Physics · Article

Scientific records, data management, and open science

Build an inspectable physics record, test and revise a bounded cart model, preserve processing history, and plan findable data with appropriate sharing conditions.

Previous article: Laboratory safety, reproducibility, and replication

Back to Physics

After reading, you can

Distinguish original entries, derived values, model predictions, interpretation and metadata.
State a physical system, reference frame, units, domain and uncertainty assumptions for a supplied record.
Read a position–time graph and compare a model prediction with explicitly bounded position information.
Revise a candidate model without rewriting the supplied observations or claiming that a successful fit proves it true.
Describe provenance, processing versions, backup, recovery and preservation in a small data-management plan.
Use the FAIR principles and open-science guidance without equating findability with unrestricted access or a persistent identifier with correctness.

Record the question and what was done

The following checkpoint is an original fictional teaching packet, CART-RECORD-v1. No apparatus was operated and no measurement was collected for this lesson. Its measurement-style values and bounds are supplied assumptions for reasoning, not validated instrument performance.

A scientific record should let another person inspect what question was asked, what was done and how the conclusion follows. Keep original entries separate from calculations, model predictions and interpretation. Clear documentation makes a claim checkable; it does not guarantee the claim is true.

Metadata are information describing a record, such as its creator, purpose, variables, units and access conditions. Provenance records its origin and the changes or processing that led to a particular result. A column of numbers without that context can answer a different question from the one you intended.

Define the cart record and its bounds

Choose the cart as the system and follow one marked point along a straight track. Use a fixed track origin, with positive x in the forward direction. Time t is measured from the stated start. The record gives reported positions r at exact time labels 0 s, 1 s, 2 s and 3 s. Millimetres, mm, and metres, m, represent the same length quantity: 1000 mm = 1 m.

For this fictional case, supply the following bounded uncertainty information: the target position at each listed time is assumed to lie in the inclusive interval [r − 0.020 m, r + 0.020 m]. These stipulated admissible-position bounds are not assigned a standard-uncertainty or confidence-interval interpretation, and no probability distribution is supplied. Time-label uncertainty is set aside as an explicit teaching assumption.

The table preserves the supplied values and their unit conversion. A reported centre is not declared to be the exact physical position. The intervals describe allowed values under our premises; they do not reveal which value occurred. The record alone also does not describe the continuous motion between the four times.

CART-RECORD-v1: fictional reported positions and stipulated inclusive bounds
Time label t	Supplied position r	Converted position r	Admissible position interval
0 s	200 mm	0.200 m	[0.180 m, 0.220 m]
1 s	500 mm	0.500 m	[0.480 m, 0.520 m]
2 s	800 mm	0.800 m	[0.780 m, 0.820 m]
3 s	1160 mm	1.160 m	[1.140 m, 1.180 m]

State a model and read its graph

Start with candidate M1: the marked point moves only forward along the straight track at constant speed, with x_M1(t) = 0.200 m + (0.300 m/s)t for every 0 ≤ t ≤ 3 s. We choose the nominal change between the first and third reported centres, (0.800 m − 0.200 m) / 2 s = 0.300 m/s, as the starting rate. The candidate's parameters are fixed exactly for testing; this calculation does not estimate their real measurement uncertainty.

At t = 3 s, M1 predicts 0.200 m + (0.300 m/s)(3 s) = 1.100 m. Multiplying rate by time gives a length, and adding the initial position gives another position. This is a prediction of the stated model, distinct from the record's reported value 1.160 m.

In the graph, horizontal coordinates give time in s and vertical coordinates give position in m. The height of a model line at a given time is its predicted position. From 0 s to 2 s, M1 rises by 0.600 m over 2 s, so its gradient is 0.300 m/s. Its line is drawn throughout the declared interval because constant motion is stipulated; four record points alone would not justify joining them into that rule.

