Skip to content

ADR-0027 — CORP-WEBSITE and IFRS-S2-CORP: permanent drop, external-only screen

Status: Accepted (2026-05-24) Supersedes: — Superseded by: — Related: ADR-0026 (baseline anchor → 100), docs/design/obs-1-scraper-sufficiency-review.md

Context

The source register has carried two "blocked/heavy" sources — CORP-WEBSITE and IFRS-S2-CORP — as deferred workstreams for corporate-site and corporate-PDF extraction:

  • CORP-WEBSITE — open-ended text extraction across arbitrary issuer websites/PDFs. The keystone source: every fully-dead positive rule (S2, S3, UN-G3, G1, UN-G2, G2, UN-G1, G4) depends solely on this source. Own-ADR, heavy engineering.
  • IFRS-S2-CORP — corporate-PDF extraction vs the IFRS S2 climate standard. Successor to the deprecated TCFD-SUPPORTERS source; feeds E4.1 as a disclosure-quality signal. Same corporate-PDF build shape as CORP-WEBSITE.

The obs-1 scraper-sufficiency review (docs/design/obs-1-scraper-sufficiency-review.md) assessed CORP-WEBSITE as the "next major workstream" after obs-1 Tier 1.

On 2026-05-24, the operator resolved to permanently drop both sources.

Decision

CORP-WEBSITE and IFRS-S2-CORP are permanently dropped. They will not be built.

The screen is now built entirely on external-assessment sources: regulators, NGOs, benchmarks, and GLEIF. Company self-reporting is out of scope by design.

Rationale

Three reasons, all deliberate and interrelated:

(1) Maintenance cost scales with universe

Corporate-site extraction is brittle by nature: each issuer's website layout, PDF structure, and disclosure vocabulary is different, and each changes over time. A universe of 79 issuers today implies 79 separate extraction patterns to maintain; at 200 issuers the maintenance burden roughly doubles. Scraper rot on corporate sites is a known class of operational failure — it produces silent data-quality problems rather than loud errors, which is the worst kind for a screening tool.

External-assessment sources (NZBA, PRB, SBTi, TNFD, TPI, IFRS-S2 mandate proxy, BankTrack, FOREST500, BHRRC, etc.) publish clean, structured lists. One scraper per source, maintained by the publisher. Maintenance cost does not scale with the universe. This is not a cost savings on build — it is a deliberate quality-and-maintainability architecture choice.

(2) Independence property — external-only screen

A screen that relies on company self-reporting to score companies has a structural conflict: an entity's disclosure quality directly affects its score. Entities with sophisticated disclosure teams and marketing budgets produce better-looking disclosures than entities with equivalent real-world performance. IFRS S2 alignment in a disclosure PDF is not the same as verified emissions reduction.

By restricting inputs to external assessments — independent organizations evaluating companies, not companies self-reporting — the screen has an independence property: companies cannot move their own scores by producing better disclosure. This is a deliberately stronger claim than "we primarily use external sources." It means the scoring function is not gameable by the subjects.

This does not mean the screen is complete or unbiased; external assessors have their own methodologies and coverage biases. But those biases are external to the scoring subject and largely observable, auditable, and replaceable.

(3) Trade-offs explicitly accepted

The permanent drop has known consequences, accepted by the operator:

  • S and G positive sides are permanently dark. Rules S2 (workforce practices), S3 (financial inclusion), UN-G3 (board diversity), G1 (board composition), UN-G2 (board independence), G2 (executive remuneration), UN-G1 (sustainability reporting quality), and G4 (transparency & disclosure quality) will never be fed by live data. These are documented in the rule catalogue as unfed.
  • E4 disclosure quality is proxy-only. The IFRS-S2 mandate proxy (E4.1, jurisdiction
  • year lookup) captures mandate presence, not disclosure quality. IFRS-S2-CORP was the path to actual disclosure-quality scoring; it is now out of scope.
  • Self-reported ESG data is out of scope. Commitments, targets, and disclosures that a company publishes about itself are not and will not be scored. The screen captures what independent assessors have determined, not what companies claim about themselves.

These are not bugs or gaps to close — they are design properties of the external-only screen. The rule catalogue will be updated to reflect unfed rules as such, and the methodology page will make this explicit.

Consequence for ADR-0026 (baseline anchor)

The permanent dark state of S and G positives is the structural reason the 100-baseline is coherent. A neutral-50 model assumes a two-sided scale: positives can push up, negatives can push down, no evidence is neutral. With the positive side of S and G permanently dark, neutral-50 caps a clean institution at 50 regardless of its actual governance behaviour. A deduction-led scale starting at 100 is the honest alternative: the screen can only assess what external assessors have found, and absence of adverse external findings reads as clean.

See ADR-0026 §Context for the full argument.

What this does NOT change

  • IFRS-S2-MANDATE (the jurisdiction-mandate proxy, already built): unaffected. This is a small static lookup table, not a corporate-PDF scraper. It lights E4.1 via mandate presence and is structurally different from IFRS-S2-CORP.
  • Any existing live sources: SBTi, NZBA, PRB, TPI, TNFD, BankTrack, BHRRC, FOREST500, InfluenceMap, CA100PLUS, CHRB, etc. — unaffected.
  • Future reconsideration path: if the external-only principle is ever revisited — e.g. a future operator decision to accept company self-reporting for specific verified sub-rules — that decision should produce a new ADR superseding this one. Do not reverse quietly.

References

  • ADR-0026 — Baseline anchor → 100 (baseline consequence of this decision)
  • docs/design/obs-1-scraper-sufficiency-review.md §4, §6 — source assessments and verdict
  • docs/design/obs-1-scraper-sufficiency-review.md §9 decision 4 — CORP-WEBSITE reclassification (initial "next major workstream" framing, superseded by this ADR)