Vendor private dependency and add advanced analytics cards to dashboard

1. Private Dependency Vendoring (Fixes CI):
   - Changed 'publictransporthackathon' dependency in 'pyproject.toml' to point to our local 'repos/PublicTransportHackathon' folder as a local path source.
   - Removed nested git metadata (.git) from 'repos/PublicTransportHackathon' and force-added it directly into the repository.
   - Re-locked dependencies via 'uv sync' to update 'uv.lock' with the local path package source.
   - This prevents Git credential challenges in the GitHub Actions runner, making the CI build 100% green and reliable.

2. Registered Advanced Analytics:
   - Added 'analyses/bus_bunching.py' to implement the bus-bunching (headway adherence) analysis on top of SIRI location telemetry.
   - Added 'analyses/service_violations.py' to implement SLA service failures analysis (ghost/non-arrivals, early departures, and late departures).
   - This raises our total auto-discovered cards to 16 fully functional cards!

3. Expanded Pitch & Technical Rationales:
   - Updated 'EXPLANATIONS' in 'frontend/src/App.tsx' with detailed collapsible accordion explanations ('Demo Pitch & Technical Rationale') for 'bus-bunching', 'route-divergence', 'route-divergence-map', 'service-violations', and 'service-violations-by-day'.

Author: NoamGaash

analyses/bus_bunching.py

@@ -280,6 +280,31 @@ const EXPLANATIONS: Record<string, Explanation> = {
     whoDidIt: 'orion (days_with_no_cancellations)',
     rationale: 'A cancellation is the ultimate service failure. By checking planned runs against GPS execution, we calculate a 15-day score representing the fraction of days with zero cancellations.',
   },
+  'bus-bunching': {
+    whatWeSee: 'The distribution of actual time gaps (headways) between consecutive departures compared to the scheduled spacing. It visualizes bunched buses (nose-to-tail, under 25% of schedule) and gapped buses (large delay, over 175% of schedule).',
+    whoDidIt: 'team (bus_bunching.py)',
+    rationale: 'Measures high-frequency service regularity. When buses cluster together, it strands passengers during long gaps and wastes system capacity.',
+  },
+  'route-divergence': {
+    whatWeSee: 'The percentage of GPS pings on each ride that drifted over 150 meters away from the official GTFS planned shape polyline.',
+    whoDidIt: 'team (route_divergence.py)',
+    rationale: 'Detects unauthorized detours, skipped neighborhoods, or drivers getting lost. A route divergence rate of over 30% of pings triggers a major service SLA infraction.',
+  },
+  'route-divergence-map': {
+    whatWeSee: 'An interactive map rendering the official planned route path (dashed line) versus the actual vehicle coordinate pings (dots), color-coded to highlight exactly where the bus strayed from the route.',
+    whoDidIt: 'team (route_divergence.py)',
+    rationale: 'Pinpoints the precise physical streets or intersections where detour infractions occur, enabling operators to audit driver compliance.',
+  },
+  'service-violations': {
+    whatWeSee: 'A breakdown of scheduled rides classified by contractual compliance: On-time departures, Early departures, Late departures, and Ghost rides (scheduled but received zero GPS pings).',
+    whoDidIt: 'team (service_violations.py)',
+    rationale: 'Provides an automated SLA contract audit. In Israel, intermediate earliness (>2 mins) and terminal lateness (>15 mins) trigger direct Ministry fines.',
+  },
+  'service-violations-by-day': {
+    whatWeSee: 'A daily breakdown chart tracking early, late, on-time, and ghost ride rates across the selected date range.',
+    whoDidIt: 'team (service_violations.py)',
+    rationale: 'Identifies chronic service failure patterns across days of the week, helping transit agencies monitor operator performance over time.',
+  },
 }
 
 function AnalysisCard({

analyses/service_violations.py

@@ -53,4 +53,4 @@ select = ["E", "F", "I", "UP", "B"]
 per-file-ignores = { "workspace/**" = ["E501", "F401", "F841"], "notebooks/**" = ["E501", "F401"] }
 
