01 · Research state
What is the governed position now?
Current accepted findings, unresolved dependencies, evidence posture, pending attention and the clearest place to resume. Useful for internal checkpoints and ongoing research management.
A professional work product may later need to survive a handover, senior review, release decision, peer challenge, institutional inquiry or audit. Reconstructing that history from a finished document or chat transcript is a weak control.
Epistamate preserves evidence identity, model-assisted work, unresolved state, human decisions and change as governed state. From that record it can produce different assurance views for different questions: what the position is now, how it got there, what still prevents reliance, and how AI was used.
The same governed state can support different consumers without changing who owns the substantive judgement.
Boundary: Epistamate does not decide whether an organisation complies with an external code, regulation, institutional policy or professional standard. It can preserve evidence about the recorded workflow that those reviewers may use.
These are deterministic or governed projections of recorded work. The exhaustive ledger remains available when someone needs proof; normal users do not have to read the whole log first.
Current accepted findings, unresolved dependencies, evidence posture, pending attention and the clearest place to resume. Useful for internal checkpoints and ongoing research management.
A bounded handover narrative over the recorded research history: iterations, material changes, major decisions and the progression to the current state. It is a zero-authority recap, not a mechanism for rewriting accepted findings.
An attention-first audit over the recorded governance state: required blockers, recommended follow-up, not-established items and passed controls. It is designed as a release/checkpoint aid, not an external compliance certificate.
Method and iteration history, recorded model-assisted workflow stages, system/model identity where recorded, pack/prompt/result fingerprints and separate researcher acceptance posture. Concise and full exports can serve different review needs.
These visuals come from the same validated three-iteration EU–India FTA research topic. They show four different review questions over one governed record rather than four disconnected reports.
Controlled product-validation case. Screenshots are evidence of observed product state, not a compliance certificate, client outcome or comparative accuracy claim.
Controlled FTA founder validation. In the three-iteration EU–India FTA research case, the disclosure recorded 20 external AI/model-assisted interactions across planning, source location, analysis, synthesis, deliverable drafting and history recap, alongside 17 separate researcher acceptance events. Sixteen were explicit fast-path decisions to advance without detailed review; one was a detailed-review acceptance. The artifact also reported that the research/evidence cutoff was not recorded and that AI use outside Epistamate could not be established.
The references below explain why transparent AI use, human responsibility, methodological traceability and careful treatment of model interactions increasingly matter. They are context for the product, not a claim that Epistamate certifies conformity with them.
The European Commission's 8 May 2026 update describes accountability, transparency and responsibility as research-integrity principles adapted to current GenAI use. It also highlights organisational awareness of hidden prompts—AI instructions hidden from human oversight.
ALLEA describes the Code as a framework for self-regulation across scientific and scholarly disciplines and research settings. The 2023 edition addresses evolving research practices; ALLEA states that the European Commission recognises it as the reference document for research integrity for EU-funded research projects.
| Review question / expectation | What Epistamate can preserve |
|---|---|
| Was AI/model assistance disclosed transparently? | Methods & AI-use Disclosure, recorded systems/models where available, workflow role counts and lineage fingerprints. |
| Can human responsibility be distinguished from model activity? | Separate researcher acceptance/review records; importing a model result does not itself create research authority. |
| Can the method and its evolution be reconstructed? | Research profile, method basis, iteration history, evidence strategy, coverage mode and recorded cutoff where available. |
| Were limitations and unresolved matters preserved? | Current research state, explicit gaps/dependencies, sufficiency decisions and Governance Audit attention. |
| Can a later reviewer understand why the present position exists? | Research History Recap plus underlying lineage and iteration delta. |
| Can source-borne AI instruction risk be surfaced for review? | Source-content trust-boundary signals on selected model-facing passages, while preserving original evidence unchanged. |
This mapping is illustrative. The cited frameworks have their own scope, language and requirements. Institutions, funders, auditors and regulators determine what evidence they require.
The 2026 ERA Living Guidelines specifically highlight hidden prompts as an emerging concern. Epistamate's current analysis path therefore carries a simple authority rule: source content is evidence data, never instructions.
Selected model-facing passages can be scanned non-destructively for signals such as zero-width Unicode, bidirectional controls, Unicode tag characters, instruction-like source text and selected hidden-markup indicators.
Original evidence remains unchanged. A flag is not proof of malicious intent, and absence of a flag is not proof of safety. The current scanner does not claim full visual PDF-layer detection or prompt-injection immunity.
Methods sections, AI-use disclosure, peer review, research-integrity review, handover and reproducibility-oriented context.
Checkpoint reviews, evidence provenance, unresolved implementation questions and a defensible trail behind policy analysis.
Senior review, recommendation challenge, inherited engagements, evidence-sensitive issue lists and decision records.
Pre-release validation and evidence packages for risk, legal, compliance, audit or other second-line review without making those functions subordinate to the model.