Update app.py

app.py

@@ -39,19 +39,15 @@ MAX_HISTORY = 6  # turns to keep in context
 DEMO_TEXT = """The ConjunctionReservoir is a document retrieval system that asks not
 "do these query terms appear somewhere in this chunk?" but rather
 "do these query terms appear in the SAME SENTENCE?"
-
 This is grounded in auditory neuroscience. Norman-Haignere et al. (2025)
 showed that auditory cortex integration windows are time-yoked at approximately
 80ms — they are fixed clocks, not expanding to cover arbitrary structure.
 The sentence is the text analog of this fixed window.
-
 NMDA receptors implement coincidence detection by requiring simultaneous
 presynaptic glutamate release and postsynaptic depolarization to open.
 This is a hard AND gate, not a weighted average.
-
 The conjunction_threshold parameter mirrors this: below the threshold,
 a sentence contributes zero score to the chunk — it is absent, not degraded.
-
 Benchmark results show ConjunctionReservoir achieves 100% Rank-1 Rate on
 conjunction-specific queries, compared to 60% for both BM25 and SweepBrain.
 It intentionally trades broad-query recall for precision on specific
@@ -325,6 +321,7 @@ Grounded in auditory neuroscience (Norman-Haignere 2025, Vollan 2025). Zero embe
                     label="",
                     height=480,
                     show_label=False,
+                    type="tuples",
                 )
 
                 retrieval_info = gr.Markdown("", elem_id="retrieval-info")