Environmental marketing claim substantiation file
Maintained by: R3CKON LLC, Estes Park, Colorado (r3ckon.com) Prepared: 2026-08-06 · Regenerated: 2026-10-22 · methodology 1.3.0 · pricing v4 (fixed baskets) Purpose: the written substantiation contemplated by Cal. Bus. & Prof. Code §17580 for the environmental marketing statements made on r3ckon.com, and the reasonable-basis record contemplated by the FTC Green Guides (16 CFR Part 260). This file is public at r3ckon.com/substantiation and is furnished to anyone who asks (info@r3ckon.com).
This file is regenerated when claims, pricing, or methodology change. The claim inventory below maps every live statement family to its evidence. Items marked TO OBTAIN are evidence not yet on file; the claims that depend on them are not made until it is. Nothing in this file is itself a marketing claim.
1. What the product claims, and what it does not
R3CKON makes activity claims only: that usage was metered; that a footprint was estimated under a published, versioned methodology with stated uncertainty; and that each paid plan's fixed monthly basket of specific third-party certificates was retired, identified by project, quantity, and vintage. The platform makes no neutrality, matching, coverage, compensation, or settled-status claims, and it does not compare the basket with any estimate. Its Terms license customers to repeat only the statements the record supports. The software scans every buyer-facing surface for banned vocabulary (test/claims.test.ts), and the claim renderer throws rather than emit a figure without its required companions.
2. Claim inventory and evidence map
| Claim family (where made) | Statement substance | Substantiation |
|---|---|---|
| Estimates (console, certificates, records, homepage) | "Your usage-phase footprint is estimated at X, 90 percent plausible range Y to Z, methodology N" | The published methodology (r3ckon.com/methodology, docs/METHODOLOGY.md): every coefficient with central, P05/P95, input range, boundary, source, vintage, and grade in src/core/methodology/configV1.ts; closed-form lognormal propagation; per-estimate persistence of resolved coefficients for reproduction. Calibration in §4. |
| Fixed retirement basket (pricing, Terms §6, certificates, emails) | "Each paid plan retires a fixed monthly basket of N gallons of BEF Water Restoration Certificates®, M kg CO2e of carbon credits, and K kWh of RECs" | Quantities pinned in src/core/config/pricing.ts (retirementBasketByTier) and published on /pricing. Accrual of exactly the basket per close: src/core/jobs/reconcile.ts (whole-unit plans) and src/core/jobs/poolClose.ts (pooled plans), pinned by test/reconcile-lots.test.ts, test/pool-close.test.ts, test/delivery-rails.test.ts. |
| Basket sizing note (methodology page only) | "Baskets were sized at the 90th percentile of a chat-shaped workload at the full allowance; coding and agentic workloads can model higher" | docs/METHODOLOGY.md "How basket sizes were chosen"; the chat-shape invariant and the coding-shape figures are asserted by test/pricing-floor.test.ts, which also checks the doc quotes the code's numbers. |
| Pooled personal plans (pricing, Terms §6, contribution certificates) | "Each month R3CKON buys enough whole certificates to meet every member's basket in full, rounding up, and allocates each account its basket as a share of the named retirement" | Round-up allocation and company-owned surplus in src/core/jobs/poolClose.ts and migration 034 (pool_rounds surplus columns, pool_round_lots, pool_member_lot_shares); allocations never exceed a lot or a member's entitlement and conserve exactly (test/pool-close.test.ts, test/pool-categories.test.ts, test/delivery-rails.test.ts). The supplier confirmed the pooled model in writing (BEF, 2026-08-10 and 2026-08-24, on file), on the condition that claims belong to the client. |
| Retirements (certificates, records, /disclosures) | "Funded the retirement of N units, project P, vintage V, reference R" | supplier_orders, supplier_order_lots, retirements; per-lot linkage ties each certificate line to the order or inventory lot that supplied it; pending orders are never presented as confirmed. Demo-tenant records are labelled synthetic everywhere. |
