Subfield Agora

Every piece of generated content on Noosaga — timelines, concept maps, articles, quizzes — will eventually drift from the field it's supposed to represent. Subfield Agora exists to catch that drift before it becomes a problem.

Rather than treat content as finished the moment it's created, Agora runs periodic review sessions that surface concrete issues and track them in discussion threads. You can find these threads in the Agora section inside any subfield.

What gets reviewed

Review sessions look for the kinds of problems that tend to creep in over time:

• Two articles covering essentially the same ground
• Missing nodes or edges in a concept map
• Learning paths that don't flow well from one topic to the next
• Quiz questions that are vague, repetitive, or testing the wrong thing
• Topics that got way more (or way less) depth than they deserve

When you have a framework selected, you can toggle between two views:

• Current framework — threads targeting that framework, plus subfield-wide threads.
• Entire subfield — every thread across all frameworks in the subfield.

Who can do what

Everyone can read threads and see the full discussion history — nothing is hidden behind a login wall.

If you're logged in, you can jump into any thread and reply. Triggering new review sessions and approving or rejecting proposed changes is reserved for admins. When there's a pending action and you're not an admin, you'll see it labeled "Pending admin approval" so it's clear something is waiting.

How a review session runs

Clicking Trigger Review starts a workflow with two stages:

1. A subfield-wide pass runs first. This includes all framework-level content for the subfield (framework articles, framework relations, concept graph structure, quiz signals), but skips full concept article bodies to control context size.
2. Then framework-specific passes run for each review-ready framework. A framework is review-ready when it has at least 5 pedagogical concept articles. These passes include full concept article context for that framework.

If no framework is review-ready, the subfield-wide pass still runs and the workflow completes normally.

The reviewer identifies the most impactful issues, then either opens fresh threads or adds findings to existing unresolved ones (tagged open or consensus). For each issue, a concrete fix gets proposed — not a hand-wavy suggestion but an actual structured action.

A second pass checks whether the proposal is sound. If it holds up, the action gets created with a pending status. From there, a human has to explicitly approve it before anything changes. Consensus alone never triggers execution.

Thread and action lifecycle

Threads move through a few states as work progresses:

• open — the issue has been identified and people are still discussing it.
• consensus — a proposed action is ready and waiting for someone to approve it.
• resolved — the action was executed and the thread is closed.

Actions themselves have their own status track: pending → approved → executed. They can also land on rejected (someone decided the fix wasn't right) or failed (something went wrong during execution).

For framework-gap threads (for example, a framework with missing foundational content), Agora can propose a run_framework_workflow action. Approving it queues the full framework pipeline (verify, relations, framework article, concept map, timeline) for that single framework.

"Truncation" reports that aren't actually bugs

You might occasionally see a thread flag an article as truncated when the article itself is actually fine. This happens because Agora excerpts article text in its review prompts to stay within token budgets.

When text gets excerpted, you'll see a marker like [excerpt ... prompt_trimmed=true] in the thread. That's the prompt being trimmed, not the article being broken. If you want to verify, just check the source article directly — if the full text is there, the truncation report is a false positive.

When a session finishes without new threads

This trips people up sometimes. You trigger a session, it runs to completion, and... nothing new appears. That doesn't mean it failed. A few things could be going on:

• The reviewer genuinely didn't find any high-confidence new issues worth raising.
• Issues it found matched existing unresolved threads and got merged in via fingerprint deduplication (so the existing threads got updated instead).
• The session hit its token or post budget before it could open new threads.

Check timestamps and thread statuses to see what actually happened during the run.

Getting the most out of Agora

Think of it as a maintenance loop rather than a one-off audit. Trigger reviews on the subfields you care about most, and do it regularly — especially after you've made significant edits. When threads reach consensus, that's your cue to look at the proposed action and either approve or reject it. Running another session after you've approved a batch of changes is a good way to confirm things actually improved.