mirror of
https://github.com/Hestia-Homes/Model.git
synced 2026-07-12 13:29:04 +00:00
2f0eb49eee
- Add SAP-accuracy sample for uprn_10093116543 (epc.json, elmhurst_inputs.md, summary/worksheet PDFs) - Persist hyde viewer stack (xvfb/fluxbox/x11vnc/novnc/websockify) and Playwright chromium in the backend devcontainer; forward noVNC 6080 - Broaden .claude/settings.local.json allowlist (display/python/grep/tail) - In-progress campaign mapper/cert_to_inputs work carried from prior cert Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
71 lines
4 KiB
Markdown
71 lines
4 KiB
Markdown
# Elmhurst RdSAP entry-tool automation (POC)
|
|
|
|
Automates most of the one manual step in `validate-cert-sap-accuracy`: re-keying
|
|
`elmhurst_inputs.md` into Elmhurst's web RdSAP entry tool and exporting the
|
|
**Input Summary** + **SAP Worksheets** PDFs (which `compare_epc_paths.py` then
|
|
consumes). Built on Playwright (Python).
|
|
|
|
Proven end-to-end on UPRN 10092973954: filled most pages automatically,
|
|
downloaded both PDFs, and `compare_epc_paths.py` reconciled to **our engine 77
|
|
vs Elmhurst worksheet 78** (engine-on-Elmhurst-inputs 79 — the calculator
|
|
reproduces accredited Elmhurst within ~1 SAP on identical inputs).
|
|
|
|
## Scripts
|
|
|
|
| Script | Does |
|
|
|---|---|
|
|
| `start_viewer.sh` | Xvfb→x11vnc→noVNC desktop on **port 6080** so you can SEE/click the headed browser (`DISPLAY=:99`). `bash scripts/hyde/start_viewer.sh` |
|
|
| `elmhurst_session.py` | `login` / `open` / `status` — persistent login (MFA by hand once; session in `.elmhurst-session/`, gitignored) |
|
|
| `elmhurst_explore.py` | dump a page's DOM (field ids, options, autopostback) or `--all` sections — how specs are built |
|
|
| `elmhurst_fill.py` | `--page <name> [--commit] [--commit-mode saveclose\|next]` — fill/clear a page from its `PageSpec` |
|
|
| `elmhurst_download.py` | downloads Input Summary + SAP Worksheets → the cert's corpus dir. **Never clicks Submit** (code-enforced) |
|
|
|
|
Typical run (browser appears on the noVNC desktop at port 6080):
|
|
```bash
|
|
bash scripts/hyde/start_viewer.sh
|
|
DISPLAY=:99 python scripts/hyde/elmhurst_session.py login --url https://members.elmhurstenergy.co.uk/
|
|
DISPLAY=:99 python scripts/hyde/elmhurst_fill.py --page property_description --commit
|
|
DISPLAY=:99 python scripts/hyde/elmhurst_download.py
|
|
```
|
|
|
|
## Hard-won mechanics (ASP.NET WebForms)
|
|
|
|
- Enter via the Guid URL (`…online…/WebFormAddress.aspx?Guid=…`); section nav is
|
|
full-form `__doPostBack` to a Guid-less subdomain — use `expect_navigation`.
|
|
- **Almost every field AutoPostBacks** (whole-page reload): set one field, wait,
|
|
re-locate the next. A 600 ms settle after each interaction stops flake.
|
|
- **Never re-click the active tab** (double-postback misbehaves) — `goto_section`
|
|
is current-page-aware.
|
|
- **Navigating away saves**; **Save & Close commits to the DATABASE** (visible to
|
|
other sessions). Next > only saves the in-progress session.
|
|
- Fields **auto-disable** when an upstream choice removes them (Room-in-Roof once
|
|
no room-in-roof; cylinder once combi) — skip them, don't fight them.
|
|
- Button ids vary per page → match by id suffix (`[id$='_ButtonSaveClose']`).
|
|
- Setup (done in container): `python -m playwright install chromium` +
|
|
`sudo python -m playwright install-deps chromium`.
|
|
|
|
## What's automated vs manual
|
|
|
|
**Automated** (in `PAGES` + special actions): Property Description, Dimensions,
|
|
Flats (position/floor/corridor), Walls (main + clear alt-wall), Roofs, Floors,
|
|
Space Heating (**boiler** via the cascade dialog → BGW combi, flue, secondary),
|
|
Water Heating (code; cylinder auto-clears with the combi), New Tech/Conservatory
|
|
(none) — plus the consolidated **clears** from the markdown.
|
|
|
|
The **boiler cascade dialog** is driven by `select_boiler_combi()`: open the
|
|
Main-Heating-code modal, set the leaf dropdown to *BGW Post 98 Combi condens.*,
|
|
click OK. The generic combi (SAP 104) moves SAP by ~0 vs the cert's exact PCDB
|
|
model, so it's accurate enough.
|
|
|
|
**Manual (🔸):**
|
|
- **Exact PCDB boiler model** (e.g. 17929) — its search field is *disabled*
|
|
(`aspNetDisabled`) in this UI state; the cascade's generic combi is the
|
|
automatable substitute (negligible SAP impact).
|
|
- **Control code 2110** and the **MV unit** — same modal-dialog pattern; can be
|
|
automated the same way as the boiler when wanted (open dialog → select → OK).
|
|
All 🔸 fields are flagged in `elmhurst_inputs.md`.
|
|
|
|
## Adding a field/page
|
|
1. `elmhurst_explore.py <guid-url> <buttonAction<Section>_Link>` to dump ids/options.
|
|
2. Add a `Field(...)` (or new `PageSpec`) in `elmhurst_fill.py`.
|
|
3. `--page <name>` (dry-run, screenshots to `elmhurst_dom/fill_<name>.png`), verify, then `--commit`.
|