The plotted record points are kept separate, with vertical segments for the stated position bounds. Those segments are not motion paths or probability error bars. Neither candidate line is extended outside 0–3 s: a formula's ability to return a number at 4 s does not extend our declared physical model.

A prediction, a bounded record and a revised candidate

M1: initial candidate
M2: revised candidate
Fictional reported positions
Stipulated inclusive position bounds

1. Inspect the mismatch
At 3 s, M1 predicts 1.100 m while the stipulated interval is [1.140 m, 1.180 m]. Correct arithmetic does not make this candidate compatible.
2. Revise the candidate, retain the record
M2 predicts 0.180 m, 0.500 m, 0.820 m and 1.140 m, all within the inclusive intervals. Keep both model versions and the original reported values.
3. Trace the conversion
Divide the numerical millimetre value 1160 by 1000 to obtain 1.160 m. Retain the wrong-divisor output as an identified earlier derivative and document the correction.
4. Share an identifiable version
Provide permitted data, methods, units, bounds and provenance with clear usage conditions. A persistent identifier identifies the record; it does not validate the motion model.

Original fictional cart checkpoint, 0–3 s. Separate reported points and their stipulated inclusive position bounds are compared with two fixed candidate lines. The bounds have no assigned probability or standard-uncertainty meaning. Agreement makes M2 compatible with this record under the premises; it does not establish the true motion.

Compare with bounds and revise a candidate

At 0 s, 1 s and 2 s, M1 predicts the reported centres. At 3 s its prediction 1.100 m is below the allowed interval [1.140 m, 1.180 m]. The difference from the reported centre is 0.060 m, larger than the stipulated bound 0.020 m. Therefore this exact candidate and the supplied bounds are incompatible at that time, if the record, units, time labels and bound assumptions are retained.

This does not establish the cause. The motion model, position or timing interpretation, acquisition process or claimed bounds might need investigation in a real case. Preserve the supplied record and change the conclusion: M1 has not passed this check. New measurements would require an appropriate approved measurement plan; altering the recorded value to make M1 agree is not a revision of the model.

A different original candidate, M2, has x_M2(t) = 0.180 m + (0.320 m/s)t on the same inclusive 0–3 s domain. Its predictions are 0.180 m, 0.500 m, 0.820 m and 1.140 m. All four lie in the stated intervals, including boundary values. This is a justified revision to a candidate compatible with this bounded record; it does not identify the true trajectory or validate the assumed bounds.

Changing only M1's rate to 0.320 m/s while keeping its initial position at 0.200 m would predict 0.840 m at 2 s, outside [0.780 m, 0.820 m]. Check every retained observation and assumption, rather than repairing only the final point. M1's failure does not rule out all constant-speed models, and M2's success does not prove constant speed between observations.

Preserve processing history and provenance

Retain the supplied input, the documented conversion and calculations, and the candidate-model version used for a conclusion. For example, CART-PROCESS-v1 accidentally divides the millimetre entries by 100 and labels the results as m. CART-PROCESS-v2 corrects the divisor to 1000. Keep the first output as an identified erroneous derivative, retain the original input unchanged, and record the correction and its reason.

For the last entry, 1160 mm × (1 m / 1000 mm) = 1.160 m. Using 100 as the divisor of the numerical millimetre value and labelling the output as m would instead produce the falsely labelled value 11.600 m. A unit symbol alone does not check the numerical conversion. Repairing this processing error does not turn M1's 1.100 m prediction into agreement with the corrected 1.160 m report and its bounds.

A provenance link should identify which input and processing version produced an output, when the change was recorded and why. If an output changes, preserve that history instead of presenting the current file as if it had always contained the new result.

Plan storage, recovery and preservation

A data-management plan assigns decisions and responsibility before a project ends. The Digital Curation Centre's checklist asks about data, documentation, formats, storage, backup, recovery, sharing and responsibilities. For our small packet, write who maintains the supplied input, derived tables, model definitions and explanation, and where an authorized reader can find them.

