Climate Disclosure Engine — done-for-you California SB 253 GHG reports and the assurance-ready evidence file behind them

Gate 1 · Low trust burden — 4/5

Big companies already outsource ESG reporting to consultancies; buying the outcome from an external expert is normal. Slight friction: emissions data is sensitive and the report carries the company's name, so trust-building is required.

Gate 2 · Low task-level judgment — 4/5

Most steps — data extraction, unit normalization, emission-factor matching, calculation, report assembly — are deterministic or rules-bound. Judgment concentrates at organizational-boundary decisions and data-gap estimation.

Gate 3 · High intelligence threshold — 4/5

Requires synthesizing messy utility/fuel/fleet data against GHG Protocol, CARB rules, and emission-factor libraries, plus defensible treatment of estimates and exclusions. Not simple labor arbitrage; quality is the product.

Gate 4 · Regulation as moat — 5/5

Statutory mandate, civil penalties, named entity list, and an assurance ramp that raises the rigor bar yearly. Regulation creates the demand and deepens it over time.

Gate 5 · No physical labor — 5/5

Entirely digital: documents, spreadsheets, ERP/utility exports, PDFs. Remote delivery; no logistics.

Gate 6 · Sam Altman test — 4/5

Better models improve extraction accuracy, factor-matching, and narrative drafting — lowering COGS and raising throughput. Watch: well-funded software incumbents could fold equivalent automation into a managed service.

Verified regulatory change

• SB 253's first-ever Scope 1&2 filing is due Nov 10, 2026 — a hard, near-term deadline with $500k/yr penalties.
• CARB finalized initial regulations in early 2026 and named ~4,160 likely covered entities.
• The assurance bar escalates on a fixed schedule (2027 limited → 2030 reasonable), so rigor only increases.

Inferred capability change

• Frontier multimodal models can now reliably parse utility bills, fuel invoices, and fleet logs into structured activity data.
• LLMs can match activity data to emission factors and draft GHG-Protocol-aligned documentation with expert review, collapsing labor per inventory.

Unverified / hypotheses

• Other states replicate California within 2–3 years (optionality, not relied upon).
• Value-chain pressure forces sub-threshold suppliers to report voluntarily, expanding the buyer pool.

Deliverable

A submitted, CARB-conformant Scope 1 & 2 GHG emissions report for the covered entity, plus a structured assurance-ready evidence binder: organizational-boundary memo, data-source register, emission-factor citations, calculation workpapers, estimation/exclusion log, and an officer-attestation package.

Acceptance criteria

• Report accepted by CARB's reporting mechanism without deficiency notice.
• Every reported figure traceable to a source document and an emission factor.
• Binder structured to the standards limited assurance will test from 2027 (ISO 14064-3 / ISSA 5000 mapping).

Customer promise

"You hand us your raw data; we hand you a filed, defensible disclosure and a binder that survives audit. One contact, fixed fee, on the deadline."

Exclusions & policy

• Excludes legal opinions on coverage (we coordinate with the client's counsel) and third-party assurance opinions (separate accredited provider).
• Rework: any CARB deficiency we caused is remediated at no charge before deadline; missed-filing-due-to-us triggers fee credit.
• Success metric: 100% on-time, deficiency-free filings; binder passes assurance with zero material findings (2027+).

1 · Intake

Secure portal + email-in. Utility bills (PDF/CSV), natural-gas & fuel invoices, fleet/telematics exports, refrigerant logs, facility lists, ERP/AP feeds, prior sustainability reports, org charts for boundary scoping.

2 · Normalization

OCR + LLM extraction into a canonical activity-data schema (site, meter, period, quantity, unit, fuel type). Unit conversion, de-duplication, period alignment, versioning, and gap flagging.

3 · Retrieval / knowledge

Indexed GHG Protocol Corporate Standard, CARB regulation text & FAQs, EPA/eGRID & IEA emission-factor libraries, GWP tables, prior-client boundary precedents, and assurance-standard checklists.

4 · AI workbench

Models classify each activity record to a scope/category, match it to the correct emission factor, compute CO₂e, draft the boundary memo and methodology narrative, and assemble the workpapers and exclusion log.

5 · Deterministic rules

Emission-factor application, unit math, GWP multiplication, eGRID subregion mapping, and report-schema validation are code, not probabilistic — every number is reproducible.

6 · Human chokepoint

A GHG-inventory specialist reviews boundary decisions, data-gap estimates, factor selection on edge cases, and sign-off readiness; a senior reviewer approves the officer-attestation package.

7 · QA layer

Automated completeness checks (every facility has data or a logged exclusion), variance vs. prior year, factor-version checks, and a red-team pass before delivery.

8 · Delivery

Filing to CARB's mechanism on the client's behalf (with authorization), plus the evidence binder and a one-page executive summary for the attesting officer.

9 · Learning loop

CARB deficiency notes, assurance findings (2027+), and reviewer edits feed back into extraction prompts, factor-matching rules, and gold-standard templates.

10 · Model portability

Provider-agnostic abstraction over extraction/drafting models; deterministic calc layer is model-independent, so we adopt better models without re-architecting.

