Scientific validation & scope
Transparent about what the evidence supports.
A dedicated red-team pass checked source provenance, circular validation, stale assumptions, reproducibility and claim wording before the conference release.
Pass with disclosed limitations
Evidence classes
Not every successful check is the same kind of validation.
| Class | Question answered | Example |
|---|---|---|
| S1 · Provenance | Can the number be traced to a named source, date and boundary? | CEA ATC; official annual electricity consumption. |
| S2 · Arithmetic / invariant | Does the transformation conserve what it should? | Daily-energy conservation; accounting identities. |
| S3 · Code verification | Does the software implement the intended equations and guards? | Unit/integration tests; state-equation checks. |
| S4 · Independent implementation | Do different implementations solve the same mathematical problem consistently? | PyPSA ↔ SciPy ↔ OSeMOSYS. |
| S5 · Held-out statistical | Does the method perform on observations withheld from fitting? | January–March intraday extrema reconstruction. |
| S6 · External physical/planning | Does independent engineering evidence support the physical inference? | CEA independently identifying Shoranur as constrained. |
| S7 · Planning-grade | Is the model calibrated enough for operational/investment decisions? | Not claimed by this study. |
What the QA changed
The audit corrected interpretation, not just formatting.
Shoranur transformer
The public model used a genuine 2023 100 MVA transformer element. Current 2026 CEA planning reports 200+100 MVA existing transformation. The hotspot geography remains independently corroborated; the old overload magnitude is not current whole-station loading.
Idukki v1.3 storage replay
Historical storage closure is algebraic because the forcing is reconstructed from storage change and generation. It verifies implementation, not hydrology.
ERA5 demand reconstruction
The held-out test evaluates intraday shape conditional on known daily energy. It is not an hourly demand forecast.
21 network lines
“Robust” means the same lines exceed screening rating across eight tested residual-import injection allocations—not across every load, dispatch, impedance or contingency uncertainty.
IEX ₹5.24/kWh
This is the FY2023–24 average MCP (~₹5.23759/kWh), not the volume-weighted MCP (~₹5.17449/kWh).
5,904 vs 5,797 MW
The first is the CEA generation-RA planning reference used by the load proxy; the second is the separately reported FY2024–25 recorded peak in the newer transmission-RA plan.
Supported conclusions
What can be stated confidently within the study boundary.
Demand growth is real.
Broad electricity consumption rose from 22.54 to 29.31 TWh between FY2020–21 and FY2024–25.
Timing matters.
Renewable spill and shortage can coexist within the same screening year; annual energy alone cannot establish adequacy.
Hydro flexibility matters.
The same hydro energy can have very different adequacy value depending on timing and transfer conditions.
4,455 MW ATC is a valid dated reference.
CEA provides this current Kerala transfer-capability figure; the model does not treat it as a guaranteed future capability.
Shoranur is a defensible hotspot location.
The public screen and current CEA planning independently point to the same geography.
Fail-closed gates are part of the result.
Unverified catchment geometry, incomplete statutory siting layers and unsupported future ATC are not promoted to stronger claims.
Scope boundary
What this study does not establish.
Not demonstrated here
- Exact operational cause of a specific 2026 load-shedding event.
- Measured hourly demand telemetry or day-ahead hourly forecast skill.
- Transient, frequency, voltage or small-signal “grid stability”.
- AC/N-1 security or operator-validated congestion.
- Hydrologically validated Idukki catchment/cascade operation.
- Statutory buildable renewable capacity.
- Probabilistic LOLP/ELCC.
- A total-system least-cost 2040 portfolio.
What the study does provide
- Source-audited historical electricity evidence.
- A bounded hourly demand reconstruction.
- Chronological adequacy and transfer sensitivities.
- Hydro timing and stateful pilot experiments.
- Public-data transmission stress screening.
- Independent implementation checks.
- Candidate response mechanisms and evidence gaps.
Reproducibility
Most released model runs are reproducible; one fitting stage remains incomplete.
Reproducible layers
Historical transformations, PyPSA screens, hydro timing experiments, cross-solver comparisons, network screens and synthetic flexibility pilots retain code/config/evidence sufficient to reproduce released calculations subject to source access.
ERA5-sensitive v2 fit
The released 8,760-hour payload, hashes, metadata and downstream use are reproducible. The original fitting script and raw ERA5 archive were not committed, so the coefficient fit cannot currently be regenerated from a fresh public clone.