #e0a800;border-radius:4px;
      padding:8px 14px;font-size:0.88em;color:#533f03;margin-bottom:12px">
      <strong>⚠ Validation issue detected for ${fmtDate(latestDate)}</strong>
      <details style="margin-top:6px">
        <summary style="cursor:pointer">Show details</summary>
        <ul style="margin:6px 0 0;padding-left:20px">
          ${failedChecks.map(c => html`<li>${c.name}${c.detail ? html` — <em>${c.detail}</em>` : ""}</li>`)}
        </ul>
      </details>
    </div>`
  }
  if (cityValidation === null) {
    return html`<div style="
      background:#e9ecef;border:1px solid #adb5bd;border-radius:4px;
      padding:8px 14px;font-size:0.88em;color:#495057;margin-bottom:12px">
      Validation status unavailable in local development
    </div>`
  }
  if (!cityValidation.city_reachable) {
    return html`<div style="
      background:#fff3cd;border:1px solid #e0a800;border-radius:4px;
      padding:8px 14px;font-size:0.88em;color:#533f03;margin-bottom:12px">
      ⚠ City of Toronto shelter census page could not be reached · <a href="${CITY_CENSUS_URL}" target="_blank">View City data →</a>
    </div>`
  }
  if (cityValidation.passed && cityValidation.city_date === cityValidation.bonquery_latest_date) {
    return html`<div style="
      background:#d4edda;border:1px solid #76b88d;border-radius:4px;
      padding:8px 14px;font-size:0.88em;color:#1a4731;margin-bottom:12px">
      ✓ Validated against <a href="${CITY_CENSUS_URL}" target="_blank">City's Daily Shelter &amp; Overnight Service Usage page</a> for ${fmtDate(cityValidation.city_date)}
    </div>`
  }
  if (cityValidation.passed && cityValidation.city_date !== cityValidation.bonquery_latest_date) {
    return html`<div style="
      background:#d4edda;border:1px solid #76b88d;border-radius:4px;
      padding:8px 14px;font-size:0.88em;color:#1a4731;margin-bottom:12px">
      ✓ Last validated against <a href="${CITY_CENSUS_URL}" target="_blank">City's Daily Shelter &amp; Overnight Service Usage page</a> for ${fmtDate(cityValidation.city_date)} · BonQuery data current to ${fmtDate(cityValidation.bonquery_latest_date)}.
    </div>`
  }
  return html`<div style="
    background:#fff3cd;border:1px solid #e0a800;border-radius:4px;
    padding:8px 14px;font-size:0.88em;color:#533f03;margin-bottom:12px">
    ⚠ Discrepancy found for ${fmtDate(cityValidation.city_date)} — BonQuery values differ from <a href="${CITY_CENSUS_URL}" target="_blank">City's published table</a>
  </div>`
}

viewof selectedDate = Inputs.select(allDates, {
  label: "Date",
  format: d => fmtDate(d)
})

dayRows = daily_occ.filter(r => r.date === selectedDate)

function n(v) { return v !== null && v !== undefined ? v.toLocaleString() : "—" }
function pct(v) { return v !== null && v !== undefined ? v.toFixed(1) + "%" : "—" }

function tableRow(r, col1Label) {
  const pad = (r.indent ?? 0) * 16
  const bold = r.is_total ? "font-weight:600" : ""
  const bg   = r.is_total ? "#e8edf5" : "#ffffff"
  if (r.col_type === "summary") {
    return `<tr style="background:${bg}">
      <td style="padding:5px 10px 5px ${pad}px;${bold}">${r.label}</td>
      <td style="padding:5px 10px;text-align:right;${bold}">${n(r.ind)}</td>
      <td></td><td></td><td></td><td></td>
    </tr>`
  }
  return `<tr style="background:${bg}">
    <td style="padding:5px 10px 5px ${pad}px;${bold}">${r.label}</td>
    <td style="padding:5px 10px;text-align:right;${bold}">${n(r.ind)}</td>
    <td style="padding:5px 10px;text-align:right">${n(r.occ)}</td>
    <td style="padding:5px 10px;text-align:right">${n(r.unocc)}</td>
    <td style="padding:5px 10px;text-align:right">${n(r.cap)}</td>
    <td style="padding:5px 10px;text-align:right">${pct(r.rate)}</td>
  </tr>`
}

