ALLSHIFT/docs

SHIFT Documentation

Every current document in the project, in markdown, in one place.

Documents that arrived as Word, PDF, Excel, or MATLAB Live Script have a markdown report here — readable in the browser, diffable in git, searchable with grep. The originals are kept alongside (root-level ones in _originals/, the rest in place under Shift Matlab Drive/); each report links back to its source and records the file's MD5.

Outdated material lives in ../superseded/, with the reason for each retirement.

---

Start here

If you want to…	Read
Understand what SHIFT is building	Project Explanation
Know what the simulator must accept and return	Simulator I/O Interface
Know what the RL agent sees, does, and is scored on	State · Action · Reward
Find which Simulink model is the live one	Simulink Model Inventory
Know what data exists and where it came from	Dataset Inventory
See what is unresolved	Open questions across the project

---

01 · Project

Document	What it covers
Project Explanation	Vision, mission, values, the five departments, benchmarking against Tibo/Zympler and the H2B2–PowiDian Rijnstate pilot, business case and routes to market
Work Packages	WP1–WP7: literature review, forecasting, simulation, RL control, evaluation, extensions, reporting
AI Efficiency Improvements	Where the losses are — ~30–40% electrolyser, ~40% fuel cell — and the scheduling changes that attack them
Reference Links	Four external sources: two IEEE/ScienceDirect papers, a review, and the H2B2–PowiDian announcement

Note

Work Packages WP6 describes "a basic crop growth/CO₂ enrichment module", "greenhouse-to-grid interactions", and "scalability tests for multiple greenhouses". SHIFT builds a hospital microgrid. WP6 is a leftover from an earlier greenhouse concept and does not describe current scope. It is kept rather than retired because WP1–WP5 and WP7 are current and still the project's work breakdown — but WP6 needs rewriting, not reading.

02 · Specifications

The contract layer. These define what gets built.

Document	Source	Status
Simulator I/O Interface	PDF, 10 pp	Draft v0.1 — 8 open questions
RL State Variables	Word	32 variables, no units
RL Action Variables	Word	16 actions; 9 have no simulator counterpart
RL Reward Function	Word	26 terms, no weights
Forecasting Requirements	PDF, 10 pp	Complete; PV, demand, and price forecasters
Rule-Based Controller	markdown	The baseline policy RL must beat

Warning

The specification set is not internally consistent. Two naming conventions are in use — the Simulator I/O sheet says SoC(t), P_PV, u_ele; the RL documents say Battery_SOC, PV_Generation, Electrolyzer_PowerSetpoint — and neither references the other. The action spaces disagree in size (5 vs 16), and five reward terms need quantities the simulator does not return. Each report documents its own mismatches; resolving them is a decision, not an edit.

03 · Energy Management

Case-study data and equipment for the two reference sites: BIDMC (Beth Israel Deaconess Medical Center, Boston) and UCSD (UC San Diego).

Document	What it covers
BIDMC/UCSD Energy Flow and Balances	Four-level dispatch priority, three-tier load model, energy balance equations, four KPI formulae
UCSD Equipment Manual	DFC3000 fuel cell, BYD battery, 30 MW CHP, planned PEM electrolyser — each parameter tagged State / Action / Reward
BIDMC Datasets Manual	Full column reference for the three MIMIC-IV demo clinical files and the synthetic building-energy CSV
BIDMC ComStock Load Profile	The hospital load proxy: 8,026 MWh/yr electricity, 15-min resolution, full end-use breakdown
BIDMC Digital Twin Readiness Checklist	8-section readiness gate. Sections 1–3 scored; 4–8 blank
PV Hourly Dataset Manual	The recommended PV dataset
Satellite PV Dataset Manual	The 167.9 MB NSRDB meteorological dataset — detailed, but not recommended for PV rates

