What is carbon footprint assessment?

Carbon footprint assessment measures Scope 1 (direct), Scope 2 (electricity), and Scope 3 (supply chain) greenhouse gas emissions from your organization, aligned with ISO 14064 standards.

What is CBAM compliance?

CBAM (EU Carbon Border Adjustment Mechanism) is EU Regulation 2023/956 requiring non-EU companies exporting to the EU to report embedded emissions and purchase CBAM certificates. Carbonaa's simulator helps calculate your exposure and compliance roadmap.

How does satellite-powered emissions monitoring work?

SatClimate uses real-time satellite imagery and AI to detect industrial emissions patterns, methane leaks, flaring activity, and fraud indicators across 100+ global datasets. This provides real-time carbon market signals and compliance monitoring.

What is carbonaa plugin?

carbonaa plugin helps stores, miners, and data centers manage carbon offsets with merchant tools, climate verification, pricing, collaboration, and 295 pages.

Model Card — contextual_signals_v1.5

EU AI Act + SR 11-7 · Platform v4.1 · 2026-05-06 · Ref: SATCLIMATE-MODELCARD-CS-2026-001

EU AI Act Art. 13 · SR 11-7 · Confidential

Model Card

contextual_signals_v1.5

SatClimate Intelligence Platform — Carbonaa SA

Model: contextual_signals_v1.5

Card Version: v1.0

Platform: v4.1

Issue Date: 2026-05-06

Doc Ref: SATCLIMATE-MODELCARD-CS-2026-001

Pipeline Stages

Non-Deterministic Step

Risks Assessed

EU AI Act Articles

Self-Assessment Notice: This model card is produced by Carbonaa SA. No independent third-party model validation has been completed. All gaps and limitations are disclosed explicitly. See SR 11-7 alignment section for validation status.

model id

contextual_signals_v1.5

model type

Hybrid deterministic + LLM-constrained probabilistic scoring

version

1.5

release date

2026-05-06

platform version

v4.1

schema version

1.5

doc ref

SATCLIMATE-MODELCARD-CS-2026-001

owner

Carbonaa SA — SatClimate Intelligence Platform

primary use

Generate directional carbon market intelligence signals (BULLISH / BEARISH / NEUTRAL) from satellite-derived environmental observations

intended users

Carbon traders, ESG investors, energy traders, regulatory compliance teams, institutional data consumers

deployment env

Production SaaS — sat.carbonaa.net — accessible via authenticated REST API

not intended for

Autonomous trading execution, real-time critical safety systems, retail investor guidance, certified regulatory submissions

Stage 1 — Environmental Observation Ingestion

Raw satellite feeds processed by satclimateDataAPI (environmental_obs_v1.0): NASA FIRMS VIIRS fire detections, Open-Meteo ERA5 climate anomalies, OpenAQ atmospheric emissions, USGS seismic activity. Each observation produces fetch_status, data_gap_flag, signal_mode, fallback_reason, and observed_at fields.

Stage 2 — Signal Parameter Derivation (Deterministic)

Market, commodity, sector, and segment input parameters are validated server-side. Invalid segment/market/commodity combinations return 400. Segment-specific context maps are applied deterministically. Input validation rejects unknown enums before any model computation.

Stage 3 — LLM Probability Scoring (Constrained)

A large language model (LLM — not user-accessible, constrained in system prompt) is invoked with a structured input prompt containing the validated parameters. The LLM outputs exactly ONE integer in range [5, 95] representing probability_score. All other fields are explicitly blocked from LLM output. This is the sole non-deterministic step.

Stage 4 — Deterministic Signal Derivation

All remaining signal fields are computed deterministically from the probability_score integer: direction (BULLISH/BEARISH/NEUTRAL via threshold mapping), confidence_score (formula: base(segment) + 0.5×|score−50| + neutral_penalty, clamped [42,94]), risk_level, color_semantic, market_impact_level, confidence_factors[], confidence_methodology string.

Stage 5 — Provenance Tagging & Audit

Each signal receives: signal_id (unique), schema_version (1.5), model_version (contextual_signals_v1.5), generated_by_engine, observed_at, generated_at, provenance_hash (FNV-1a deterministic hash of direction+confidence+market+commodity+schema_version), data_sources[] array. AuditLog entry created for every signal delivery.

Stage 6 — Fallback Governance

If LLM invocation fails, signal_mode is set to "fallback", fallback_reason is populated, confidence_score is set to 0, color_semantic is set to "grey". Fallback signals are served — they are not blocked — but are visually and semantically distinguished. Redis circuit breaker is FAIL-CLOSED: Redis unavailable → 503, never serves unprotected quota.