Specify a working copy and a recoverable backup, who maintains them, and how a restoration will be checked. Preserving only the latest converted table would lose the supplied millimetre entries and the history of the wrong divisor. Preserve the documentation needed to read the data too; a copied file without its units can preserve bytes while losing the intended meaning.

Plan an understandable file format and its field definitions, required access protection and the retention decision appropriate to the institution and project. Backup, long-term preservation and publication are different tasks. Neither a file on a laptop nor a public link alone is a complete recovery or preservation plan; this lesson sets no universal storage duration or legal rule.

Make reuse possible through FAIR principles

FAIR means Findable, Accessible, Interoperable and Reusable. The original principles address both human and computational reuse, including identifiers, descriptive metadata, access protocols, usage licences and provenance. The table gives a small-packet application, rather than claiming that our example has passed a formal FAIR assessment.

Accessible does not mean every file must be unrestricted. FAIR explicitly permits authentication and authorization where needed, and calls for metadata to remain accessible when data are no longer available. A record can explain how access is requested without disclosing protected contents.

UNESCO's open-science recommendation encompasses sharing scientific knowledge, tools and processes with justified access restrictions. Openness requires consideration of permissions and the people or communities affected. Making an unexplained file public is different from enabling responsible scrutiny and reuse.

Apply the four FAIR concerns to the fictional packet
Concern	Question for this packet
Findable	Can a reader identify and locate the exact packet?
Accessible	Does its record explain the access method and any required authorization?
Interoperable	Are fields, units and formats described for use with other records?
Reusable	Are provenance, intended scope and usage conditions clear?

Worked examples

Predict and check the cart record

Using candidate M1, predict the marked point's position at 3 s and compare it with CART-RECORD-v1's supplied position bounds.

State the candidate and domain: x_M1(t) = 0.200 m + (0.300 m/s)t for 0 ≤ t ≤ 3 s. Both parameters are exact candidate choices, not independently known physical values.
At 3 s, the displacement term is (0.300 m/s)(3 s) = 0.900 m, so the predicted position is 1.100 m.
Convert the reported entry: 1160 mm = 1.160 m. Its assumed interval is [1.160 m − 0.020 m, 1.160 m + 0.020 m] = [1.140 m, 1.180 m].
The prediction lies 0.040 m below the lower endpoint. Its difference from the reported centre is 0.060 m. The candidate is outside the supplied bounds, although the arithmetic is correct.
Record that incompatibility and the premises checked. It identifies no cause and is not a statistical significance test.

M1 predicts 1.100 m, outside the assumed [1.140 m, 1.180 m] interval. Revise the claim about this candidate while preserving the original record and bound assumptions.

Revise the model without rewriting the data

Does candidate M2, x_M2(t) = 0.180 m + (0.320 m/s)t on 0–3 s, meet all four supplied inclusive position intervals? Does that establish the true motion?

At 0 s, M2 predicts 0.180 m, exactly the lower endpoint of [0.180 m, 0.220 m]. The interval is inclusive, so this is allowed.
At 1 s, M2 predicts 0.180 m + 0.320 m = 0.500 m, within [0.480 m, 0.520 m].
At 2 s, M2 predicts 0.180 m + 0.640 m = 0.820 m, exactly the included upper endpoint of [0.780 m, 0.820 m].
At 3 s, M2 predicts 0.180 m + 0.960 m = 1.140 m, exactly the included lower endpoint of [1.140 m, 1.180 m].
Retain M1 as the earlier failed candidate and record M2's parameters, domain and successful bounded check. Do not replace the reported centres with the predictions or claim that unobserved motion must follow the line.

M2 meets all four stipulated intervals. It is a compatible candidate under those premises, not proof of the true trajectory, correct acquisition or valid bounds.

Correct a derived table and keep its history

CART-PROCESS-v1 divides the last 1160 mm entry by 100 and labels it as m. How should the conversion and record history be corrected?

