UK Battery Storage – Sponsor Track Record as Deliverability Predictor
What 2,931 labelled outcomes from the NESO TEC Register (2012–2026) reveal is that a sponsor's track record of prior grid connections is the single strongest predictor of whether a battery storage project actually reaches completion. (Source: NESO TEC Register, 2012–2026, n=2,931 labelled outcomes.) Sponsors with two or more prior connections complete at roughly 73%, against a population baseline of approximately 23%.
6.0/10
Hostile score
Adversarial validation panel
0.889
Temporal AUC
Holdout 2021+ only
73%
Completion (experienced)
vs 23% baseline
1. Sponsor track record outperforms technology class
Technology class (BESS, solar, wind, hybrid) on its own tells you very little about whether a TEC project will actually reach grid connection. When we run the logistic model on technology class alone, discriminatory power is low. But adding the binary sponsor experience flag (prior_connections ≥ 2) moves the needle more than anything else we tested. (Source: Manathric model, logistic regression, prior_connections feature lift.) Sponsors with two or more prior connections complete at approximately 73%, against a population baseline of around 23%. (Source: NESO TEC Register, 2012–2026, n=2,931.) The experienced cohort is small – 144 sponsors across the register – but cleanly identified.
This is consistent with a capability explanation: experienced sponsors have navigated grid connection processes before, hold established relationships with DNOs and transmission operators, and are more likely to resource the legal, technical, and financial requirements that frequently cause attrition in less experienced project teams.
2. Temporal validation preserves signal integrity
The model was trained exclusively on pre-2021 data. The holdout set covers 2021 onward, a period of material policy and market change in UK BESS that includes the Contracts for Difference exclusion reform and network charging reviews. (Source: Manathric model, training cut pre-2021, holdout 2021–2026.) The temporal AUC of 0.889 on this holdout confirms that the sponsor experience signal generalises across regulatory regimes, not only within the training window. (Source: Manathric model, expected_mw v3, holdout 2021+.)
The temporal split was applied precisely to prevent future data leakage, a structural flaw in V1 of this signal, which used a random train/test split. In V1, sponsor entities from the same firm appeared in both training and holdout, inflating out-of-sample metrics and masking the true predictive boundary. V2 corrects this by enforcing strict time separation. Any prior model or screen built on V1 data should be treated as unreliable.
3. Implications for project screening
The deliverability gap between experienced and inexperienced sponsors is roughly 50 percentage points, which is large enough to drive material differences in expected portfolio MW if screening is applied at origination. (Source: NESO TEC Register, 2012–2026; experienced cohort ~73%, population baseline ~23%.) A fund deploying into UK BESS pipeline that does not condition on sponsor track record is implicitly accepting a population completion rate of approximately 23%, versus a screened rate of around 73% for the experienced cohort.
This signal is UK-specific and BESS-focused. It should not be applied directly to other geographies or technologies without re-estimation. The small size of the experienced sponsor group (n=144; Source: NESO TEC Register entity resolution) means that new entrants with only one prior connection fall into the inexperienced bucket, even if that single connection was a large or complex project. Refinement of the threshold is planned for H002 V3.
Hostile validation
The hostile validation panel raised concerns on: sample size of the experienced cohort (n=144), UK-only scope, and the binary threshold for prior connections. (Source: Manathric adversarial validation panel.) The panel accepted temporal split methodology and entity resolution corrections as substantive improvements over V1 (score 4.2/10). RN-004 supersedes RN-001.
RN-001 (V1, hostile score 4.2/10) contained an incorrect sponsor entity merge that inflated completion rates for the experienced cohort. All conclusions from RN-001 should be treated as unreliable. The corrected analysis is this note. (Source: Manathric adversarial validation, V1 score 4.2/10, V2 score 6.0/10.)
Investment implications
Investors
Sponsor track record is a first-order screen. Projects led by sponsors with fewer than two prior grid connections should carry a deliverability haircut in underwriting models, regardless of technology class or region. (Source: NESO TEC Register, 2012–2026, n=2,931.)
Developers
Experienced developer status (≥2 prior connections) confers a measurable completion premium. Platform or partnership strategies that pool track record across subsidiaries should be disclosed and tested separately.
Research teams
Signal H002 V2 supersedes V1. Any model trained on the uncorrected sponsor merge should be retrained. The corrected dataset and label file are available on request.