CK-28 Provenance tracking and audit logging (W3C (World Wide Web Consortium) PROV; lineage; record keeping)
Record which source, prompt, model version, settings, reviewer and decision produced each output, and produce a provenance graph or log that traces any claim back to its source.
What it does
Records which source, prompt, model version, settings, reviewer and decision produced each output, as a graph of entities, activities and agents. Artefact: a provenance graph or log. Read the derivation chain to find whether two sources trace to one origin, and to retrace any claim to its source.
When to use it
Use when you must retrace a claim to its source, show how an output was made, or find whether two sources trace to one origin.
When it does not fit
Fits when derivation links can be recorded as they happen. Misfit: provenance is missing or added afterwards. PROV records provenance but does not judge reliability. NIST and the Australian standard require record keeping for tests and results, and give no field list.
Result format
The result is a provenance graph or log that links each output to the source, prompt, model version, settings, reviewer and decision that produced it, as entities, activities and agents. It states whether any two sources trace to one origin and shows the derivation chain from a claim back to its source.
Requirements on the material
| Property | Requirement type | Requirement |
|---|---|---|
| Unit of analysis | Always applies | requires=source; requires=document |
| Appraises source credibility | Applies in some cases | no |
Not restricted on: Structure, Content type, Variable types, Time structure, How the sample was chosen, Level of detail available, Number of sources, Whether sources can be joined, Amount of material, Known data problems, Producer interest, Source dependence, Origin, Document genre, Language, Personal information, Security classification, Licence.
| Minimum material | derivation links recorded as work happens |
|---|---|
| Additional sources needed | none |
| Not for (rules) | none |
| Not for (text) | Wrong choice when provenance is missing or added afterwards. A provenance record does not judge reliability, and no source gives a field list. |
Choosing and ordering
| Kind of technique | Preparation technique, runs in step 5.2 |
|---|---|
| Phase tag | gather |
| Other phase tags | verify and prepare; analyse; validate; report |
| Kind of finding | risk |
| Question answered | who |
| Strongest statement it supports | describe |
| Needs as input |
|
| Makes as output |
|
| Must exist first | None |
| Loop target | proceed with limit stated |
Library checks by stakes
Each stakes level includes the checks of the level below it. Checks that need a person apply at high stakes only, so the process leaves them out at lower stakes (decision 38). Repeatability: deterministic.
| Stakes | Check | Person involved | Counts as a different method |
|---|---|---|---|
| exploratory | input validation | none | no |
| decision-informing | input validation | none | no |
| decision-informing | spot check | optional | no |
| decision-informing | provenance log | none | no |
| high-stakes | input validation | none | no |
| high-stakes | spot check | optional | no |
| high-stakes | provenance log | none | no |
| high-stakes | independent review | needed | no |
Failure modes and limits
| Model suitability | model plus code |
|---|---|
| Typical model failure | Infers a derivation link from similar wording that was never recorded, writes a plausible reconstructed account of its steps instead of logging real ones, or treats same-origin sources as independent. |
| Error sources | input data; data transformation |
| Certainty limits | Certainty drops when links were reconstructed afterwards (risk of bias) and when recorded steps are incomplete (indirectness). |
Related techniques
- pairs with IS-06 Source dependence and copy detection (source independence analysis): derivation chains show whether sources share an origin
- pairs with CK-29 Scripted reproducible workflow (reproducible analysis; version control; literate programming): a scripted workflow records code and environment alongside provenance
Evidence and tag basis
Documented. PROV describes provenance as information about the entities, activities and people involved in producing data, used to judge quality and reliability. (W3C PROV overview; W3C PROV-DM; W3C PROV-O; NIST AI 600-1; Voluntary AI Safety Standard, 10 guardrails)
Tag basis: input profile inference; selection inference; sequencing inference; robustness inference. All tag values are inference unless a source is named.