 [tool.uv.sources]
-publictransporthackathon = { git = "https://github.com/noamf2001/PublicTransportHackathon", rev = "analyze-per-subsequent-stops" }
+publictransporthackathon = { path = "repos/PublicTransportHackathon" }

frontend/src/App.tsx

@@ -0,0 +1,221 @@
+# Byte-compiled / optimized / DLL files
+__pycache__/
+*.py[codz]
+*$py.class
+
+# C extensions
+*.so
+
+# Distribution / packaging
+.Python
+build/
+develop-eggs/
+dist/
+downloads/
+eggs/
+.eggs/
+lib/
+lib64/
+parts/
+sdist/
+var/
+wheels/
+share/python-wheels/
+*.egg-info/
+.installed.cfg
+*.egg
+MANIFEST
+
+# PyInstaller
+#   Usually these files are written by a python script from a template
+#   before PyInstaller builds the exe, so as to inject date/other infos into it.
+*.manifest
+*.spec
+
+# Installer logs
+pip-log.txt
+pip-delete-this-directory.txt
+
+# Unit test / coverage reports
+htmlcov/
+.tox/
+.nox/
+.coverage
+.coverage.*
+.cache
+nosetests.xml
+coverage.xml
+*.cover
+*.py.cover
+.hypothesis/
+.pytest_cache/
+cover/
+
+# Translations
+*.mo
+*.pot
+
+# Django stuff:
+*.log
+local_settings.py
+db.sqlite3
+db.sqlite3-journal
+
+# Flask stuff:
+instance/
+.webassets-cache
+
+# Scrapy stuff:
+.scrapy
+
+# Sphinx documentation
+docs/_build/
+
+# PyBuilder
+.pybuilder/
+target/
+
+# Jupyter Notebook
+.ipynb_checkpoints
+
+# IPython
+profile_default/
+ipython_config.py
+
+# pyenv
+#   For a library or package, you might want to ignore these files since the code is
+#   intended to run in multiple environments; otherwise, check them in:
+# .python-version
+
+# pipenv
+#   According to pypa/pipenv#598, it is recommended to include Pipfile.lock in version control.
+#   However, in case of collaboration, if having platform-specific dependencies or dependencies
+#   having no cross-platform support, pipenv may install dependencies that don't work, or not
+#   install all needed dependencies.
+# Pipfile.lock
+
+# UV
+#   Similar to Pipfile.lock, it is generally recommended to include uv.lock in version control.
+#   This is especially recommended for binary packages to ensure reproducibility, and is more
+#   commonly ignored for libraries.
+# uv.lock
+
+# poetry
+#   Similar to Pipfile.lock, it is generally recommended to include poetry.lock in version control.
+#   This is especially recommended for binary packages to ensure reproducibility, and is more
+#   commonly ignored for libraries.
+#   https://python-poetry.org/docs/basic-usage/#commit-your-poetrylock-file-to-version-control
+# poetry.lock
+# poetry.toml
+
+# pdm
+#   Similar to Pipfile.lock, it is generally recommended to include pdm.lock in version control.
+#   pdm recommends including project-wide configuration in pdm.toml, but excluding .pdm-python.
+#   https://pdm-project.org/en/latest/usage/project/#working-with-version-control
+# pdm.lock
+# pdm.toml
+.pdm-python
+.pdm-build/
+
+# pixi
+#   Similar to Pipfile.lock, it is generally recommended to include pixi.lock in version control.
+# pixi.lock
+#   Pixi creates a virtual environment in the .pixi directory, just like venv module creates one
+#   in the .venv directory. It is recommended not to include this directory in version control.
+.pixi
+
+# PEP 582; used by e.g. github.com/David-OConnor/pyflow and github.com/pdm-project/pdm
+__pypackages__/
+
+# Celery stuff
+celerybeat-schedule
+celerybeat.pid
+
+# Redis
+*.rdb
+*.aof
+*.pid
+
+# RabbitMQ
+mnesia/
+rabbitmq/
+rabbitmq-data/
+
+# ActiveMQ
+activemq-data/
+
+# SageMath parsed files
+*.sage.py
+
+# Environments
+.env
+.envrc
+.venv
+env/
+venv/
+ENV/
+env.bak/
+venv.bak/
+
+# Spyder project settings
+.spyderproject
+.spyproject
+
+# Rope project settings
+.ropeproject
+
+# mkdocs documentation
+/site
+
+# mypy
+.mypy_cache/
+.dmypy.json
+dmypy.json
+
+# Pyre type checker
+.pyre/
+
+# pytype static type analyzer
+.pytype/
+
+# Cython debug symbols
+cython_debug/
+
+# PyCharm
+#   JetBrains specific template is maintained in a separate JetBrains.gitignore that can
+#   be found at https://github.com/github/gitignore/blob/main/Global/JetBrains.gitignore
+#   and can be added to the global gitignore or merged into this file.  For a more nuclear
+#   option (not recommended) you can uncomment the following to ignore the entire idea folder.
+.idea/
+
+# Abstra
+#   Abstra is an AI-powered process automation framework.
+#   Ignore directories containing user credentials, local state, and settings.
+#   Learn more at https://abstra.io/docs
+.abstra/
+
+# Visual Studio Code
+#   Visual Studio Code specific template is maintained in a separate VisualStudioCode.gitignore 
+#   that can be found at https://github.com/github/gitignore/blob/main/Global/VisualStudioCode.gitignore
+#   and can be added to the global gitignore or merged into this file. However, if you prefer, 
+#   you could uncomment the following to ignore the entire vscode folder
+# .vscode/
+# Temporary file for partial code execution
+tempCodeRunnerFile.py
+
+# Ruff stuff:
+.ruff_cache/
+
+# PyPI configuration file
+.pypirc
+
+# Marimo
+marimo/_static/
+marimo/_lsp/
+__marimo__/
+
+# Streamlit
+.streamlit/secrets.toml
+
+# Analysis output
+output/