function sectionHeader(label, col1, col2, col3) {
  return `<tr style="background:#1a3c6b;color:white;font-weight:600;font-size:0.85em">
    <td style="padding:7px 10px">${label}</td>
    <td style="padding:7px 10px;text-align:right">Individuals</td>
    <td style="padding:7px 10px;text-align:right">${col1}</td>
    <td style="padding:7px 10px;text-align:right">${col2}</td>
    <td style="padding:7px 10px;text-align:right">${col3}</td>
    <td style="padding:7px 10px;text-align:right">Occupancy Rate</td>
  </tr>`
}

function summaryHeader() {
  return `<tr style="background:#1a3c6b;color:white;font-weight:600;font-size:0.85em">
    <td style="padding:7px 10px">Occupancy Summary</td>
    <td style="padding:7px 10px;text-align:right">Individuals</td>
    <td colspan="4"></td>
  </tr>`
}

dailyTable = {
  const s = (key) => dayRows.find(r => r.key === key)

  const allShelterInd = s("all_shelter") ? s("all_shelter").ind : null
  const totalAccomRow = `<tr style="background:#e8edf5;font-weight:600">
    <td style="padding:5px 10px">Total People Accommodated †</td>
    <td style="padding:5px 10px;text-align:right">${n(allShelterInd)}</td>
    <td></td><td></td><td></td><td></td>
  </tr>`
  const bridgingRow = `<tr style="background:#f5f5f5;color:#999;font-style:italic">
    <td style="padding:5px 10px 5px 26px">Bridging &amp; Triage Programs</td>
    <td style="padding:5px 10px;text-align:right;font-size:0.85em">Not published to open data</td>
    <td></td><td></td><td></td><td></td>
  </tr>`

  const summaryKeys = ["all_shelter","room_based","singles_sector","singles_shelter","allied_summ","temp_summ","iso_summ"]
  const roomKeys    = ["fam_total","fam_emerg","fam_trans_r","fam_hotel","sng_hotel"]
  const bedKeys     = ["singles_total","emerg_total","mix_emerg","men_emerg","wom_emerg","yth_emerg",
                       "trans_total","mix_trans","fam_trans_b","men_trans","wom_trans","yth_trans"]
  const alliedKeys  = ["allied_total","respites","dropin"]

  return html`
    <div style="overflow-x:auto;margin:0 0 1rem">
    <table style="border-collapse:collapse;width:100%;font-size:0.88em;font-family:'Segoe UI',Arial,sans-serif;min-width:620px">
      ${summaryHeader()}
      ${totalAccomRow}
      ${bridgingRow}
      ${summaryKeys.map(k => s(k) ? tableRow(s(k)) : "").join("")}

      ${sectionHeader("Shelter Programs, Room-Based", "Occupied Rooms", "Unoccupied Rooms", "Actual Room Capacity")}
      ${roomKeys.map(k => s(k) ? tableRow(s(k)) : "").join("")}

      ${sectionHeader("Shelter Programs, Bed-Based", "Occupied Beds", "Unoccupied Beds", "Actual Bed Capacity")}
      ${bedKeys.map(k => s(k) ? tableRow(s(k)) : "").join("")}

      ${sectionHeader("Allied Services", "Occupied Spaces", "Unoccupied Spaces", "Actual Space Capacity")}
      ${alliedKeys.map(k => s(k) ? tableRow(s(k)) : "").join("")}