04 · Simulations

Document	What it covers
Simulink Model Inventory	Every .slx, with real lineage from Simulink's internal revision metadata. Identifies the current model
Rule-Based Controller Implementation	controller-claude.m — the dispatch cascade in MATLAB, and why it does not yet run
PEM Electrolyser and Fuel Cell Models	The MathWorks Simscape packages and their full parameter sets
PV + Battery Simulink Model	The Excel→Simulink vertical slice, and its known time-dependency bug
MATLAB Live Scripts	All five .mlx files transcribed
Generation Profiles Workbook	Both sheets of the profiles spreadsheet, plus the PowerWorld results analysis
PowerWorld Microgrid Model	The campus-scale network study
Shift Input Data	The per-component variable list the simulator contract grew from
Existing EST Model	The vendored MIT-licensed TU/e model, and SHIFT's gravity-storage experiments on top

05 · Business & Economics

Document	What it covers
Cost Analysis	A 48-row parameter register — value, source, why it matters, and a Confirmed / Estimated / Derived / Missing status per row. The most rigorous single artefact in the project
Social Cost-Benefit Analysis	SCBA methodology: the questions a robust analysis must answer, discounting, NPV in Excel

The SCBA document is methodology, not a SHIFT analysis — it is course material on how to run an SCBA, worked through a generic wind-farm example. The SHIFT-specific analysis has not been written yet. The parameter register in Cost Analysis is what would feed it.

06 · Data

Document	What it covers
Dataset Inventory	Every data file: size, rows, provenance, licence obligations, and the redundancies

07 · Team & Operations

Document	What it covers
Interview Questions	The recruitment question template — general, technical, business, PR
Team Design	Polo/sweatshirt and business card design, logo concept

---

Open questions across the project

Collected from the individual reports. Each links to where it is documented in full.

Blocking

Question	Where
Timestep: hourly or sub-hourly? Affects RL compute cost, battery fast-response fidelity, and every energy/power conversion	Simulator I/O §0
Reward weights: 26 terms, no relative magnitudes beyond "critical load dominates"	RL Reward
Naming convention: two incompatible variable vocabularies across the spec set	RL State
Critical / non-critical load split: can the hospital profile actually be split, or is a proxy fraction needed?	Simulator I/O §5
PEM models are still at vendor defaults — the "10× bigger" sizing TODO was never done, and the fuel cell still runs an automotive drive cycle	PEM Models
The dispatch controller does not compile — undefined variables, outputs unassigned on several paths	Controller Implementation
Readiness checklist §4 (Safety) is unscored — the document's own rule makes any FAIL here a deployment blocker	Readiness Checklist

Data

Question	Where
No measured hospital load — every figure is simulated or synthetic	Dataset Inventory
No price or carbon-intensity series — both are required simulator inputs	Dataset Inventory
PV array configuration is inconsistent — 4 kW / 100 kWp / 460 kW across three documents	Dataset Inventory
Boston or Netherlands? — all datasets are Boston; the price forecaster targets the Dutch market	Dataset Inventory
Licence and citation files missing for MIMIC-IV demo (ODbL), ComStock, NSRDB, PVWatts	Dataset Inventory

Modelling

Question	Where
Tank model — isothermal + ideal gas, or full pressure/temperature dynamics?	Simulator I/O §5
Battery degradation — modelled or ignored?	Simulator I/O §5
Should simultaneous electrolyser + fuel cell operation be structurally forbidden, or merely penalised?	Simulator I/O §5
Two-tier or three-tier load model? The specs and the energy-flow doc disagree	Energy Flow
Is the clinical/patient layer in scope at all?	BIDMC Datasets
Was chal's _CLEAR_CONNECTIONS cleanup (rev 2.6) ever merged into rev 2.7?	Simulink Inventory

---

Conventions used in these reports

Every report generated from a non-markdown source opens with a provenance block:

Field	Meaning
Source	Repo-relative path to the original file
Format	File type and size
MD5	Checksum at the time of writing — lets you tell whether the original has changed since
Status	Current, superseded, or draft
Report generated	The date this markdown was produced

Where machine extraction was lossy — PDF tables that interleave columns, Word equations, images — the report says so explicitly and points back to the original. Reconstructed tables are labelled as reconstructed. Author asides and first-person notes from the sources are preserved as quotes rather than smoothed away, because they usually mark exactly the parts that are uncertain.

Observations that go beyond the source — inconsistencies between documents, unit errors, scale mismatches — are marked with callouts and attributed to this audit, not to the original author.