Retrospective evidence cases

Published evidence examined through the Violet framework.

Each case separates the original result from Violet’s interpretation, practical suggestions, and unresolved questions. These are different from Applied Projects, where Violet is used on a live problem and records its prediction before later results exist.

Published field evidence Healthcare Updated July 23, 2026

NHS appointment reminders

When complete information still does not produce attendance

A reminder stating the approximate cost of a missed appointment reduced non-attendance from 11.1% to 8.4% in one trial. A second trial found 8.2% non-attendance for a specific-cost message versus 9.9% for a general cost statement.

Read the living analysis →

Violet interpretation

The essential appointment facts remained similar, but the concrete consequence may have changed Stake, credibility, memory, or why the receiver believed the number was included.

What it does not prove

The trials establish a local effect of message assignment on attendance. They did not directly establish which psychological mechanism changed.

Published field evidence Campaigns Updated July 23, 2026

WeChat social advertising

Why a familiar source is not equally persuasive for every decision

Showing that a friend endorsed an advertisement increased clicking on average, but the effect varied sharply by product and by the relationship, status, and involvement of the displayed friend.

Read the living analysis →

Violet interpretation

Source is not a badge attached to a message. Its value depends on who the source is to this receiver, for this decision, in this context.

What it does not prove

A click does not establish belief, product benefit, or a durable relationship effect.

Public-record retrospective Campaigns Updated July 25, 2026

The Harris 2024 Attention Gap

What voter issue salience, campaign message allocation, and issue ownership can—and cannot—tell us about the 2024 campaign

Pre-election surveys made the economy unusually consequential, and voters generally gave Donald Trump an advantage on the economy and immigration. Yet Kamala Harris had a detailed economic platform and her advertising frequently discussed economic issues. The useful question is therefore not whether the campaign mentioned the economy, but whether its communication created recall, credible ownership, and an actionable choice for the receiver groups where those issues mattered most.

Read the living analysis →

Violet interpretation

The evidence does not support a simple “wrong issues” verdict. It supports a testable alignment question. A campaign can possess Substance—a policy—and allocate substantial airtime to it while still failing to produce receiver-side recall or Source ownership. Conversely, a rival can dominate ownership through repetition, prior party reputation, news context, or dissatisfaction with incumbent conditions. Violet would locate the gap by receiver group, issue, date, and channel rather than infer it from the final result.

What it does not prove

The 2024 result had many plausible causes, including economic evaluations, policy preferences, incumbency, turnout, candidate perceptions, campaign organization, media coverage, partisanship, and events outside either campaign’s control. Public ad data do not cover every channel or private exposure, survey responses are not direct measures of attention or causal effects, and issue mention does not establish comprehension. This case does not claim that a different message allocation would have changed the winner.

Published experiments Public communication Updated July 23, 2026

Headlines and accuracy prompts

Clicking, believing, and sharing are different transitions

Large headline experiments show that wording changes can alter clicking for the same underlying story. Separate experiments found that briefly redirecting attention toward accuracy improved the quality of later sharing.

Read the living analysis →

Violet interpretation

The cue that wins attention can change the transition. Clicking, judging accuracy, sharing, and benefiting are separate outcomes with different thresholds.

What it does not prove

Results from headline and sharing environments should not be treated as universal evidence for every communication context.

Documented public event Public communication Updated July 23, 2026

One Growth Plan, Four Receivers

How an intended growth signal became a credibility, security, fairness, and political-survival signal

On 23 September 2022, Chancellor Kwasi Kwarteng presented the Truss government’s Growth Plan. Sterling fell, gilt yields rose, the Bank of England intervened on financial-stability grounds five days later, public opinion deteriorated, and major measures were subsequently reversed.

Read the living analysis →

Violet interpretation

The phrase “growth plan” supplied the sender’s intended frame, but receivers inferred consequences through their own jobs and priors. Substance and Source were inseparable: large fiscal changes arrived without a simultaneous Office for Budget Responsibility forecast, while the medium-term fiscal plan was deferred. High-Stake downside questions became salient before the promised growth path became credible. Repetition or rhetorical confidence could not, by itself, close that evidence gap.

What it does not prove

This sequence does not prove that wording caused the market or public reaction. Policy substance, the absence of a simultaneous independent forecast, global economic conditions, monetary-policy expectations, market structure, political priors, and later decisions all contributed. Violet’s redesign is a communication-governance hypothesis, not evidence that different presentation would have prevented the outcome.

Published field evidence Campaigns Updated July 23, 2026

Facebook election mobilization

When a social cue moves online expression and offline participation

A randomized Election Day experiment assigned mobilization messages to roughly 61 million Facebook users and linked treatment to online activity and validated voting records. Social messages that included friends’ faces produced more impact than informational messages alone.

