See Epistamate in action · Controlled policy-research example

Inside a governed research iteration.
EU–India FTA, Iteration 2.

This is one controlled example of the broader Epistamate research workflow. Follow a policy-research run from decision framing through evidence acquisition, source-bounded analysis, optional deeper coverage of already represented evidence, findings and the next research agenda — with the changing research state kept visible rather than only the final brief.

Start walkthrough →

The research question

What opportunities could the concluded EU–India Free Trade Agreement create for German/EU manufacturers considering India manufacturing expansion in automotive and pharmaceuticals, and what material commercial, regulatory, supply-chain and implementation challenges would remain?

The governing boundary

FTA-attributable effects were kept separate from broader India manufacturing conditions, and negotiated provisions were kept separate from rules not yet legally in force. The researcher retained authority over framing, evidence admission and whether the current state was sufficient to continue.

Controlled product experiment, not a client case study, investment recommendation, policy recommendation or comparative accuracy benchmark. The EU and India concluded FTA negotiations on 27 January 2026; as of this experiment, completion of the parties’ legal procedures still separated the negotiated agreement from entry into force.

Exit walkthrough
Stage 1 · Policy question

Start from the decision, not just the topic.

1 / 9
Epistamate policy research framing for an EU–India FTA manufacturing assessment.
Actual Iteration 2 framing state. The question, intended use, scope and evidence route are explicit before analysis begins.
What Epistamate preserved

The manufacturing decision, 3–5 year horizon, automotive/pharmaceutical scope and the requirement to distinguish FTA effects from broader India conditions.

Human authority

The framing is not inferred from later findings. The researcher owns the question, intended use and boundaries that subsequent evidence must answer.

Stage 2 · External research

Turn the framing into an explicit evidence plan.

2 / 9
Epistamate Iteration 2 policy evidence plan showing the accepted research question, intent, scope and external research state.
Accepted Iteration 2 planning state. External research remains a bounded response to the question and scope rather than an open-ended search for persuasive material.
What happened

The accepted plan carried 10 external targets. At this point 7 had been acquired; the plan itself remained visible as the reference for what the iteration set out to obtain.

What remained open

Plan acceptance did not imply complete acquisition, sufficient evidence or a finding. Those later transitions remained separate.

Stage 3 · Policy evidence

Continue with an incomplete corpus without pretending completeness.

3 / 9
Epistamate policy Evidence Room showing 10 planned targets, 7 acquired, 3 deferred by reviewer and 0 unresolved.
Explicit acquisition state. Seven of ten targets were acquired; three were deliberately deferred by the reviewer; no retrieval remained unresolved.
What Epistamate did

Preserved the acquired material and the reviewer’s decision to stop pursuing three targets, instead of converting “work stopped” into “corpus complete.”

What the researcher decided

Whether those explicit limitations were acceptable for the present iteration, or whether acquisition should resume.

Stage 4 · Source-bounded analysis

Let frontier reasoning propose claims; keep admission separate.

4 / 9
Epistamate policy analysis showing 27 candidates from 7 represented sources with researcher admission required.
Candidate review state. The run produced 27 candidates from 7 represented sources. Lineage was validated, but researcher admission was still required.
Division of responsibility

Frontier supplied candidate semantics; Epistamate kept them source-bounded and used structured research state to route attention.

Authority boundary

A model-generated candidate did not become admitted evidence merely because it was plausible or well written. Admission remained a researcher action.

Stage 5 · Progressive coverage

Inspect more of what you already have — without pretending the first pass was complete.

5 / 9
Controlled run record · same Iteration 2 corpus
Pass 1

67 passages
27 admitted claims

Pass 2

+60 non-overlapping
+16 claims · 43 cumulative

Pass 3

+60 non-overlapping
+18 claims · 61 cumulative

The represented corpus contained 2,168 passages. Epistamate kept the unanalysed remainder explicit and allowed the researcher to request another bounded pass without starting a new research iteration. These counts describe one controlled product experiment, not a completeness or quality score.

Researcher-controlled depth

The first model-facing packet was deliberately smaller than the represented corpus. Additional non-overlapping passes could be requested over evidence already in hand.

What changed

Later passes still produced useful claims, but by Pass 3 more of the value was refinement and deeper support for an increasingly stable research position. A changed analytical basis required fresh synthesis rather than silently retaining the old one.

No completeness claim

Logged passage coverage remained a processing fact. It did not mean the remaining passages were irrelevant or that the research was complete.

Stage 6 · Findings

Keep negotiated opportunity separate from legal effect.

6 / 9
Epistamate provisional policy finding stating that EU–India FTA benefits remained prospective rather than legally operative.
Provisional finding, not automatic conclusion. The synthesis treated negotiated market-access benefits as prospective while the required legal procedures still separated the agreement from entry into force.
Evidence characterization

The interface exposed that this procedural point rested on direct official EU evidence but a single source path, rather than flattening it into a generic confidence number.

Human authority

“Researcher judgement still required” remained visible. Importing a structured synthesis did not silently turn it into an accepted finding.

Stage 7 · Decision readiness

Ask whether the research is usable — and show what was lost or deferred.

7 / 9
Epistamate further-analysis state showing current research usable, 15 agenda items and a first-loss audit.
Usable did not mean complete. The first-loss audit exposed 7/10 acquired targets, 7 sources and 67 passages, 27/27 admitted claims, and a synthesis of 10 findings with 9 gaps.
What Epistamate routed

The accepted synthesis was converted into 15 further-analysis agenda items rather than disappearing into the final prose.

What the researcher still chose

Whether to generate a cited report from the current state or strengthen the research with another iteration.

Stage 8 · Iteration delta

Make the change in research position inspectable.

8 / 9
Epistamate policy research iteration delta showing strengthened, narrowed or qualified, reframed and new findings.
Iteration 2 did not overwrite Iteration 1. The delta recorded 3 strengthened findings, 3 narrowed or qualified, 1 reframed and 3 new findings, alongside 5 persistent/refined gaps and 4 new gaps.
Why this matters

A new report can hide how the underlying research position moved. The delta preserves that evolution as research state rather than asking a later reviewer to infer it from two pieces of prose.

What remained unresolved

Strengthening some findings did not erase open gaps. Persistent and newly discovered uncertainty stayed attached to the iteration.

Stage 9 · Next research focus

Turn residual uncertainty into the smallest useful next research actions.

9 / 9
Epistamate recommended next research focus showing 15 further-investigation items and explicit authority limits.
The next agenda was routed, not declared authoritative. Epistamate identified 15 further-investigation items from accepted synthesis and review state while explicitly stating that this deterministic routing did not create a new finding or evidence authority.
What Epistamate did

Prioritized where additional evidence could most improve the current research position, including finding-specific validation needs.

Human authority

The researcher still decides whether to continue, which validation actions to pursue and whether the expected value of another iteration justifies the work.

Stage 1 of 9

The output was not the only product.

The experiment produced a research position that could be inspected, challenged and continued: evidence acquisition limits remained visible, model-proposed claims stayed separate from researcher admission, the researcher could deliberately inspect more of an already represented corpus, and unresolved gaps became explicit next-research work.

Back to the Policy example →How the evidence model works →