| Holder of record (certificates, records, Terms §6) | "Instruments are retired in R3CKON LLC's name for the benefit of the named account" | The supplier's retirement certificate honours a single named purchaser (sample on file); the reason line agreed with BEF is "R3CKON LLC, retired on behalf of platform users for usage in {month}" (draft agreements of 2026-10-22). |
| Timing (pricing, Terms, confirmation email) | "Retirements typically confirm, and certificates arrive, three to six weeks after the month ends" | Supplier operations (docs/plans/supplier-operations.md, from BEF 2026-08-24): invoice within 24 hours, retirement about every two weeks after payment; R3CKON pays on invoice. Draft agreements: delivery within 30 days of payment. |
| Carbon projects (/disclosures, certificates) | Project identity, location, type (emission reduction, not removal), registry, protocol and version, start date, crediting period, verified reductions, verifier, vintage, CCP label status, and the additionality and legal-requirement result (Green Guides 260.5(c)) | src/core/suppliers/basketProjects.ts, all retrieved 2026-10-24, with the source documents kept in docs/evidence/registry/. CAR512: Climate Action Reserve project page; vintage 2024 Verification Report and Opinion (Agri-Waste Technology, Inc., reasonable assurance, 88,872 tCO2e) showing the Legal Requirement and Performance Standard tests passed; developer's Attestation of Regulatory Compliance. ACR1200 (registry title "A-Gas V22"): ACR registry record; Validation and Verification Report and Opinion (TÜV SÜD America, Inc., 2025-12-31, positive, 1,690,094 tCO2e for 2025-01-08 to 2025-06-18) showing the Regulatory Surplus Test and Performance Standard passed; GHG Project Plan V5. EPA's Emissions Reduction and Reclamation rule (published 2024-10-11) first requires reclaimed HFCs for servicing on 2029-01-01, after this vintage; any later HFC vintage is checked against it before it is listed. |
| CCP label (certificates, /disclosures) | "ICVCM CCP label carried / not carried at purchase, per the supplier's menu; confirmed against the registry record for the serials retired" | BEF carbon menu, August 2026 (on file); registry tag checked per lot at confirmation (TO OBTAIN per lot). |
| BEF Water Restoration Certificates® (certificates, records) | "N BEF Water Restoration Certificates®, N x 1,000 gallons of water restoration as documented by BEF"; "restored gallons are streamflow benefit in the named river, not comparable with water used to serve AI requests" | BEF WRC menu (August 2026) and retirement certificate sample (on file) naming the verifier per project (Watercourse Engineering; National Fish and Wildlife Foundation). |
| RECs (certificates, records, home page) | "N RECs (N MWh) from {resource} generation, vintage V, from BEF's Green-e® Energy certified portfolio; the certification covers BEF's RECs, not R3CKON; a REC is not a statement about the electricity that served the usage" | BEF REC draft agreement (Green-e® Energy certified; product content label RY2025/2026, on file); Green-e® Code of Conduct v2.4 posture in src/core/config/claims.ts (GREEN_E). Per-lot resource type and vintage come from the delivery document (TO OBTAIN per lot; certificates say "pending" until then). |
| Automatic renewal (pricing cards, /subscribe, Stripe checkout, emails, Terms §6a) | Renewal terms, cancellation online, no refund of paid periods except where required, 7 to 30 day price-change notice | One source of copy (src/lib/renewalTerms.ts); separate consent recorded with the copy shown (consent_records, migration 035); acknowledgment email awaited and recorded; renewal reminders (src/core/billing/renewalReminders.ts); price-change notice route (src/app/api/admin/price-change-notice). |
| Metadata-only boundary (docs, Terms) | "The platform never receives prompt or completion content" | Deep payload scan rejecting content-shaped keys with HTTP 422 before parsing; no content columns in the schema; enforced by tests. |
| Scope naming | "Usage-phase footprint," never "the footprint of AI" | Metric definition and out-of-scope disclosure (training, hardware lifecycle beyond the embodied adder, network, devices) in the methodology. |
Retired claim families, kept for certificates issued under them: the v2 certificate-share floor ("at least 50 percent of subscription revenue") and the P90 accrual model. Neither is made on any live surface.
