Skip to content

ADR-0014 — CHRB source instrument clarification

Field Value
Status Accepted
Date 2026-05-21
Supersedes —
Superseded by —
Related ADR-0013 (seed compilation scope); cycle 11 commit c8d90d4 (CHRB wiring); cycle 11.5 chat session

Context

Cycle 11.5 seed compilation surfaced that WBA's instrument landscape had shifted since the cycle 11 scaffold note was written. The Social Benchmark (2024) is a new WBA instrument covering 2,000 SDG2000 companies at shallower depth, scored 0–20. CHRB continues as a distinct, sector-deep instrument covering ~100–130 companies in high-risk sectors, scored 0–100 natively.

The risk considered: were Social Benchmark and CHRB the same instrument under a new name? If so, the scaffold (threshold 40/100, sectors apparel/extractives/food/ICT/auto) would have been wrong — the ×5 normalisation needed to compare a 0–20 instrument against a 40/100 threshold would have been a silent methodology error.

Additionally, cycle 11.5 noted that CHRB publishes in a biennial two-sector-split pattern (2022: food & agri, ICT, automotive; 2023: apparel, extractives) rather than a single annual release. Treating 2022 + 2023 as the "5th iteration combined" requires a methodology statement.

Decision

CHRB and Social Benchmark are treated as separate sources. CHRB is the source feeding NF-S3.3 and NF-S4.3. Social Benchmark is not currently a wired source.

CHRB 5th iteration is compiled as 2022 + 2023 combined. The two years are different sectors of the same iteration per WBA's own framing. Refresh next when 6th iteration publishes (expected 2026).

Threshold remains chrb_score_rounded < 40 / 100. No ×5 normalisation needed (the Social Benchmark workaround that was investigated during the chat session and discarded once CHRB was confirmed to still exist as a distinct instrument).

Rationale

(a) CHRB and Social Benchmark are complementary, not substitute instruments. WBA's own documentation treats CHRB as "deeper, sector-specific analysis of around 100 high-risk companies" and Social Benchmark as covering "fundamental steps" across 2,000 companies in the SDG2000. These are different depths of analysis, different company scopes, and different scoring scales (0–100 vs 0–20). A substitution would have required score normalisation and would have lost the sector-specific depth that makes CHRB useful for NF-S3.3 (coverage) and NF-S4.3 (poor performance).

(b) 2022 and 2023 CHRB use the same revised methodology. Both editions use the revised methodology published in 2021 and the same 0–100 scoring scale. Combining them is methodologically sound — the two years are complementary sector coverage, not competing assessments of the same companies.

(c) Threshold calibration deferred, not changed. The 40/100 threshold was set in cycle 11 as a scaffold and is retained unchanged. With 8 unique institutions matched (7 below threshold, 1 above), there is insufficient distribution to calibrate against. Revisit at cycle 16 (PRB seed completeness + RAG calibration) or when the universe expands significantly in cycle 15.

Consequences

  • C11.5 seed compiled from 2022 + 2023 combined: 9 entries, 8 unique institutions.
  • Social Benchmark integration not pursued at this time.
  • Refresh script (scripts/refresh-chrb-seed.js) must read both 2022 and 2023 datasets until 6th iteration replaces them.
  • C16 (next CHRB refresh decision): if 6th iteration is methodology-continuous with the 5th, replace the seed. If not, write a new ADR. WBA has signalled a "unified 0–100 scoring scale" for 2026 benchmarks; CHRB already uses 0–100 so no break expected, but confirm methodology continuity before replacing.
  • VW LEI mismatch (seed LEI 529900NNUPACCH7MJ151 ≠ config.lei 529900NNUPAGGOMPXZ31) addressed by migration 035 (chrb_match_name = 'Volkswagen' on INST-0022).

References

  • ADR-0013 (seed compilation scope)
  • Cycle 11 commit c8d90d4 (CHRB + FOREST500 build)
  • Cycle 11.5 chat session compilation
  • WBA Benchmark Hub: worldbenchmarkingalliance.org/benchmarks/2026-benchmark-hub
  • 2026 scoring approach: worldbenchmarkingalliance.org/research/scoring-approach-2026-benchmarks