Tines The single, secure environment for agents, apps, and automations.
Promoted
Industry Research Framework
Source-backed research protocol for AI agents
8 followers
Source-backed research protocol for AI agents
8 followers
Open-source protocol for AI agents doing longform industry research: brief gates, state files, source/claim discipline, staged drafting, review loops, hard stops, gotchas, and offline evals with regression fixtures.
How does the protocol handle conflicting sources when claims can't be cleanly reconciled, and is there a built-in way to surface that uncertainty back to the user or does each agent have to roll its own approach?
@diyar113588 Great question. The protocol treats conflicting sources as a first-class research state, not something the agent should quietly smooth over.
It separates sources, claims, and uncertainty. When sources disagree, the agent should record the conflict, mark confidence/unknowns, and either resolve it with stronger evidence or surface the disagreement clearly in the draft. The hard rule is: don’t turn conflicting evidence into a clean narrative unless the evidence actually supports that reconciliation.
The hard stops and gotchas sections show real respect for the reader's time. Most frameworks gloss over the messy parts, so seeing them laid out plainly with offline evals and regression fixtures feels like the work of people who've actually been burned in production.
@ymet_k37866 Thank you. That was exactly the motivation. In long agent research, the messy parts are usually where the work fails, so I wanted hard stops, gotchas, evals, and regression fixtures to be part of the public framework, not just private operator notes.
Love that this is open source and actually cares about source/claim discipline, not just vibes. The hard stops between stages feel like the kind of structure I always end up hacking together myself.
@ceylanupy3 Thanks Ceylan. I had the same experience: rebuilding ad hoc gates for every serious research task. The goal here is to make source/claim discipline and stage gates reusable across agents while keeping the protocol lightweight enough to drop into Codex, Claude, Gemini, Cursor, and similar tools.
the brief gates and gotchas sections are a really thoughtful touch, shows the team has actually felt the pain of agent research going off the rails
@brahimbedk4vop Thanks İbrahim. Yes, the brief gates and gotchas came directly from agent research runs that went off track. I’m trying to turn those failure patterns into explicit contracts so the agent has to slow down before scope, evidence, or depth collapses.