      <tr style="background:#1a3c6b;color:white;font-weight:600;font-size:0.85em">
        <td style="padding:7px 10px">Temporary Programs, Bed/Space-Based</td>
        <td style="padding:7px 10px;text-align:right">Individuals</td>
        <td style="padding:7px 10px;text-align:right">Occupied Beds/Spaces</td>
        <td style="padding:7px 10px;text-align:right">Unoccupied</td>
        <td style="padding:7px 10px;text-align:right">Actual Capacity</td>
        <td style="padding:7px 10px;text-align:right">Occupancy Rate</td>
      </tr>
      ${s("temp_resp") ? tableRow(s("temp_resp")) : ""}

      ${sectionHeader("Temporary Programs, Room/Unit-Based", "Occupied Rooms", "Unoccupied Rooms", "Actual Room Capacity")}
      ${s("hotels") ? tableRow(s("hotels")) : ""}

      ${sectionHeader("Temporary Isolation/Recovery Programs", "Occupied Rooms", "Unoccupied Rooms", "Actual Room Capacity")}
      ${s("iso") ? tableRow(s("iso")) : ""}
    </table>
    <p style="font-size:0.8em;color:#666;margin:4px 0 0">
      † Total People Accommodated excludes Bridging &amp; Triage programs,
      which are tracked in the City's operational system but not published
      to open data.
    </p>
    </div>`
}

This table reproduces the City of Toronto’s Daily Shelter & Overnight Service Usage report using the City’s open data. Two known differences from the City’s live dashboard: it does not include Bridging & Triage programs, which are not published to open data (29 people on May 14, 2026); and for emergency programs, “Actual Bed Capacity” counts beds currently available for use and excludes beds temporarily offline, which the City’s live dashboard includes. Occupancy and individual counts match the City’s figures exactly. A full side-by-side comparison is available on the data validation page.

Contains information licensed under the Open Government Licence – Toronto. Source: City of Toronto Open Data — Daily Shelter & Overnight Service Occupancy & Capacity. Aggregated by BonQuery.

BonQuery’s aggregated data is available for download from 2021 to present: JSON · CSV

Historical Occupancy Spike Detection

These spike-detection charts flag days when Toronto’s shelter system accommodated a lot more people than its seasonal pattern and long-run trend would’ve predicted, not merely days that happened to be busy. Spikes are identified by removing two predictable forces (seasonality and long-run trend) with Seasonal-Trend decomposition using LOESS (STL) and measuring what is left over.

The amber band marks moderate anomalies (2 MAD-SD above the expected value), the red band marks pronounced ones (3 MAD-SD). The analysis counts individuals accommodated rather than the occupancy rate, and both thresholds are calibrated against the system’s own behaviour over the preceding two years. It is retrospective and descriptive, not predictive. Full methodology and caveats are on the linked page.

Historical Occupancy Spike Detection →

Open in a new tab →

Contains information licensed under the Open Government Licence – Toronto. Source: City of Toronto Open Data — Daily Shelter & Overnight Service Occupancy & Capacity. Analysis and charts by BonQuery.

KEY FINDING

The most affected group — and the one we hear least about

_allFlow = shelter_flow
  .filter(d => d.population_group === "All Population"
            && d.gender_male != null && d.gender_female != null)

// Most recent month — for the hero ratio and sub-line
_latestFlow = _allFlow.sort((a, b) => b.flow_date.localeCompare(a.flow_date))[0]

_menCount   = _latestFlow ? _latestFlow.gender_male : 0
_womenCount = _latestFlow ? _latestFlow.gender_female : 0
_transCount = _latestFlow ? _latestFlow.gender_trans_nb_two_spirit : 0
_ratio      = _womenCount > 0 ? (_menCount / _womenCount).toFixed(1) : "—"
_flowMonth  = _latestFlow
  ? new Date(_latestFlow.flow_date + "T00:00:00")
      .toLocaleString("default", {month: "long", year: "numeric"})
  : ""

// --- Time-series statistics ---
// These are population counts, not random samples, so a single month carries no
// sampling error. The meaningful uncertainty is month-to-month volatility.
// We treat each month as one observation and compare group means across months.

