Watt, our embedded energy market analyst, answers from verified data rather than from memory. This page sets out the guardrails that keep its answers honest, in plain language, and is candid about what we do and do not claim.
gridIQ runs on AI, and people make real decisions on what Watt tells them about prices, emissions, contracts and compliance. That only works if the answer is trustworthy. We did not bolt a chatbot onto a dataset and hope for the best. The way Watt is built is itself the guardrail, and the measures below are the ones actually implemented in the platform today.
A fabricated price or emission factor is not a quirk on an energy platform. It is a liability, and we treat it as one. Watt is constrained on several layers at once.
When you ask a question, Watt calls specialist tools that query verified sources: AEMO dispatch prices for the NEM, WEM prices for Western Australia, Clean Energy Regulator registers and your own data. A generic assistant writes an answer from its training data. Watt assembles one from real figures. There is no retrieval guesswork and no vector search approximating a result.
Watt is prohibited from stating any price, generation value, emission factor or historical figure that did not come from a tool call. This applies to recalled general knowledge too. Watt will never present a number it remembered as a gridIQ figure, and it will not invent the counts or denominators behind a percentage.
Watt explains why a price moved only when a tool supplies the driver: the generation mix, a market event, a binding constraint, a rebid or an interconnector flow. If it has the number but not the cause, it reports what the data shows and stops. A confident guess about causation is treated as a hallucination even when the number itself is real.
Some warnings are too important to leave to an instruction. AEMO constraint codes, for example, are equation references rather than plain-English causes, so the caveat that says so travels inside the tool response itself, attached to the field that could be misread. The model cannot ignore a warning that is part of the data it is reading. This is defence in depth.
If a tool returns 7.8%, Watt reports 7.8%, not 7.83%. It will not tell you how conditions will trend over coming months unless a forward-looking tool produced that trajectory, and it carries an indicative or screening-grade label through to the answer instead of overselling it.
When a query returns a long series, such as a year of five-minute prices, we trim it to whole records and tell the model exactly how many were left out. Watt never sees a dataset cut off mid-record, so it can caveat or decline a calculation rather than quietly averaging a biased sample.
A responsible system knows the edge of its own knowledge and says so. When Watt does not have the data, it tells you rather than filling the gap with a guess. Prices are shown at five-minute resolution and generation at half-hourly, with the timestamp noted so you can see how current the answer is. Emission factors are stamped with their regulatory vintage so a calculation can be audited and reproduced.
Estimates are labelled as estimates. A PPA backtest runs at daily resolution and is presented as a screening-grade figure, not a bankable dispatch model. A battery arbitrage backtest carries a documented capture-factor haircut. And the disclaimers are surfaced, not buried:
On a trading or contract question: “This is market analysis, not financial advice. Consult your risk team before acting on any position.”
On a regulatory obligation: “This is general guidance. Verify compliance requirements with your legal or regulatory team.”
gridIQ is a facts-and-data platform. We do not editorialise on energy sources. Coal is not described as dirty and renewables are not described as clean in any gridIQ output. Watt reports emission intensity, capacity factor, generation output and cost, and lets you draw your own conclusions.
This is not a soft preference. It is enforced across every AI surface: the chat, the daily narratives, the newsletter, the Grid Brief and the event analysis. Our audience includes coal generators, gas traders, sustainability teams and renewable developers in equal measure, and the data reads the same for all of them. Watt also holds no opinion on energy policy. It analyses market mechanics, not politics.
Watt runs on a deliberate, version-pinned model setup. Models are not swapped silently, because the choice affects accuracy, cost and behaviour.
Emission factors and similar reference values live in peer-reviewed code, not in anything the model writes freehand. The AI calculates with these values. It never invents them.
Real Watt conversations are reviewed for weaknesses. When we find one, we fix it by tightening the underlying tool and bounding what the model can claim, then we re-test to confirm the fix holds.
Errors are captured on the server, in the browser and at the edge, so a broken data feed surfaces to the team instead of silently degrading an answer.
Your sites, contracts, consumption and compliance information are scoped to your account and visible only to you. Your conversations with Watt and the data you upload are processed to generate your answers. They are not used to train the underlying AI models. What each plan can do is enforced on the server, so the AI cannot grant itself a capability your plan does not include, and every message is validated and length-limited before it reaches the model.
Watt is built for Australian energy markets and Australian disclosure frameworks: the NEM across the eastern states and South Australia, the WEM in Western Australia, NGER emissions reporting and AASB S2 climate disclosure. Our approach reflects the principles Australia expects of trustworthy AI. It is reliable and safe, transparent about its reasoning and its limits, fair to every market participant and accountable to the people who rely on it.
No AI system is infallible, and we will not pretend Watt is. The measures on this page are designed to make a wrong answer rare and a fabricated number far rarer, but they are not a guarantee. That is exactly why Watt shows its workings, labels its estimates and tells you to verify what matters.
gridIQ provides market data and analytics, not regulated financial or legal advice. PPA, Scope 2 and compliance outputs are designed to support your decisions and your reporting. Verify them with a qualified professional before you act or lodge.
We are happy to walk your team through how Watt is built and where its limits are.