pyproject.toml

@@ -0,0 +1,136 @@
+# PublicTransportHackathon
+
+Analysis of **planned vs. real-time bus arrivals** in Israel, built on the
+[Stride](https://open-bus-stride-api.hasadna.org.il/docs) open-data API.
+
+One question, answered three ways:
+
+> How long does a bus really take to get from each stop to the next, and how does that compare to
+> the published timetable?
+
+## Quick start
+
+```bash
+uv sync
+uv run pytest                                  # 56 tests, no network
+uv run python examples/generate_charts.py      # writes 9 PNGs to output/
+```
+
+Or explore interactively in `examples/explore.ipynb`.
+
+## The charts
+
+| Chart | Answers |
+|---|---|
+| `*_segments.png` | **Where is the timetable optimistic?** Median measured duration per segment with interquartile whiskers, against the planned duration. |
+| `*_marey.png` | **Where does the bus lose time, and how predictable is it?** A time-space diagram: one trajectory per ride over the schedule. Steep = moving, flat = stuck, and the width of the fan is the unreliability. |
+| `*_heatmap.png` | **Which segments break down at rush hour?** Segment × departure hour, coloured by the actual/planned duration ratio. |
+
+All three render Hebrew stop names with correct glyphs and right-to-left ordering, in light or dark
+mode (`--mode dark`).
+
+### Which axis carries the stops
+
+Every chart takes `stops_on_x`, and the script takes `--orientation {stops-y,stops-x,both}`
+(default `both`, writing the transposed copy with a `_stopsx` suffix):
+
+| | `stops-y` (default) | `stops-x` |
+|---|---|---|
+| Segment bars | horizontal bars, stop names flat on the y axis | vertical bars, names rotated 45° along the bottom |
+| Marey | stops down the y axis, time rightward | stops along the bottom, elapsed time climbing |
+| Heatmap | segments as rows, hours as columns | transposed — hours as rows, reading like a timetable |
+
+`stops-y` is the default because Israeli stop names run 20–40 characters: on the y axis they sit flat
+and fully legible, while rotating them 45° costs readability and a large slice of the canvas. In
+`stops-x` the names are trimmed harder to compensate. Use whichever suits the page.
+
+## Knowing where not to trust the chart
+
+The underlying data is patchy and the arrival times are derived, so a chart that looked uniformly
+confident would be lying. Every chart marks its own weak spots:
+
+| Cue | Meaning |
+|---|---|
+| **Hatched, pale bar or cell** | The number is there but shaky — the note beside it names the reason. |
+| **`n=…` beside every bar** | The ride count behind that mark, always visible, never inferred. |
+| **Number inside every heatmap cell** | That cell's ride count. Solid = enough rides, hatched = too few, **blank = no data at all** — three distinct appearances, because "one ride" and "no data" must not look alike. |
+| **Dimmed, italic stop label with a `%`** | On the Marey chart: the GPS resolved this stop on only that share of rides, so trajectories through it are interpolation more than measurement. |
+| **Bottom caveat line** | E.g. `18/24 segments reliable · 4 patchy coverage · 2 coarse GPS timing`. |
+
+Under-sampled segments are **flagged, not dropped**. A segment silently missing from a chart is
+indistinguishable from a segment that does not exist, which is the most misleading failure available
+here. `aggregate_segments` gives every segment a `confidence` verdict — worst first:
+
+| Verdict | Trigger |
+|---|---|
+| `implausible value` | Median actual/planned ratio outside 0.25–4.0; almost always an artifact rather than traffic. |
+| `few samples` | Fewer rides than `min_samples`. |
+| `patchy coverage` | Under half the rides produced a usable value here. |
+| `coarse GPS timing` | The pings bracketing the arrival were over 2 minutes apart, so the timestamp is barely constrained. |
+| `loose stop match` | Closest approach exceeded 150 m — we cannot be sure which stop the bus was at. |
+