function _mean(arr) { return arr.reduce((a, b) => a + b, 0) / arr.length }
function _sd(arr) {
  const m = _mean(arr)
  return Math.sqrt(arr.reduce((s, x) => s + (x - m) ** 2, 0) / (arr.length - 1))
}
function _se(arr) { return _sd(arr) / Math.sqrt(arr.length) }

// Normal CDF — Abramowitz & Stegun 26.2.17, accurate to ~7 dp.
// Used for Bonferroni-adjusted two-tailed p-values.
function _Phi(z) {
  const c = [0.319381530, -0.356563782, 1.781477937, -1.821255978, 1.330274429]
  const t = 1 / (1 + 0.2316419 * Math.abs(z))
  const tail = t*(c[0]+t*(c[1]+t*(c[2]+t*(c[3]+t*c[4])))) * Math.exp(-z*z/2) / Math.sqrt(2*Math.PI)
  return z >= 0 ? 1 - tail : tail
}

_menSeries   = _allFlow.map(d => d.gender_male)
_womenSeries = _allFlow.map(d => d.gender_female)
_transSeries = _allFlow.map(d => d.gender_trans_nb_two_spirit).filter(v => v != null)

_menMean   = _mean(_menSeries);   _menSE   = _se(_menSeries)
_womenMean = _mean(_womenSeries); _womenSE = _se(_womenSeries)
_transMean = _mean(_transSeries); _transSE = _se(_transSeries)

_n1 = _menSeries.length;   _s1 = _sd(_menSeries)
_n2 = _womenSeries.length; _s2 = _sd(_womenSeries)
_n3 = _transSeries.length; _s3 = _sd(_transSeries)

// Welch's t-test helper — unequal variances, returns {t, df}
function _wt(m1, s1, n1, m2, s2, n2) {
  const v1 = s1**2/n1, v2 = s2**2/n2
  const t  = (m1 - m2) / Math.sqrt(v1 + v2)
  const df = (v1 + v2)**2 / (v1**2/(n1-1) + v2**2/(n2-1))
  return {t, df}
}

// All three pairwise tests
_pw_mw = _wt(_menMean,   _s1, _n1, _womenMean, _s2, _n2)  // Men   vs Women
_pw_mt = _wt(_menMean,   _s1, _n1, _transMean, _s3, _n3)  // Men   vs Trans/NB/2S
_pw_wt = _wt(_womenMean, _s2, _n2, _transMean, _s3, _n3)  // Women vs Trans/NB/2S

// Bonferroni-adjusted two-tailed p (3 comparisons → multiply by 3, cap at 1)
function _pAdj(t) { return Math.min(1, 3 * 2 * _Phi(-Math.abs(t))) }
function _pLabel(t) {
  const p = _pAdj(t)
  return p < 0.0001 ? "< 0.0001 ***" : p < 0.001 ? "< 0.001 **" : p < 0.05 ? p.toFixed(3) + " *" : p.toFixed(3)
}

_yearRange = _allFlow.length > 0
  ? _allFlow[_allFlow.length - 1].flow_date.slice(0, 4) + "–" + _latestFlow.flow_date.slice(0, 4)
  : ""

// Minimum monthly share of men, used in the closing paragraph.
// Math.floor ensures the displayed integer is a true lower bound: every
// month on record had men's share above this number.
_menShareSeries = _allFlow.map(d => d.gender_male / d.actively_homeless)
_menShareMinPct = Math.floor(Math.min(..._menShareSeries) * 100)

// Average monthly gap: men minus (women + transgender/NB/2S), full record.
_gapAvg    = Math.round(_menMean - (_womenMean + _transMean))
_gapAvgStr = _gapAvg.toLocaleString()

// Earliest and latest years in the data, for the "on record (X to Y)" phrase.
_recordStartYear = _allFlow.map(d => d.flow_date.slice(0, 4)).sort()[0]
_recordEndYear   = _latestFlow.flow_date.slice(0, 4)

html`<div class="crisis-hero">
  <div class="hero-number">${_ratio}×</div>
  <div class="hero-text">
    <div class="hero-label">more <strong>men</strong> than women are actively homeless in Toronto's shelter system</div>
    <div class="hero-sub"><strong>${_menCount.toLocaleString()} men</strong> · ${_womenCount.toLocaleString()} women · ${_transCount.toLocaleString()} transgender, non-binary, or two-spirit · ${_flowMonth}</div>
  </div>
</div>`