Standardization

• One canonical activity-data schema across all clients.
• Boundary-scoping SOP with a decision tree (operational vs. financial control).
• Required-evidence checklist per facility type generated at intake.

Variance elimination

• Automated completeness gate: no facility ships without data or a logged, reasoned exclusion.
• Confidence scoring on every extracted record; low-confidence routes to an exception queue.
• Year-over-year variance flags catch anomalies before filing.

Control & auditability

• Full version control and immutable audit trail on every figure and edit.
• Reviewer-assignment logic by sector/complexity.
• Gold-standard binders as templates and training data.

Continuous improvement

• Root-cause analysis on every CARB deficiency or assurance finding.
• Postmortem loop converts each recurring manual fix into a rule, template, or model-prompt change.

Primary model

Fixed per-entity annual report fee (tiered by facility count / data complexity), plus an assurance-readiness retainer that grows as the 2027/2030 assurance bar rises. Scope 3 (2027) priced as an add-on module.

Why not hourly

Hourly punishes our own automation and caps margin. A fixed outcome fee lets every efficiency gain accrue to us while the customer gets price certainty — the core AI-native flywheel.

Pricing legality

No contingency/percentage issues here (this isn't recovery). The only boundary: we don't sell legal coverage opinions or assurance opinions — both are separate, so no unauthorized-practice or independence conflicts. Clean fixed-fee service.

Compliance-safe structure

We prepare; the client's officer attests; an independent accredited firm assures (2027+). Keeping preparation and assurance separate preserves assurer independence and is itself a selling point.

Why this GTM

The buyer is enumerated (CARB's ~4,160-entity list), penalty-motivated, and deadline-bound. This rewards precise outbound and warm intros from the lawyers, auditors, and consultants already advising these companies on the rule.

First 50 prospects

$1B–$10B firms on/adjacent to CARB's covered list with no visible sustainability hire (LinkedIn signal) and a California nexus — concentrated in industrials, retail, food & bev, and real estate.

Trigger events

CARB covered-entity notice; audit-committee agenda items; investor/customer ESG-data requests; approaching Nov 10 deadline; competitors' first disclosures appearing.

Channel strategy

Referral partnerships with ESG/securities counsel, mid-tier audit firms (who can't staff prep but will assure later), and benefits/risk consultancies — they hand off prep, we hand assurance back to them.

Outreach wedge

"You're likely on CARB's list with a Nov 10 filing and no team for it. We file it for you, fixed fee, and the evidence binder is built to survive the assurance starting in 2027."

Conversion path & metrics

Free 20-minute coverage-and-readiness triage → fixed-fee proposal → data intake → filing. Track: triage-booked rate, triage→proposal, proposal→close, time-to-file, and net revenue retention (Scope 3 + assurance expansion).

Pilot cap

Hard cap of 5–8 covered entities for the Nov 10, 2026 cycle. Refuse more even if demand exceeds it — the goal is a hardened engine, not maximum logos.

Pilot ICP

$1B–$5B firms with 5–40 facilities, mostly Scope 1&2-relevant operations, and an internal champion who will give us real data fast.

Success criteria

• 100% on-time, deficiency-free CARB filings.
• Measured automation rate ≥ target by client #5.
• ≥3 references willing to vouch + 1 channel partner activated.

Demand-trap discipline

• Instrument every manual workaround; log minutes by step.
• Refuse bespoke client asks that don't generalize.
• Convert each repeated human fix into a rule/template/prompt before scaling.
• Define the process-hardening gate that must close before client #9.

What our AI/operators do

Extract activity data, classify to scope, apply emission factors, compute CO₂e, draft methodology and workpapers, validate the report schema, and assemble the evidence binder.

What stays with humans

Operators review boundary and estimation decisions; a senior reviewer approves sign-off readiness. The client's officer attests to the report; the client's counsel opines on coverage; an independent accredited firm provides assurance (2027+).

Prohibited claims

We do not provide legal opinions, do not issue assurance, and do not guarantee a specific regulatory interpretation. We prepare a defensible inventory; we don't certify it ourselves.

Controls

Data-processing agreement and confidentiality on sensitive operational data; immutable audit logs; clear authorization before any filing; independence firewall between our prep work and any assurer.

Stack

• Intake: secure upload portal + email ingestion; object storage with per-client isolation.
• Extraction: multimodal LLM + OCR into a typed activity-data schema (Pydantic-validated).
• Knowledge: vector index over GHG Protocol, CARB text, emission-factor libraries (eGRID/IEA), GWP tables; retrieval returns record IDs for citation.
• Calc: deterministic Python service (no model in the math path) with unit tests + factor-version pinning.
• Workbench: reviewer UI with confidence scores, exception queue, diff view, audit trail.
• Delivery: report generator + CARB filing handoff; binder export (PDF/structured).

Build sequence

• Wk 1–3: schema + extraction + factor library + deterministic calc core.
• Wk 4–6: reviewer workbench, completeness/QA gates, binder template.
• Wk 7–9: run pilot client #1 end-to-end; instrument every manual step.
• Wk 10–12: harden the top recurring manual fixes into rules/templates; onboard clients #2–5.
• Throughout: regression tests on factor updates; deficiency/finding feedback loop.