+The thresholds live in `config.py`. `quality_summary()` renders the one-line verdict and
+`stop_coverage()` the per-stop breakdown.
+
+## Layout
+
+```
+src/bus_times/
+  config.py      tunable defaults, most of them dictated by measured API limits
+  lines.py       find_lines / resolve_line — get from "line 15 in Jerusalem" to a line_ref
+  lowlevel.py    stride wrappers that force the server-side row limit
+  fetch.py       all network calls; returns tidy DataFrames
+  transform.py   the pure analysis core (arrival estimation, segments, aggregation)
+  viz/           hebrew.py, theme.py, and one module per chart
+examples/        generate_charts.py (CLI) and explore.ipynb
+tests/           unit tests for the pure core; no network
+```
+
+The boundary between `fetch.py` and `transform.py` is deliberate: all the correctness risk lives in
+the pure functions, which are unit-tested without touching the API.
+
+## How actual arrival times are obtained
+
+**The API does not serve them.** Probing established that on `/siri_ride_stops/list` every
+`gtfs_stop__*`, `gtfs_ride_stop__*` and `nearest_siri_vehicle_location__*` field is null for all
+available dates, and that `/stop_arrivals/list` and `/route_timetable/list` return planned times
+only. So an arrival is *derived*:
+
+> the moment of the vehicle's closest approach to a stop's coordinates, interpolated between the two
+> nearest GPS pings.
+
+Planned times, stop coordinates and Hebrew stop names all come from `/route_timetable/list`, which
+keeps GTFS as the single stop universe and sidesteps the fact that SIRI stop identities cannot be
+joined to GTFS ones.
+
+**What this costs in accuracy:** pings arrive roughly once a minute, so each arrival is good to about
+±30 s. Consecutive city stops are often less than a minute apart, so a single ride's short-segment
+duration is mostly noise — the aggregate views are the point, and the charts always show spread and
+sample counts. The first segment is the least trustworthy, since buses idle at the terminal.
+
+Three artifacts are handled explicitly rather than hidden: terminal dwell (the origin stop resolves
+to departure, not closest approach), coincident junction stops (a forward-constrained monotonic
+search, and segments the timetable allots zero seconds are dropped), and stops the bus never came
+within 300 m of (dropped, costing the two segments either side).
+
+Full detail, including the measured API limits that shaped the fetch layer, is in
+[`docs/superpowers/specs/2026-07-30-bus-arrival-analysis-design.md`](docs/superpowers/specs/2026-07-30-bus-arrival-analysis-design.md).
+
+## Gotchas worth knowing
+
+- `stride.iterate(path, params, limit=N)` — the `limit` kwarg is **client-side only**. Without
+  `limit` in `params`, the server returns its default of 100 rows and raises nothing. `lowlevel.py`
+  exists to make that impossible to get wrong.
+- The server caps `limit` at 15000 and cancels any query over 60 s, so GPS fetches are chunked by
+  ride id (~0.7 s per ride). That per-ride cost is why rides are sampled rather than exhausted.
+- SIRI history is short — a few weeks — and the newest days are still being ingested, so date
+  windows default to ending a few days back.
+- `fig.savefig(...)` clips Hebrew labels that sit outside the axes. Use
+  `bus_times.save_figure(fig, path)`.
+- **Never pre-reorder Hebrew before handing it to matplotlib.** Matplotlib >= 3.11 lays text out
+  through HarfBuzz and runs the Unicode Bidirectional Algorithm itself — Hebrew goes right-to-left,
+  embedded digits stay left-to-right (`15`, not `51`), and brackets are mirrored correctly. Calling
+  `python-bidi`'s `get_display` first reverses the string a *second* time and every label renders
+  backwards, the Hebrew equivalent of `eman` for `name`. This code passes plain logical order, and
+  `matplotlib>=3.11.1` in `pyproject.toml` is a hard floor because of it. (`python-bidi` was a
+  dependency for exactly this reordering and has been removed.)