mirror of
https://github.com/Hestia-Homes/Model.git
synced 2026-07-12 13:29:04 +00:00
44 lines
2.7 KiB
Markdown
44 lines
2.7 KiB
Markdown
# Resume prompt — finish the Hyde portfolio-796 property_overrides run (after Khalim review)
|
|
|
|
Paste the block below to continue. It tells the assistant to review the unknown-override
|
|
decisions with me, verify them, confirm before writing, then run the remaining steps.
|
|
|
|
---
|
|
|
|
We paused the Hyde property-overrides bulk load to review the UNKNOWN classifications with
|
|
Khalim. Pick it back up.
|
|
|
|
**Context (already done):**
|
|
- Target is **portfolio 796** in DevAssessmentModelDB (NOT 795 — 795 is empty).
|
|
- Script: `scripts/hyde/build_property_overrides.py`. Pass 1 (`classify`) is DONE — the
|
|
`landlord_*_overrides` ledger is populated; re-running classify is free (cache hits).
|
|
- The 19 unresolved descriptions are documented in `scripts/hyde/unknowns_review.md`, with
|
|
proposed values already written to `overrides_edits.csv` (gitignored).
|
|
- Env (DB creds + `OPENAI_API_KEY`) is in `/workspaces/home/github/Model/.env`; load it with
|
|
python-dotenv and set `POSTGRES_DRIVER=psycopg2`. Writes are idempotent upserts (unique on
|
|
`property_id, override_component, building_part`) — safe to re-run, never duplicates.
|
|
|
|
**Do this, in order:**
|
|
1. **Ask me what Khalim decided** for the unknowns. The one real judgement call is the
|
|
flat-roof reading: `Flat: As Built` (1,172 rows) + `Flat: Unknown` (194) → which of
|
|
`Flat, no insulation (assumed)` / `Flat, insulated (assumed)` / `Flat, limited insulation
|
|
(assumed)`. The `construction_age_band` bands (29,829 rows) are deterministic (band = first
|
|
letter) — keep as-is unless I say otherwise. Confirm the other roof/wall proposals too.
|
|
2. **Update `overrides_edits.csv`** (`corrected_value` column) to match Khalim's decisions.
|
|
3. Run `validate --edits overrides_edits.csv` and fix anything it rejects.
|
|
4. **Show me the final edits + the planned write counts, and WAIT for my explicit go-ahead
|
|
before any `--apply`.** Do not write to the DB before I confirm.
|
|
5. On my go-ahead:
|
|
- `apply-edits --edits overrides_edits.csv --portfolio-id 796 --apply` (user corrections → ledger)
|
|
- `write --excel scripts/hyde/hyde_property_overrides.xlsx --portfolio-id 796` (DRY RUN —
|
|
report unmatched org_refs + unresolved across all 31,773 first)
|
|
- then the same `write ... --apply`
|
|
6. `verify --portfolio-id 796 --org-ref <a few org_refs>` to confirm property_overrides +
|
|
overlays landed.
|
|
7. Remind me about the deferred **age-classifier prompt-hint fix** for the production lambda
|
|
(the live frontend will hit the same `"D: 1950-1966"` → UNKNOWN until that lands).
|
|
|
|
Every DB command loads env from `/workspaces/home/github/Model/.env`. Read-only checks
|
|
(`verify`, dry-run `write`) are fine to run unprompted; anything `--apply` needs my confirm.
|
|
|
|
---
|