A clear walkthrough of the diagnostic. The questions, the maths, what's validated, what's modelled and what the score does and doesn't tell you about your organisation.
A note on what's shared here
This is a light public summary of the Signal Pulse model. It covers the structure of the diagnostic, the headline maths and the validation sample. It doesn't include item-level weights, segment baselines, intervention mappings or anything else that's commercially sensitive. If you need a deeper methodology pack for procurement, internal audit or technical due diligence, please get in touch and we'll share a full version under NDA.
The diagnostic is short by design. Long instruments train respondents to satisfice. The 18 items cluster into three drag dimensions and three capacity dimensions, plus one cultural-state item. Every item uses the same 1, 5 scale.
Workflow drag covers decisions reversed, duplicate work, coordination breakdown. Data drag covers reporting friction, version disputes, time spent assembling information instead of using it. Leadership drag captures whether decisions get unmade above the level of the person doing the work.
Higher scores mean more friction. The three dimensions average into the headline Complexity Tax.
Clarity covers shared understanding of priorities. Speed and scale covers the organisation's ability to move once it's decided. Leadership enablement captures whether the senior layer removes blockers or creates them.
These items are reverse-scored so that higher always means better. They combine into the Efficiency Dividend.
A single question that captures whether the organisation still feels the friction or has accepted it as the way things are. Sensitive cultures (1, 2) make recoverable friction visible. Normalised cultures (4, 5) bake it in. This item doesn't change the Tax score itself but multiplies the Net Signal up or down by 20%, since recoverable friction is worth more than friction that's already invisible.
Three are external (statutory and market data), two are internal (the validated research sample and the aggregated submission base) and one is the live diagnostic response itself. Where a source is live in production we mark it. Where it's on the roadmap we say so. Nothing on this page is aspirational.
Real-time API lookup against the UK statutory registry. Returns company number, status, registered address, incorporation date and officer-level data where relevant. Used to pin every UK submission to a real legal entity.
For any submission resolved to a public ticker, we pull a 33-metric financial snapshot at the moment of submission. Revenue, operating profit, EBITDA, margins, headcount, sector classification, year-on-year deltas and a benchmark index for the relevant exchange.
A curated lookup that maps company names to exchange tickers across LSE, NYSE, NASDAQ, Euronext, SIX, Deutsche Börse, Tokyo, Hong Kong and the major Australian and Indian exchanges. Index pairing happens automatically so YoY performance gets compared to the right benchmark (FTSE, S&P, DAX, CAC, SMI, etc).
The US equivalent of Companies House for any organisation that files with the SEC. On roadmap for 2026 to give US private-company submissions the same statutory anchoring UK submissions already have via Companies House.
The instrument was fitted on 443 clean responses collected via Prolific across three research stages (January to April 2026). UK and US knowledge workers across function, seniority and sector. This is what gives every score a real population reference rather than a synthetic baseline.
Every completed snapshot adds to the aggregated, anonymised reference base, segmented by industry, company size band, role family and function. As volume grows the segment-level baselines tighten. The score isn't just compared to the original n=443, it's compared to the population that resembles your respondent.
The 18 scored items and 5 open-text fields the respondent submits in the 10-minute snapshot. Each open-text field is interpreted by an LLM with constrained driver vocabulary so qualitative signal feeds the score rather than sitting separately.
Timestamps, time-per-section, pauses and item revisits. We don't score this directly but we use it to flag patterns: respondents who race through the instrument, or who pause heavily on specific items, get the result framed with the right caveat. Quality control, not surveillance.
The Complexity Tax is the equal-weighted mean of the three drag dimensions. The Efficiency Dividend is the mean of the two capacity dimensions, multiplied by a leadership coefficient (0.8 to 1.2) that reflects whether leadership amplifies or dampens it. The Net Signal is the difference between the two, multiplied by a normalisation coefficient (also 0.8 to 1.2). Everything's bounded, defined and reproducible.
The headline number is plain arithmetic so any reader can replicate it. The layer beneath is where the real statistics live. Here's what's running inside the engine. Each technique is named precisely so a technical reader can stress-test it.