_genderStats = [
  {label: "Men",          mean: _menMean,   ci_lo: _menMean   - 1.96 * _menSE,   ci_hi: _menMean   + 1.96 * _menSE,   isFocus: true},
  {label: "Women",        mean: _womenMean, ci_lo: _womenMean - 1.96 * _womenSE, ci_hi: _womenMean + 1.96 * _womenSE, isFocus: false},
  {label: "Trans/NB/2S",  mean: _transMean, ci_lo: _transMean - 1.96 * _transSE, ci_hi: _transMean + 1.96 * _transSE, isFocus: false}
].sort((a, b) => b.mean - a.mean)

Plot.plot({
  width, height: 300, marginBottom: 36, marginLeft: 70, marginRight: 20,
  title: `Mean monthly actively homeless count by gender — ${_yearRange} (${_n1} months)`,
  caption: "Error bars show 95% confidence intervals of the monthly mean. Source: City of Toronto Shelter System Flow dataset (open.toronto.ca).",
  x: {label: null, domain: _genderStats.map(d => d.label)},
  y: {label: "Mean individuals actively homeless / month", grid: true,
      tickFormat: d => d.toLocaleString(),
      domain: [0, Math.max(..._genderStats.map(d => d.ci_hi)) * 1.12]},
  marks: [
    // Bars — mean monthly count
    Plot.barY(_genderStats, {
      x: "label", y: "mean",
      fill: d => d.isFocus ? "#ff3b30" : "#4a6fa5",
      fillOpacity: d => d.isFocus ? 1 : 0.55
    }),
    // Vertical error bar line (ci_lo → ci_hi)
    Plot.ruleY(_genderStats, {
      x: "label", y1: "ci_lo", y2: "ci_hi",
      stroke: "#222", strokeWidth: 1.8
    }),
    // Horizontal cap at ci_hi (top)
    Plot.tickY(_genderStats, {
      x: "label", y: "ci_hi",
      stroke: "#222", strokeWidth: 1.8,
      insetLeft: 18, insetRight: 18
    }),
    // Horizontal cap at ci_lo (bottom)
    Plot.tickY(_genderStats, {
      x: "label", y: "ci_lo",
      stroke: "#222", strokeWidth: 1.8,
      insetLeft: 18, insetRight: 18
    }),
    // Value label above each bar
    Plot.text(_genderStats, {
      x: "label", y: "ci_hi",
      text: d => Math.round(d.mean).toLocaleString(),
      dy: -10, textAnchor: "middle", fontSize: 13,
      fill: d => d.isFocus ? "#cc1f15" : "#333",
      fontWeight: "600"
    }),
    Plot.ruleY([0])
  ]
})

{
  // Bayesian results from brms model (pre-computed, saved in data/bayesian_gender.json)
  const bg   = bayesian_gender
  const rr   = bg.rate_ratios
  const pd   = bg.pairwise_diffs
  const diag = bg.diagnostics
  const fmt1 = x => Number(x).toFixed(1)
  const fmtN = x => Math.round(x).toLocaleString()

  const bayesPairs = [
    {comp: "Men vs Women",         rr: rr.men_vs_women,       diff: pd.men_vs_women,       note: false},
    {comp: "Men vs Trans/NB/2S",   rr: rr.men_vs_transnb2s,   diff: pd.men_vs_transnb2s,   note: true},
    {comp: "Women vs Trans/NB/2S", rr: rr.women_vs_transnb2s, diff: pd.women_vs_transnb2s, note: true}
  ]
  const bayesRows = bayesPairs.map(p => html`<tr>
    <td style="padding:3px 10px 3px 0">${p.comp}${p.note ? html`<sup> †</sup>` : ""}</td>
    <td style="padding:3px 10px;text-align:right;font-variant-numeric:tabular-nums">
      ${fmt1(p.rr.median)}&times; (${fmt1(p.rr.hdi[0])}&ndash;${fmt1(p.rr.hdi[1])})
    </td>
    <td style="padding:3px 0;text-align:right;font-variant-numeric:tabular-nums;font-weight:${p.note?"400":"700"}">
      ${fmtN(p.diff.median)} (${fmtN(p.diff.hdi[0])}&ndash;${fmtN(p.diff.hdi[1])})
    </td>
  </tr>`)

