Repository navigation
Research roadmap: integrate Jev appropriately across energy intelligence - #23
Conversation
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 24ffe0c9c7
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
|
|
||
| Build Luminus as a **point-in-time energy intelligence and decision-support system**. Use physical/statistical forecasting for numbers, deterministic code for mathematics and market rules, an LLM for extracting complex documents and explaining research, and Jev only for bounded semantic judgments that demonstrate incremental value in controlled tests. **Do not make Jev a compulsory dependency or market-forecasting oracle.** No autonomous real-money trading, unsupervised battery dispatch or grid-safety control in the proposed releases. | ||
|
|
||
| The repository already exposes weather (`get_weather_forecast`, `get_era5_weather`), demand and renewable forecasts, prices, balancing/imbalance, REMIT/outages, gas and carbon, flows, GIS and BESS screens. It also has a thin Python SDK and MCP profiles. Existing [R&D roadmap](rnd-roadmap.md) documents lack of point-in-time storage, BESS hindsight/perfect-foresight limitations and the `src/tools/prices.ts` timestamp/currency/resolution issue. Correct these **before** fitting models: `prices.ts` labels a source currency as EUR, treats each position as an hour, sorts different periods by repeated positions and can corrupt downstream spread analysis. Adopt the existing `extractSeriesPoints` interval expansion and add coverage/DST/multi-period fixtures. None of this is solved by Jev. |
There was a problem hiding this comment.
When an A44 price period has a missing point, the recommended extractSeriesPoints helper silently fills that slot with the previous value: src/lib/entsoe-timeseries.ts applies forward-fill to every period without checking curveType (lines 117–120). Reusing it directly for auction prices can therefore fabricate observations and prevent the proposed coverage checks from detecting gaps, contaminating spread and backtest results. Restrict this behavior to A03 curves or use a price-specific extraction path that preserves missing intervals.
Useful? React with 👍 / 👎.
Adds a detailed, research-backed companion to the existing R&D roadmap covering Jev's documented primitives and limitations, data correctness/point-in-time contracts, weather, demand, three volume definitions, price and trading research, BESS/GIS/REMIT/document intelligence, provider-optional architecture, calibration, ablations, source/market-data feasibility, release gates and 30-day backlog.
This is documentation only: no API keys, Jev code, model results or live trading are claimed. The model is optional and shadow-only until a matched-baseline evaluation demonstrates incremental value. Main roadmap and default branch remain untouched pending review.
Primary sources and implementation caveats are linked in the document.