Guides
Science and technology announcements: a reporting guide
How to turn a science or technology announcement into a verifiable story, from the paper and dataset to funding, corrections and later versions.
What to take away
- Open the paper, dataset, standard or test report before you interview anyone outside the team.
- Write the claim as one measurable sentence, then ask what result would make it false.
- Label statussubmitted, accepted, published, corrected, withdrawn, draft or final.
- Name the stage of readiness. A simulation, a component test and a shipped product are three different stories.
- Track corrections, retractions and revised versions, and change the headline when the record changes.
A release is written to draw attention. A paper is written to defend a narrow claim. Those are different documents with different review, and the gap between them is where most bad science coverage starts.
The story selection checklist is the first filter: if you cannot reach the underlying item, you do not yet have a story.
Build an announcement record
| Field | Question |
|---|---|
| Announcer | University, agency, company, journal, laboratory, or team? |
| Underlying item | Paper, preprint, dataset, standard, patent, prototype, award, or policy? |
| Status | Submitted, accepted, published, corrected, withdrawn, draft, or final? |
| Method | What was observed, built, modeled, or tested? |
| Comparison | Baseline, control, benchmark, prior version, or none? |
| Scope | Which materials, users, settings, or conditions were studied? |
| Uncertainty | Error, interval, sensitivity, failed tests, or unresolved mechanism? |
| Interests | Funding, employment, patent, ownership, or collaboration? |
Fill this in before you call anyone. The release often omits exclusions, negative results and engineering constraints, and it rarely distinguishes a planned milestone from a completed one. Instrument records and decision records raise the same questions, and government decision documents compared puts both against one field list.
Announcement record fields
- Announceruniversity, agency, company, journal?
- Underlying itempaper, dataset, patent?
- Statussubmitted, accepted, published, corrected?
- Methodobserved, built, modeled, tested?
- Comparisonbaseline, control, benchmark, prior?
- Scopematerials, users, settings, conditions?
- Uncertaintyerror, interval, failed tests?
- Interestsfunding, employment, patent, ownership?
Identify who speaks for what
An institutional spokesperson describes the institution's position. An author explains the study. An outside expert assesses the methods. Those three roles are not interchangeable, and a quote is only as strong as the capacity you name with it.
NASA's policy on releasing information to news media routes public information through scientific, technical, programmatic and public-affairs review, and limits who may speak officially on policy, program and budget matters. Attribute accordingly.
Test the central claim
Rewrite the claim as a measurable sentence. Then ask what result would make it false, and go find the sample, the comparator, the outcome, the time window and the analysis that produced it.
For a device, the questions shift. They include:
Device claim test questions
- Test conditions
- Reliability
- Power draw
- Safety
- Scale
- Maintenance
- Independent replication
- What exactly was tested, and which version or build?
- In what environment did it runlab, bench, field, or production?
- How many units ran, and for how long?
- Compared with what baseline, control, or prior version?
- Who operated it, and under what supervision?
- Which failure modes appeared, and which were excluded?
- Which safety, certification, or regulatory gate remains?
A demonstration that works once, in one lab, under supervision, is a data point.
Worked example, with illustrative numbers. Claim as issued: 'Our new battery charges in five minutes.' Measurable rewrite: 'A 1 Ah prototype cell reached 80 percent charge in five minutes at 25 C, over 100 cycles.'
Falsifying result: charge fell to 55 percent by cycle 100. The claim then fails, and the supportable story covers the first cycle only.
Avoid first, proves, solves, revolutionary and ready unless the evidence defines and supports them. Health desks run a parallel typology; the comparison of health evidence and alert types does the same status work for medical claims.
Separate stages of readiness
Stages of readiness
- Simulation is not a physical test
- Component test is not integrated-system test
- Demonstration is not mass production
- Patent is not evidence an invention works
- Grant funds proposed work, not completed finding
Most announcements sit at one stage and imply the next.
| Stage | What it means | What it is not |
|---|---|---|
| Simulation | Model or numerical study only | A measured result |
| Prototype | A working unit tested on a bench or in a lab | A product |
| Pilot | Limited use with real users or conditions | A full service |
| Certified | Passed the named safety or regulatory gate | Proof of performance at scale |
| Shipped | On sale in a stated market | Proven under every condition |
State the stage and the next gate. If timing depends on approvals, manufacturing, funding or further testing, name the dependency and who controls it.
Report uncertainty without fog
Give the estimate and its uncertainty together. Describe the largest practical limitation in plain words. Ask the authors what alternative explanation they tested and what data would change their conclusion. Common failures repeat across daily explainers, and daily explainer problems collects them.
Maintain the record
Track corrections, retractions, updated preprints, and revised datasets. Also track standard drafts, software versions, and mission status. If the underlying record changes the claim, revise the headline and summary, not only a note at the bottom.
Sample fix, using the battery example above. First headline: 'Lab battery charges in five minutes.' Corrected headline: 'Lab cell reaches 80 percent charge in five minutes, first cycle only.' Corrected summary: 'One prototype cell charged to 80 percent in five minutes in a single lab test. Capacity fell over 100 cycles, and no independent test has run.'
The Science journals' editorial policies commit to correcting errors and, where necessary, retracting papers; they distinguish a minor correction from a citable erratum, and both from a retraction that follows when core conclusions are invalidated. Match the article's language to the record's current status, not its first version.
Common questions
Is an institutional release a scientific source?
It is a source for what the institution announced, and nothing more. Use the underlying research record for methods and findings, and say which one you are quoting.
Does peer review prove a result is correct?
No. It is one review stage, and it varies by journal. Study design, data, analysis, replication and later correction still decide whether the finding holds.
Is a prototype a product?
Not necessarily. Report its tested functions, the environment it ran in, the scale it reached, its limits, and the next development stage.
Should funding be included?
Yes. Funding and relevant financial or institutional interests let readers weigh the work's context, and they belong in the same paragraph as the claim.
