Busabase
Busabase 是一个开源(MIT)数据库与 AI 智能体工作空间——将数据、文档、技能和应用集成于一体。您可以本地运行、在网络内自托管或使用 Busabase Cloud,从而开启新时代。
AI 翻译
Школьное исследование: агентная модель дневного присутствия людей и индекс разрыва TSG — где в Талдыкоргане не хватает услуг в тот час, когда там люди. Предварительные результаты.
翻译准备中
数据GitHub @imangeldinaamina-cloud在线运营者未确认
试用过了吗? 点一下就好,无需登录。
尚未收集仓库信息。
尚未检测到更新。GitHub 发布版本后会显示在这里。
TSG Map — Temporal Service Gap index for Taldykorgan: agent-based model, video check (YOLOv8) and a web map for choosing business locations by time of day
翻译准备中
正在加载完整 README。
查看完整 README ↗A web app for Taldykorgan (Kazakhstan) that shows where and when services are missing, using the Temporal Service Gap index:
TSG = (D − S) / (D + S) range −1 … +1, positive = shortage
The app has two parts:
/city) — the spatial-intelligence workspace: WorldPop population on a 250 m grid,
OSM facilities and walk network, a configurable temporal presence model, the Service Gap Index and a
candidate-location comparison. Data sources and limitations: /city/data./map) — observation points measured in the field (camera counts), ranking and admin import.Stack: Next.js 15 (App Router, TypeScript) · Tailwind CSS 4 · MapLibre GL + OSM raster tiles · Supabase (Postgres + admin auth) · Vitest.
| Part | Status | Where | Notes |
|---|---|---|---|
| TSG formula | ✅ Real | src/lib/tsg.ts | Tested: D=S=0 → undefined (null), only D → +1, only S → −1 |
| Services (S) | ✅ Real, OSM | src/data/osm/services.json | 575 places, extracted 2026-11-05 via an Overpass mirror; OSM snapshot date 2026-06-01 (the mirror lags). OSM coverage is incomplete. Re-run npm run osm:fetch. |
| People present (D, dynamic) | ⏳ No real data yet | admin import → observations table | Imported via /admin CSV. Until then the map shows demo data. |
| Demo points + counts | ⚠️ DEMO, synthetic | src/data/demo/demo-data.ts | 10 arbitrary locations, invented numbers. Always shown with a "DEMO DATA" banner and badges. Never mixed with real data. |
| Residents (D, static) | ⚠️ PLACEHOLDER | src/data/population.ts | All null for real points → static mode shows "population not provided". Demo points have demo population. |
| Detector accuracy | ⚠️ Not verified | methodology page | YOLOv8n counts not yet manually checked. |
| Validation (holdout, block CV, survey) | 📝 Plan only | methodology page | No results yet. |
| Feedback, analytics, waitlist | ✅ Real (stored) | Supabase or .data/ locally | No personal data except waitlist email with consent. |
| City analysis | |||
| City boundary | ✅ Real, OSM | public/data/taldykorgan/boundary.json | OSM relation 3392135 (city akimat, admin_level 6) — includes surrounding settlements. |
| Walk network | ✅ Real, OSM | roads.json, src/data/city/graph.json | 5 867 walkable ways, ~2 055 km, largest connected component (OSM snapshot 2026-11-05). Street centrelines, sidewalks rarely mapped. |
| Facilities (6 categories) | ✅ Real, OSM | facilities.json | 585 objects; tag rules in scripts/build-city.mjs and src/lib/city/types.ts. OSM is incomplete (e.g. only 44 pharmacies) → gaps are overstated. |
| Residents | ✅ Real, WorldPop | grid.json | WorldPop Global2 R2025A constrained 100 m, 2021 (CC BY 4.0): 176 584 people in the boundary, summed into 2 475 cells of 250 m. A model, not a count. |
| Census 2021 total | ⏳ Not loaded | model settings | Optional proportional rescaling once you enter the verified figure. |
| Presence by time of day | ✅ Agent model (default) | data/agent-model/, agent-presence.json | The author's agent-based model (≈17 k agents, 100 runs) laid onto the 250 m grid (npm run city:agent). Users can switch to their own assumptions (src/lib/city/model.ts). Video check (H2) not confirmed yet — see research/H2_REPORT.md. |
| Walking access | ✅ Computed | access.json | Dijkstra on the walk network, ≤1 200 m, from cell centres. |
| Service Gap Index | ✅ Computed | src/lib/city/gap.ts | TSG formula on city-wide shares; tested. |
| Location assessment | ✅ Rule-based | src/lib/city/location.ts, /api/location | Explainable comparisons with numbers; no revenue prediction. |
| Kazakh translation | ⚠️ Draft | src/i18n/kk.ts | Needs review by a native speaker. |
| Payments | ❌ Not built (by design) | — | Freemium is a test: "Pro" gate + waitlist + click counting + free preview. |
npm run data:all rebuilds everything the app serves from raw open data
(OSM → grid, walk graph, facilities; WorldPop → residents; street index; agent model → grid).
The agent model itself: research/agent_model/ (Python, see its README).npm test (Vitest) — TSG formula properties, gap index, street search, agent-model totals,
weighted median; Python property test research/agent_model/code/test_tsg_properties.py.research/h2_calibrate.py — model vs video (H2) with bootstrap CIs and the video→schedule calibration loop
with leave-one-point-out validation → research/H2_REPORT.md.research/yolo_error.py + research/manual-count/index.html — YOLOv8 accuracy against a manual count.research/entrepreneur_check.py — precision/recall of TSG classes against entrepreneurs' "good/bad" places.以上为 README 的开头部分。其余内容请阅读原文。 查看完整 README ↗