The stated unit relationship is 1000 mm = 1 m. Divide the numerical millimetre value by 1000, giving 1.160 m.
The old output 11.600 m differs from the correct conversion by 10.440 m. The original 1160 mm entry is not changed.
Create the identified corrected derivative CART-PROCESS-v2, record the divisor change from 100 to 1000 and its reason, and retain v1 as an erroneous prior output.
Recompute any conclusion that used the old output. M1 still predicts 1.100 m at 3 s, outside the corrected report's stated interval; M2 meets the interval at its lower endpoint.
Keep the original input, both identified processing versions and the revised conclusion linked. This is a processing correction, not a new physical measurement.

Use 1.160 m and preserve the original 1160 mm plus the old and corrected derived outputs. Correct processing and scientific model checks remain distinct.

Try it yourself

A file contains only the values 200, 500, 800 and 1160. What context is needed before interpreting them as this cart record?
Show the answer and reasoning
Identify the packet and its fictional status, the marked point and fixed axis, millimetre units, time labels and reference, supplied bounds and their meaning, and the origin and processing history.
Numbers alone do not name a quantity, distinguish an entry from a prediction or define uncertainty. Metadata and provenance make that interpretation inspectable.
What does M1 predict at 2.5 s? Does the same declared model authorize a prediction at 4 s?
Show the answer and reasoning
At 2.5 s, M1 predicts 0.950 m. The declared 0–3 s model supplies no physical prediction at 4 s.
0.200 m + (0.300 m/s)(2.5 s) = 0.200 m + 0.750 m = 0.950 m. The algebra can be evaluated beyond the interval, but using it there would add a model assumption not provided here.
In a separate fictional comparison at 3 s, the reported position is 1.120 m with the same stipulated ±0.020 m admissible bound. Is M1's 1.100 m prediction allowed?
Show the answer and reasoning
Yes: the allowed interval is [1.100 m, 1.140 m], and M1 is exactly its included lower endpoint.
The endpoints come from subtracting and adding 0.020 m. Compatibility at one supplied time proves no trajectory or confidence level, and this separate input does not overwrite CART-RECORD-v1.
At 1.5 s, what do M1 and M2 predict, and how far apart are their predictions?
Show the answer and reasoning
M1 predicts 0.650 m and M2 predicts 0.660 m; their difference is 0.010 m.
M1 gives 0.200 m + 0.450 m. M2 gives 0.180 m + 0.480 m. These are two exact candidate outputs within their domains, not two measurements or a measurement error.
An analyst replaces 11.600 m with 1.160 m but deletes the original millimetre file and records no reason for the change. What should the retained packet contain?
Show the answer and reasoning
Keep the original 1160 mm entry, the identified erroneous and corrected derivatives, the conversion rule, a recorded correction and its reason, and revised conclusions linked to their inputs.
The corrected value is necessary but is not the entire record. Without the original input and history, a reader cannot inspect whether the change was a unit correction, a new measurement or an unexplained alteration.
A researcher says that FAIR requires every protected file to be public and that a repository identifier proves the cart model true. Are those conclusions justified?
Show the answer and reasoning
No. FAIR allows authorized access when needed, and a persistent identifier identifies a record for citation rather than proving its contents true.
Review permitted access and usage terms, describe the record and provenance, cite the exact version and retain the model's conditions and limitations. Findability, permissions, preservation and scientific validity are distinct checks.

What to carry forward

Retain original entries separately from derived values, predictions and interpretations.
Specify the system, tracked point, axes, units, time reference, domain and meaning of uncertainty information.
Compare predictions under explicit assumptions; correct arithmetic and a compatible candidate do not establish physical truth.
Preserve correction history and link every output to its input, method and model version.
Plan responsibilities, metadata, access, storage, recovery and preservation before the project ends.
FAIR and open science support inspectable reuse with appropriate permissions; cite the exact record version and state what remains unverified.

References

Sources for the concepts and teaching approaches in this article.