Project update — June 2026: Canary 2.0 concluded on 4 June 2026 following completion of its pre-registered hypothesis test. The content on this page remains live as a record of the project methodology. For the full project conclusion and results, see the Project Conclusion and Final Report.
Sentiment, Scoring & Methodology
What does KnowEntry mean by “sentiment”?
Sentiment here means the measurable tone of news coverage, not investor positioning or market prices. Each morning Canary read hundreds of AI sector articles and scored them on a scale from -1.0 to +1.0, where -1.0 was clearly negative coverage and +1.0 was clearly positive. Before scoring, each article also got a relevance rating from 4 to 10 based on how directly it covered the topic in question. The daily figure on the dashboard was a relevance-weighted average, so articles that were squarely about AI chips counted for more than articles that mentioned AI chips in passing. This was editorial sentiment, how journalists and analysts were framing developments, not a market signal.
How is KnowEntry sentiment different from market sentiment?
Market sentiment is about investor behaviour, positioning, and prices. KnowEntry sentiment was about how journalists and analysts were framing AI sector developments in financial and technology news. The two can diverge significantly and often do. News narratives, investor positioning, and prices update on different timescales and respond to different information. A run of strongly positive Canary scores did not imply rising prices, and a run of negative scores did not imply falling ones. Canary measured the narrative environment, not the investment environment.
Was Canary a trading or investment signal?
No. It measured narrative tone and narrative change across AI sector news coverage. It did not predict prices, valuations, or market outcomes. Any use of Canary data in financial decision-making is entirely at the user’s own discretion and risk. That use is not implied or encouraged by KnowEntry.
What are Canary’s eight narrative frames?
Every article was assigned to one of eight frames: Growth Momentum, Technical Breakthrough, Financial Results, Regulatory Risk, Geopolitical Risk, Competitive Threat, Market Correction, and Macro Environment. The frame captured what kind of story was being told, not just whether the coverage was positive or negative. Two days with identical sentiment scores can represent entirely different situations if the mix of story types differs. A day dominated by Financial Results stories carries different implications from a day dominated by Regulatory Risk stories, even when both produce a similar headline number.
What is the Semantic Volatility Index?
It was a composite score from 0 to 100 that was designed to detect narrative instability before it shows up in sentiment scores. It aggregated five sub-components with different timing characteristics: Sentiment Volatility over a rolling 7-day window (lagging), Within-Day Dispersion of article scores (concurrent), day-over-day Narrative Shift (concurrent), Vocabulary Drift measured using Jensen-Shannon Divergence (leading), and Frame Distribution Shift also using Jensen-Shannon Divergence (leading). The composite was weighted toward the two leading components because the whole point was to surface change early. A score above 75 indicated a possible regime shift.
Is the Semantic Volatility Index a proprietary system?
No. The SVI is an open research methodology developed as part of the Canary 2.0 project. The sub-components, weightings, calibration approach, and analytical framework are all documented on this site. There is no commercial product, no licence, and no restriction on discussing or building on the concepts. Canary 2.0 was a public experiment in AI sector narrative monitoring, not a vendor offering. If you have seen it described elsewhere as proprietary, that is a mistake worth correcting.
What is a narrative regime shift?
It is when the fundamental character of AI sector coverage changes, not just the level of sentiment but the vocabulary and framing. A shift from earnings-driven Growth Momentum coverage to Geopolitical Risk coverage can happen at similar sentiment levels but represents a qualitatively different environment. The SVI was designed to detect that kind of change early, before it becomes obvious in the headline numbers.
How often was Canary updated?
Scores were published each morning, based on the previous day’s coverage, from February 2026 until the project concluded on 4 June 2026. The dashboard now shows the final readings from that date.
What six categories did Canary cover?
AI Chips and Hardware, AI Platform Hyperscalers, AI Enterprise Software, Data Centre Infrastructure, AI GPU Cloud, and Pure Play AI. These represent distinct layers of the AI value chain from infrastructure through to application. Category-level divergence, where one layer strengthens while others weaken, was often more informative than the overall headline score because different parts of the stack respond to different news drivers.
What does the 95% confidence interval on Canary scores mean?
Each daily score came with a confidence interval representing statistical uncertainty around that day’s estimate. A narrow interval means the articles scored that day were broadly consistent in tone. A wide interval means either fewer articles were available or sources disagreed significantly in their framing. A wide interval is a signal to interpret cautiously, not an error in the system.
How should historical Canary scores be compared?
Each category is most meaningful compared against its own recent history rather than against other categories or fixed thresholds. Canary used Welch’s t-test to flag days that are statistically unusual relative to that category’s 30-day baseline. A p-value below 0.05 means there is less than a 5% probability the reading is random variation. A score of +0.3 can be unremarkable for one category and significant for another, which is why the statistical flag matters more than the raw number. Sentiment direction, frame distribution, and the confidence interval should all be read together.
Where did Canary’s data come from?
Three commercial news APIs: NewsAPI, NewsData, and Finnhub. Between them they covered most of the major financial and technology publishers. After fetching, articles went through deduplication to remove cross-source repeats, then a relevance filter. Typically 450 to 650 articles were scored each day from a raw fetch of 800 to 900. The same model that scored sentiment also made the relevance judgment, so the filtering and scoring were consistent.