Responsible research
Research integrity is a core requirement of this site, not an afterthought. This page sets out the principles we hold ourselves to and how each one shows up in the work.
Principles
- 01
Evidence before claims
We publish a claim only when the evidence supports it, and we state plainly what that evidence is.
- 02
Reproducibility where practical
Where practical we share the setup, data and code needed to reproduce a result.
- 03
Honest limitations
Every research page states what it does not establish, as prominently as what it does.
- 04
Clear provenance
Results are traceable to their inputs, methods, versions and the people responsible.
- 05
Community participation
Work that touches a community is done with that community's participation and consent — not about them without them.
- 06
Privacy by default
We minimise, redact and contain sensitive data; the default is to collect nothing we do not need.
- 07
Responsible release
We release code, weights and data only when the licence, safety and consent conditions are met.
- 08
Transparent conflicts
Commercial and other interests are disclosed on the page where they are relevant.
- 09
Human accountability
A person, not a model, is accountable for every published decision.
- 10
Correction over concealment
When we are wrong we correct the record openly rather than quietly editing it away.
How we practise it
Research integrity & reproducibility
Publications are versioned and carry documented methodology, contribution statements (CRediT) and canonical citation metadata. We do not fabricate authorship, results, citations, peer review or ethics approval. Where practical, evaluation setups and artefacts are shared so a result can be interpreted and reproduced.
Research integrity policyEthical research & community participation
Research that involves a community is shaped by that community's consent and governance. Our Auslan work is explicitly community-informed: any dataset or model would require informed, revocable consent and Deaf-community governance before it exists. We collect no human-subjects data through this website.
Auslan research & ethicsPrivacy & data governance
We use no advertising or tracking cookies and no invasive analytics by default. Datasets carry explicit ownership, licence and access conditions; third-party data is used, not owned; community-governed data is subject to consent and takedown. We publish no personal, health or sign-language video data.
Data governanceLicensing & intellectual property
We display the exact licence for every released artefact and mark ‘Licence not yet confirmed’ where unresolved. Public access is not a public-domain dedication; a dataset is not ‘open’ when it has restrictions; non-commercial licences prohibit commercial use.
Licensing & IPConflicts of interest & funding
Octopus Research Institute receives infrastructure, engineering and funding from Octopus Core. Where a programme, publication or dataset touches an area in which Octopus Core has a commercial interest, that interest is disclosed on the relevant page. Findings, including negative results, are not altered to suit a funder.
Independence & fundingResponsible publication & release decisions
Preprints and non-peer-reviewed material are labelled as such. Model weights are not released unless a licence permits distribution; dataset downloads are not published without confirmed licensing. Metrics appear only where verified, with dataset and version context.
Model cardsVulnerability disclosure & research misuse
Report a vulnerability to the security contact and allow reasonable time for remediation before public disclosure. We weigh the misuse potential of any release and withhold detail where the risk outweighs the benefit.
Security & disclosureCorrections & retractions
We record corrections and retractions rather than concealing them. Publications are versioned and their version history is shown. If you believe something here is inaccurate, tell us and we will correct it and, where appropriate, publish a correction notice.
Corrections & retractionsWhere this shows up in the work
These are not abstract values — they are encoded in how this site is built. Research statuses, disclosures, limitations and claim boundaries are first-class, and several are required by the content model rather than left optional. See the Responsible AI research area for the programmes behind them.
Responsible AI research