Research Frame
CoreTex reflects several recurring retrieval and agent-memory concerns:
| Theme | Design response |
|---|---|
| Dense embeddings can miss answer-bearing evidence | Qwen reranking and graded qrels check evidence beyond surface similarity |
| Long-horizon memory changes over time | Temporal validity and stale-memory rejection are substrate surfaces |
| Multi-hop memory needs structure | Relation and category-routing regions make useful paths explicit |
| Retrieval quality depends on hard negatives | Corpus generation includes plausible wrong documents and near-collision cases |
| Memory systems need compression | The 32 KB substrate creates a fixed state budget |
| Public benchmarks invite overfitting | Hidden packs, canaries, seed reveal, and post-reveal replay separate search from verification |
The design focus is retrieval structure: finding the right memory, at the right time, under a small public state budget.