  return html`<div style="font-size:0.81rem;color:#444;margin:0.6rem 0 0;line-height:1.6">
    <p style="margin:0 0 0.3rem">
      <strong>Bayesian analysis (negative-binomial regression, fit with
      <code>brms</code> ${bg.brms_version} on ${bg.n_months} months of data):</strong>
      Weakly informative priors were used throughout.
      Sampling diagnostics: R&#x0302;&nbsp;${diag.max_rhat},
      ESS&nbsp;${diag.min_ess_bulk.toLocaleString()},
      ${diag.n_divergent} divergent transitions.
    </p>
    <table style="border-collapse:collapse;width:100%;font-size:0.8rem;margin-bottom:0.3rem">
      <thead><tr style="border-bottom:1px solid #ccc;color:#666">
        <th style="text-align:left;padding:3px 10px 3px 0;font-weight:600">Comparison</th>
        <th style="text-align:right;padding:3px 10px;font-weight:600">Rate ratio (95% HDI)</th>
        <th style="text-align:right;padding:3px 0;font-weight:600">Difference in count (95% HDI)</th>
      </tr></thead>
      <tbody>${bayesRows}</tbody>
    </table>
    <p style="margin:0 0 0.5rem;color:#777">
      † The Trans/NB/2S group averages ~${Math.round(_transMean)} individuals per month,
      a small count that is also likely under-reported in administrative shelter records.
      Tests involving this group should be interpreted with caution.
    </p>
    <p style="margin:0 0 0.3rem">
      <strong>Men vs Women:</strong>
      Men's expected monthly count is roughly
      <strong>${fmt1(rr.men_vs_women.median)}&times; that of women</strong>
      (95% HDI ${fmt1(rr.men_vs_women.hdi[0])}&ndash;${fmt1(rr.men_vs_women.hdi[1])}&times;),
      an average gap of about
      <strong>${fmtN(pd.men_vs_women.median)} people per month</strong>
      (95% HDI ${fmtN(pd.men_vs_women.hdi[0])}&ndash;${fmtN(pd.men_vs_women.hdi[1])}).
      Across a wide range of reasonable priors, the posterior probability that the true mean
      monthly count of men exceeds women is indistinguishable from 1.
    </p>
    <div class="plain-eng">
      <div class="pe-label">Plain English</div>
      A posterior probability indistinguishable from 1 means the estimated probability is
      close to 100% that men's true monthly count exceeds women's. The 95% HDI also indicates
      the gap is consistently large — men's running about ${fmt1(rr.men_vs_women.median)}&times;
      women's, or roughly ${fmtN(pd.men_vs_women.median)} more people per month.
      <strong>Men are the largest and most consistently over-represented group in the city's shelter system.</strong>
    </div>
  </div>`
}

md`Men have accounted for at least **${_menShareMinPct}%** of Toronto's actively homeless population every single month on record (${_recordStartYear} to ${_recordEndYear}). That figure is the **lowest single-month share of men observed across the entire data record**; in most months the share is higher. The average monthly gap between men and everyone else (women plus transgender, non-binary, and two-spirit individuals combined) is roughly **${_gapAvgStr} people**, and it has barely moved as the total has risen and fallen. The pattern is not new, not subtle, and not hidden in the data. Yet it rarely makes headlines, and few of the city's public conversations about homelessness lead with it.`

The impact on men is also highlighted here: Occupancy Spike Detection →

Have a humanitarian dataset you think should get this treatment? Reach out at info@BonQuery.ca.