Each of the 17 scored items has a population β coefficient fitted via standardised regression against an external cultural anchor (the normalisation item). We use β against an external anchor specifically to avoid circularity, regressing on the Tax score itself would just describe how we computed it.
contribution = (item_value - 3) × |β_pop|
Every respondent is compared not just to the full sample but to the segment that resembles them. We compute z-scores against pre-fitted segment baselines (company size, role family, function, seniority) so a CFO's score is read against CFO norms, not against everybody.
z = (your_score - segment_μ) / segment_σ
The four bands (Lower, Moderate, High, Severe) use fixed thresholds for the headline placement, plus hysteresis on borderline cases so a score of 3.39 doesn't flip Moderate or High based on a rounding accident. The exact threshold values are published in the formula section above.
band = max(t ∈ T : tax ≥ t)
Two multiplicative coefficients shape the headline numbers. Leadership enablement multiplies the Efficiency Dividend up or down by ±20%. The cultural normalisation flag multiplies the Net Signal up or down by ±20%. Bounded, defined, documented in the code.
net_signal = (dividend − tax) × q_norm_mult
Open-text responses are interpreted by gpt-4o under a locked taxonomy: each response is scored on a 1-5 friction scale, tagged with a primary and secondary driver drawn from a fixed 10-driver vocabulary, and weighted by the relevant population β before flowing into the driver-attribution layer. Multi-label output, weighted contribution. On model failure or malformed JSON, a deterministic neutral fallback fires so a submission never returns a blank score.
driver ∈ {rework, handoff, tool sprawl, ...} × |β_pop|
The exposure pound figure on the results screen is an explicit Fermi estimate, headcount × hourly-rate proxy × friction coefficient derived from published research. We show a low, mid and high range, not a single number, and we label it "modelled" so a sophisticated reader knows what they're looking at. Order-of-magnitude reliable, not pinpoint accurate.
exposure ≈ headcount × rate × hrs(tax) × 52
The instrument was developed and validated through three rounds on Prolific between January and April 2026, pooling 443 clean responses across UK and US knowledge workers. The driver-level weights were fitted via standardised regression against a held-out cultural anchor (the normalisation item), avoiding circularity with the Tax score itself. Segment baselines for company size, role type and function are computed from the same dataset.
Across 511 respondents pooled from the research panel and the live diagnostic, two complexity drivers dominate the top of the list: decision delay (32.5% name it as their number one driver) and workflow friction (31.3%). Together, they account for 63.8% of all primary drivers. They sit in the same family. Both describe a room that can't reach a decision, and the work that piles up around it. We group them as "decision friction" because the underlying mechanism is the same: the meeting that should produce an answer doesn't, the approval loop that should close doesn't, the handoff that should resolve doesn't. The decision exists. The friction is the gap between the decision and the action that follows.
Why this matters. Most complexity diagnostics treat all drivers as equal. The data says they're not. Decision friction is structurally dominant. It travels across sectors, headcounts, and business models. The Pulse names the specific shape of decision friction in your organisation, alongside the dimensions where it concentrates. That's what makes the output actionable rather than diagnostic theatre.
The peer comparison shown on the results screen draws from a reference set of large consumer-goods organisations. If you operate in CPG or an adjacent sector, the benchmark is directly relevant. If you operate outside those sectors (technology, financial services, pharma, retail), the peer comparison should be read as directional rather than precise, since the underlying reference set isn't a perfect sector match. We're actively building out sector-specific reference sets and will note in the results when one is available for your category.
The headline pound figure ("modelled exposure: £X, £Y million") is a Fermi-style estimate, not a measurement. It combines your Tax band, an assumed average compensation level for the relevant company size and a friction-time coefficient derived from published research on organisational drag in knowledge work. It's calibrated to be directionally sensible rather than pinpoint accurate. Two organisations with identical scores can have meaningfully different real exposures depending on factors the instrument can't capture in 10 minutes.
Use the Tax band and driver list as the primary read. They're validated against a real population. Use the financial figure as a sense-check on order of magnitude: is this a £100k problem or a £10 million problem? It'll tell you which side of that line you're on, with confidence. The exact number is a model not a measurement, and the basis line under the figure states this explicitly.
The item-level regression coefficients, segment baseline tables, the driver-to-intervention mapping, the BX narrative prompts and the deeper engineering of the model aren't in this public summary. They're available under NDA for client procurement, internal audit or any other due diligence context where the detail matters.