Directional Accuracy (EU ETS)Not yet formally validated

Backtesting framework exists (crisisForecastEngine); external validation against realized outcomes not completed

Confidence Score CalibrationNot externally calibrated

Formula: base(segment) + 0.5×|prob−50| + neutral_penalty. No empirical calibration against historical outcomes.

Fallback RateMonitored per deployment

signal_mode field tracks live vs fallback. Fallback triggered by LLM timeout or Redis circuit OPEN.

Provenance Reproducibility100% deterministic

Given stored AuditLog parameters, any historical signal can be reproduced exactly from code.

LLM Output Constraint Compliance100% — validated at runtime

LLM output parsed as integer; any non-integer output triggers fallback. No free-text fields derived from LLM.

Quota Enforcement AccuracyAtomic — no known bypass under concurrent load

Redis Lua INCR — single atomic operation. Fail-closed circuit breaker prevents bypass on Redis failure.

LLM HallucinationContained

LLM output is constrained to a single integer [5–95]. All other fields are deterministic. Hallucination cannot propagate beyond the probability_score value.

Model Drift (LLM)Disclosed

LLM model version is not pinned — may change as provider updates. model_version field (contextual_signals_v1.5) refers to the SatClimate schema version, not the underlying LLM. No formal drift scanning in place.

Confidence Score OverinterpretationMitigated by disclaimer

confidence_methodology string published verbatim in every signal response. Disclaimer field explicitly states: "Confidence reflects signal consistency, not probability of financial outcome."

Data Gap PropagationMitigated by flags

data_gap_flag, fetch_status, signal_mode, and fallback_reason fields expose all upstream data failures. Confidence_score = 0 when data unavailable.

Segment/Market Coherence FailureBlocked at input layer

Server-side enum validation rejects invalid combinations before any model computation. 400 error with valid values returned.

Quota Bypass Under Concurrent LoadResolved (v4.1)

Redis atomic Lua INCR + fail-closed circuit breaker. Quota cannot be bypassed even under concurrent requests or Redis unavailability.

Autonomous Decision-MakingContractually prohibited

Terms of Engine v3.0 §1 explicitly prohibits use for autonomous trading. Disclaimer in every signal: "Human interpretation required. Not for autonomous execution."

Art. 6 — Risk Classification

Designed toward

Signals are informational intelligence tools, not autonomous decision systems. Not classified as high-risk AI per Annex III. Formal conformity assessment not completed.

Art. 13 — Transparency

Met

confidence_methodology published in every response. disclaimer field present. model_version, model_inputs, provenance_hash exposed. No hidden transformations.

Art. 14 — Human Oversight

Met

Explicit "human interpretation required" disclaimer. No autonomous execution path. Signals feed human decision processes only.

Art. 15 — Accuracy & Robustness

Partial

Deterministic derivation from validated inputs provides robustness. Formal accuracy validation against realized outcomes not completed. Fallback governance prevents silent failures.

Art. 17 — Quality Management

Partial

AuditLog, provenance_hash, model_version tracking present. Formal quality management system not documented. No certified third-party audit.

Model InventoryMet

contextual_signals_v1.5 registered in this model card with unique DOC_REF. environmental_obs_v1.0 documented separately.

Model Development DocumentationMet

Architecture pipeline documented in 6 stages. Input validation, LLM constraint mechanism, deterministic derivation, and fallback governance all documented.

Model ValidationGap

No formal independent model validation completed. Internal backtesting framework exists but results not published. Third-party validation not yet engaged.

Model Performance ScanningPartial

AuditLog records all signal deliveries. Fallback rates monitored via signal_mode field. No automated drift detection or performance alarming.

Model Risk RatingPartial

Qualitative risk assessment completed in this model card. No formal quantitative model risk rating assigned by an independent risk function.

Ongoing ScanningPartial

AuditLog + SecurityAuditLog provide activity scanning. No formal periodic model review schedule defined.

Permitted Uses

• Internal research and market analysis
• Informing proprietary investment research processes
• Regulatory reporting with SatClimate as disclosed source
• Due diligence and ESG screening support
• Academic research with proper attribution

Prohibited Uses

• Autonomous trading execution without human oversight
• Representing outputs as certified or audited data
• Real-time critical safety or infrastructure systems
• Redistribution or white-labelling as proprietary data
• Creating competing indices or benchmarks

SATCLIMATE-MODELCARD-CS-2026-001 · Model Card v1.0 · Platform v4.1 · contextual_signals_v1.5 · Carbonaa SA · amin@carbonaa.net

This model card is provided for institutional due diligence under EU AI Act Art. 13 and SR 11-7. Self-assessed — not independently audited.

Model Card

1. Model Metadata

2. Model Architecture Pipeline

3. Performance Disclosure

4. Risk Register & Mitigations

5. EU AI Act Alignment

6. SR 11-7 Model Risk Management Alignment

7. Intended Use & Constraints