3. Coefficient provenance (summary; the config file is the record)
Every value in src/core/methodology/configV1.ts carries its source, vintage, and derivation inline. Principal anchors at 1.3.0:
- Per-token energy: two evidence families that disagree by roughly an order of magnitude: top-down provider disclosures (Google 2025, arXiv:2508.15734; Epoch AI 2025; Mistral 2025) and parametric fits (EcoLogits). Per-token P05 values sit at the top-down family and centrals at the geometric midpoint of the two; for a 1,000 input / 400 output frontier request the published 0.24 Wh and about 0.3 Wh per prompt fall below the modeled P05.
- Idle/reserve uplift: central 1.571, modeled 1.30 to 1.90 (fit floor 1.30, elicited range 1.30 to 2.4), from the fleet-level accounting in the Google measurement paper.
- PUE / WUE: operator fleet reporting (Google, Microsoft, AWS) at fleet level, every PUE floored at 1.0. Footprint water is defined as consumption; AWS and Meta report withdrawal, used unconverted as an upper bound and flagged as such.
- Off-site water: 1.95 L consumed per facility kWh (arXiv:2607.02531, built on Macknick et al. 2011 and Meldrum et al. 2013 consumption factors).
- Grid carbon: EPA eGRID2023 rev.2 subregion CO2e rates (US); Ember 2025 national intensities, lifecycle CO2e (non-US). Regions without a recognized mapping use the unrecognized-region default, the maximum over mapped zones (671 gCO2e/kWh).
- Embodied carbon adder: 43 [25, 100] gCO2e/kWh (NVIDIA H100 product carbon footprint, Gupta et al., Boavizta).
- Hidden-reasoning multiplier: 4.0, modeled range 1.1 to 14.6, floor 1.0 (ML.ENERGY, arXiv:2505.06371); applied to unclassified models too, so the unclassified class is the most conservative row.
4. Calibration against the published measurement
The only public comprehensive per-prompt measurement is Google's median Gemini Apps text prompt (May 2025): 0.24 Wh, 0.26 mL. test/calibration.test.ts gates every release on the modeled 90 percent interval for a same-shaped prompt (250 input / 100 output) containing both anchor values, and on the frontier-reasoning median staying within stated factors of the anchor. For a larger 1,000 / 400 request the anchor sits below P05, which the methodology page states plainly.
5. Known limitations, stated where the claims are made
- The interval covers parameter uncertainty only; model form, tokenizer, the treated-as-independent PUE/WUE pair, and excluded scope sit outside it.
- Hidden reasoning volume, closed-model parameters, and site-level grid mix are unobservable from outside.
- Water restoration in one basin does not un-consume water in another, and restored gallons are a different quantity from water used; every water statement carries the geography disclosure.
- Provider-level REC matching can overlap customer REC retirements; disclosed in the methodology, the Terms, and on every REC line.
- A single supplier (BEF) serves all three categories. Substitution of a carbon project happens only after it is listed on /disclosures.
6. Section 17580(a) items
- Reasons the product is marketed with environmental statements: the service retires third-party environmental certificates; the statements describe that activity.
- Source of the claims: the evidence in §2; supplier documents on file.
- Adverse environmental impacts of the product: the service itself is software; its own usage-phase footprint is not yet estimated, and no claim about it is made.
- Measures taken to reduce environmental impacts: none claimed.
- Violations of federal, state, or local permits: none.
- Whether the product conforms to the Green Guides: designed to; see §1.
7. Records retention
Substantiating records live in the production database (usage events, estimates with resolved coefficients, supplier orders and lots, retirements, pool rounds and shares, certificate JSON, consent records) and in the repository (this file, the methodology and its changelog, the claims configuration, the test suite, and docs/reviews/). Consent records are kept at least three years and survive account deletion. Historical estimates remain reproducible under the methodology version that produced them.
8. Requests
Requests for this file or the underlying records: info@r3ckon.com.