At first glance, the logic seems self-evident: a brand's media peak in a given month is the audience's reaction to an event that occurred in that same month. Klarna announced AI-driven layoffs in February 2024 — and we observe a spike in February. PayPal cut 9% of its workforce on January 30 — and the January peak reflects the reaction. The logic appears linear: event → reaction → peak.

Working with ICDS data across two structurally different sectors — fintech and consumer brands — we identified a systematic pattern that challenges this assumption. Comparing the tonality profile of individual peaks against the preceding period, we found that the aggregate tonality of a media peak is not a clean imprint of a single event. It is a mixture — a layered cake — composed of two fundamentally distinct information streams.

The first stream is inertial. It encompasses the ongoing organic conversation about the brand: product reviews, user questions, feature discussions, reshares of earlier content. This stream does not stop when a crisis or significant event occurs: people continue writing about what concerned them the previous month, carrying with them the tonal profile of that prior period.

The second stream is reactive. It consists of new mentions directly triggered by the event: news commentary, analytical breakdowns, emotional social media reactions. This stream reflects what the audience actually thinks about the event.

This paper formalises the model, applies it to 11 fintech media peaks and 8 cases from consumer categories, and demonstrates that the masking effect is not only reproducible across sectors but differs systematically between them.

A brand's media space is not an empty hall where sound appears and disappears instantaneously. It is more like a hall with a constant background noise level. At any given moment, thousands of people are discussing PayPal not because something new has happened, but because PayPal is part of their everyday financial lives — they complain about fees, praise Venmo, ask about refunds. This background stream is stable and predictable: it approximates the brand's average monthly baseline mention volume.

When a trigger event occurs, the reactive stream overlays the inertial one — it does not replace it. A blogger who was preparing a product review in January will still publish it. Its tonality will be shaped by personal product experience, not by the layoff announcement. But in the aggregate data, that review will blend with thousands of reactions to the crisis — and dilute the overall picture.

In consumer categories, this effect is structurally more pronounced. Fashion, beauty, consumer electronics, and K-Beauty brands generate a continuous high-volume organic content stream — reviews, unboxings, tutorials, lifestyle content — produced independently of any corporate event. This is why, as we show below, the inertial layer in consumer categories reaches a median of approximately 62% of peak volume — nearly twice the fintech figure.

We propose the following decomposition. Total peak volume (V_t0) equals the inertial layer (V_base) plus the reactive layer (V_t0 − V_base), where V_base is the brand's average monthly baseline mention volume over the observation period.

Solving for pure reactive tonality:

For baseline model verification, we used a sub-sample of 11 media peaks across 9 fintech companies (2024). The sample covers the full event spectrum: positive milestones (Stripe valuation growth, Nubank's 100 million customers), ambiguous signals (Klarna's IPO filing, Wise's full-year financial results), and crises (PayPal layoffs, Robinhood's SEC Wells Notice, Klarna's AI-driven headcount reduction, Synapse bankruptcy, Evolve Bank data breach).

The inertial layer accounts for between 20.6% and 40.8% of total peak volume. Catastrophic crises (Synapse: 20.8%, Evolve Bank: 20.6%) generate such a powerful reactive stream that the inertial layer is effectively overwhelmed — it represents only one-fifth of total volume. Moderate events (Wise: 40.8%, Nubank: 39.8%) leave roughly 40% of space to the inertial background.

In 10 of 11 cases, the inertial layer dilutes the reactive signal. Three polarity reversal cases are particularly instructive — where a positive inertial background masked the genuine negativity of the event:

The single exception is Wise, where the inertial layer amplified rather than diluted observed tonality. Reactive tonality was mildly positive (+0.16), but residual positivity from the prior month (+0.38) lifted the observed result to +0.25. The market saw through it: the share price fell 17.8% despite aggregate positive tonality.

We applied the same methodology to 8 media peaks drawn from the ICDS consumer brand corpus: YouScan, December 2023 — November 2025. The sample was deliberately selected to span different categories and event types — from viral positive moments to advertising scandals, from K-Beauty influencer waves to post-sanctions product launches.

The first observation is a systematically elevated inertial layer. Unlike fintech (range 20–41%), consumer categories show an inertial layer of 54.6% to 89.3%, with a median of approximately 62%. In a typical consumer brand peak, roughly two-thirds of all mentions are organic background continuing independently of any event.

The second finding is proportionally stronger masking. The median masking in the FMCG/Consumer sample (~52%) is roughly double that of fintech (~21%). This is a direct consequence of inertial layer depth: the larger the ballast, the more heavily it suppresses the reactive signal.

Two polarity reversals illustrate the mechanism most vividly:

Joint analysis of both corpora yields three patterns that appear structural rather than incidental.

Pattern one: the inertial layer grows with the density of organic content in a category. Fintech generates mentions predominantly around events — regulatory news, product launches, earnings releases. Consumer brands — particularly in fashion, beauty, and K-Beauty — generate a continuous lifestyle stream regardless of corporate events. Accordingly, all else equal, a media peak in consumer categories carries a heavier inertial load.

Pattern two: catastrophic crises suppress the inertial layer in both sectors. Synapse's bankruptcy (inertia 20.8%) and the Evolve Bank data breach (20.6%) both produced inertial layer figures at the level of a fintech crisis, not fintech normal. The same mechanism should operate in consumer categories: a sufficiently severe crisis can overwhelm the organic stream. Our current sample contains no such cases — a direction for further investigation.

Pattern three: observed tonality is a systematically biased metric. In the FMCG/Consumer corpus, median masking is approximately 52% — meaning that a brand analyst looking at aggregate peak tonality sees less than half of the real emotional signal. For brands with dense organic backgrounds and moderate events, observed tonality carries almost no informational content about the triggering event.

For the publicly traded companies in the fintech sample, we compared reactive and observed tonality against price volatility. Reactive tonality correlates with volatility more closely than the observed aggregate figure.

Robinhood is the most instructive case. Observed tonality (−0.11) pointed to near-neutrality, implying minimal market reaction. Reactive tonality (−0.25) signalled a substantially more negative picture. Actual price volatility — 76.5%, the highest in the sample — aligns far more closely with the reactive measure. Wise confirms the inverse mechanism: the artificially inflated positivity of observed tonality did not deceive the market, and the share price fell 17.8%.

The most intriguing question is whether different market participant types react to observed or reactive tonality. Algorithmic traders scanning news feeds in real time most likely work with the reactive stream — filtering out background noise. Swing traders most likely observe the aggregate (inertia-diluted) tonality. This creates a potential information asymmetry that may explain part of intra-month price volatility. The hypothesis requires verification on daily-granularity data.

The Layered Cake model introduces a methodological correction to standard tonality monitoring practice. Aggregate media peak tonality is not a clean signal of audience reaction to an event. It is a mixture of reactive and inertial streams: in fintech the inertial layer accounts for 20–41% of peak volume; in consumer categories, 55–89%.

Verification across two independent corpora — fintech and FMCG/Consumer — demonstrates that the masking effect is reproducible across sectors but differs systematically in scale. Median masking in FMCG/Consumer (~52%) is roughly double that of fintech (~21%). For brands with dense organic content streams — particularly in fashion, beauty, and lifestyle categories — observed tonality is a systematically biased indicator: it mutes the true force of both positive and negative events by approximately half.

The practical implication is straightforward: when analysing media peaks, practitioners should perform the decomposition and compute reactive tonality. This is especially critical in polarity reversal situations — where the tonal background of the preceding period diverges sharply in sign from the current event.