Read the living analysis →

Violet interpretation

The friend cue may have changed Source, norm perception, identity relevance, and Stake at the moment of action. The effect also demonstrates why a social cue should be modeled relative to the receiver’s relationship network rather than as generic proof.

What it does not prove

This does not show that every campaign should display friends, that the same effect transfers to other elections or platforms, or that social pressure is ethically appropriate in every context.

Published field evidence Healthcare Updated July 23, 2026

Pharmacy vaccine reminders

Why “a vaccine is waiting for you” outperformed expert intuition

A megastudy randomized 689,693 Walmart pharmacy patients across 22 text-reminder conditions and a business-as-usual control. Tested reminders increased influenza vaccination by an average of 2.0 percentage points; the strongest approach used repeated messages saying a shot was waiting for the patient.

Read the living analysis →

Violet interpretation

The winning message may have increased expected relevance, made availability concrete, and shifted the operative frame from abstract recommendation to an already-prepared next step. Repetition may also have improved retrieval at a convenient moment.

What it does not prove

The result belongs to the participating pharmacy population, influenza season, delivery channel, and operational context. It does not establish a universal vaccine message or explain the mechanism by outcome data alone.

Violet applied demonstration Recruiting Updated July 23, 2026

Resume redesigned for a specific role

From a generic career history to a receiver-specific evidence path

This living demonstration shows how the same verified experience can be reordered and clarified for two transitions: rapid recruiter classification and deeper hiring-manager confidence.

Read the living analysis →

Violet interpretation

The intervention should change sequence and accessibility, not invent experience. Spotlite moves role fit and verified outcomes earlier; Strain falls when evidence is grouped around the receiver’s decision; Substance remains bounded by what the candidate actually did.

What it does not prove

This is an applied demonstration, not a reported client outcome or evidence that formatting can overcome discrimination, missing qualifications, labor-market conditions, or a genuine role mismatch.

Published field evidence Recruiting Updated July 23, 2026

Resume screening and racialized names

Why a stronger resume does not enter a neutral receiver system

A field experiment sent fictitious resumes to employers in Boston and Chicago. Resumes assigned White-sounding names received about 50% more callbacks than otherwise comparable resumes assigned African-American-sounding names.

Read the living analysis →

Violet interpretation

Receiver-conditioned interpretation includes harmful and unjust priors. A communication framework must not redescribe discrimination as a candidate messaging failure or imply that the burden belongs entirely to the applicant.

What it does not prove

The study does not identify every mechanism behind employer decisions or establish the size of discrimination in every contemporary market. It clearly warns against attributing all resume outcomes to document quality.

Published multi-site evidence Product adoption Updated July 23, 2026

Online-learning persistence interventions

Why a promising intervention can weaken or reverse across contexts

A large multi-course study tested behavioral interventions intended to support online-learning persistence. Effects varied across courses and student contexts, demonstrating the risk of scaling an average message as if receiver and environment were stable.

Read the living analysis →

Violet interpretation

A cue configuration is conditional on receiver state and context. Scaling changes the information field, population, exposure history, and often the bottleneck itself.

What it does not prove

The results do not mean behavioral interventions never work. They show that context-sensitive effects require theory, measurement, and replication across the environments where a program will operate.

Published field evidence Healthcare Updated July 23, 2026

Clinical preventive-care reminders

When the decision support appears inside the clinician’s active workflow

A randomized controlled trial embedded computerized reminders in hospital order entry. Ordering increased substantially for influenza and pneumococcal vaccination and also improved for other preventive therapies.

Read the living analysis →

Violet interpretation

The reminder arrived when the decision schema and action controls were active. It increased Spotlite and immediate Stake while reducing the Strain of translating guidance into an order.

What it does not prove

The intervention combined several design features and occurred in one hospital system. Increased ordering does not by itself establish appropriate delivery, patient benefit, or transfer to modern clinical systems.

Violet applied demonstration Organizational change Updated July 23, 2026

AI workplace change announcement

How “efficiency” can become a threat signal before the plan is understood

This applied case audits a common announcement: leadership introduces AI to improve efficiency while employees need to know what changes in their work, how performance will be judged, and whether roles are at risk.

Read the living analysis →

Violet interpretation

Repeating benefits before addressing credible risk can increase mistrust. The path should clarify what is decided, what remains uncertain, who is accountable, what changes now, what does not change, and where employees can challenge or shape implementation.

What it does not prove

This is an applied demonstration, not a client result. The appropriate message depends on the actual system, labor context, governance, risks, and decisions already made.

Violet’s standard

Separate the outcome from the mechanism.

A changed attendance rate, click, or share establishes a local outcome under the study conditions. It does not by itself establish attention, trust, comprehension, or benefit. Violet uses those distinctions to design the next measurement.

One Message → Free Audit