CogCanvas: Verbatim-Grounded Artifact Extraction for Long LLM Conversations
Paper
•
2601.00821
•
Published
conversation_id
stringlengths 8
8
| question_id
stringlengths 8
8
| question
stringclasses 30
values | answer
stringclasses 1
value | question_type
stringclasses 1
value | keywords
stringclasses 1
value |
|---|---|---|---|---|---|
eec25369
|
31ec57c3
|
Why did we choose use existing templates for approach?
|
[]
|
||
eec25369
|
82169ade
|
Why did we choose AWS for hosting?
|
[]
|
||
eec25369
|
a6197368
|
What decisions were affected by our deadline (in 2 weeks)?
|
[]
|
||
eec25369
|
9a2630bd
|
What decisions were affected by our budget ($2,000)?
|
[]
|
||
c7a73fd4
|
69006c9f
|
Why did we choose AWS for hosting?
|
[]
|
||
c7a73fd4
|
221ef9d0
|
Why did we choose go with MVP only for approach?
|
[]
|
||
c7a73fd4
|
421f555d
|
What decisions were affected by our budget ($10,000)?
|
[]
|
||
c7a73fd4
|
6488185c
|
What decisions were affected by our deadline (in 2 weeks)?
|
[]
|
||
69187df4
|
36f6d61e
|
Why did we choose Flask for framework?
|
[]
|
||
69187df4
|
580ec0ba
|
Why did we choose use existing templates for approach?
|
[]
|
||
69187df4
|
40bb8ab8
|
What decisions were affected by our team_size (2 developers)?
|
[]
|
||
69187df4
|
ce1b14b3
|
What decisions were affected by our deadline (in 2 weeks)?
|
[]
|
||
7401f420
|
450e060c
|
Why did we choose Spring Boot for framework?
|
[]
|
||
7401f420
|
8977bea3
|
What decisions were affected by our team_size (15 people)?
|
[]
|
||
fd955595
|
f055eb65
|
Why did we choose AWS for hosting?
|
[]
|
||
fd955595
|
9dda6482
|
Why did we choose Django for framework?
|
[]
|
||
fd955595
|
5ea3638b
|
What decisions were affected by our budget ($50,000)?
|
[]
|
||
fd955595
|
c7930311
|
What decisions were affected by our team_size (5 engineers)?
|
[]
|
||
24f7d1f0
|
8f346924
|
Why did we choose FastAPI for framework?
|
[]
|
||
24f7d1f0
|
624a6d8f
|
Why did we choose follow best practices for approach?
|
[]
|
||
24f7d1f0
|
ff03d1ad
|
What decisions were affected by our team_size (2 developers)?
|
[]
|
||
24f7d1f0
|
dcd0c46c
|
What decisions were affected by our deadline (end of Q2)?
|
[]
|
||
cfcf96a8
|
6ef6b26a
|
Why did we choose build properly with tests for approach?
|
[]
|
||
cfcf96a8
|
d35313ec
|
Why did we choose FastAPI for framework?
|
[]
|
||
cfcf96a8
|
5e62154e
|
What decisions were affected by our deadline (end of Q2)?
|
[]
|
||
cfcf96a8
|
bc46bff9
|
What decisions were affected by our team_size (5 engineers)?
|
[]
|
||
f1e2b5b1
|
e6d091ac
|
Why did we choose Azure for hosting?
|
[]
|
||
f1e2b5b1
|
c5807bca
|
Why did we choose use existing templates for approach?
|
[]
|
||
f1e2b5b1
|
2591a64b
|
What decisions were affected by our budget ($10,000)?
|
[]
|
||
f1e2b5b1
|
8cb88741
|
What decisions were affected by our deadline (in 2 weeks)?
|
[]
|
||
606c86de
|
4610e5d3
|
Why did we choose Spring Boot for framework?
|
[]
|
||
606c86de
|
ea85009b
|
Why did we choose Google Cloud for hosting?
|
[]
|
||
606c86de
|
7271c2b7
|
What decisions were affected by our team_size (a team of 10)?
|
[]
|
||
606c86de
|
8f1b3b34
|
What decisions were affected by our budget ($50,000)?
|
[]
|
||
ec1995c3
|
d892a9fe
|
Why did we choose AWS for hosting?
|
[]
|
||
ec1995c3
|
c498bc17
|
Why did we choose Spring Boot for framework?
|
[]
|
||
ec1995c3
|
d70cb090
|
What decisions were affected by our budget ($2,000)?
|
[]
|
||
ec1995c3
|
eb6d5b5a
|
What decisions were affected by our team_size (a team of 10)?
|
[]
|
||
44a26508
|
a027c7b3
|
Why did we choose FastAPI for framework?
|
[]
|
||
44a26508
|
785a7c5c
|
Why did we choose build a robust solution for approach?
|
[]
|
||
44a26508
|
268fd8d3
|
What decisions were affected by our team_size (2 developers)?
|
[]
|
||
44a26508
|
defe6556
|
What decisions were affected by our deadline (by year end)?
|
[]
|
||
9e241f74
|
410289b5
|
Why did we choose use a proven stack for approach?
|
[]
|
||
9e241f74
|
7b981fd2
|
Why did we choose Spring Boot for framework?
|
[]
|
||
9e241f74
|
acffb245
|
What decisions were affected by our deadline (next month)?
|
[]
|
||
9e241f74
|
7e2abda7
|
What decisions were affected by our team_size (a team of 10)?
|
[]
|
||
6082dd3b
|
9c2b308f
|
Why did we choose Azure for hosting?
|
[]
|
||
6082dd3b
|
6301877b
|
What decisions were affected by our budget ($50,000)?
|
[]
|
||
6e74320a
|
3bfa3ce9
|
Why did we choose AWS for hosting?
|
[]
|
||
6e74320a
|
e2e5b4bd
|
Why did we choose use existing templates for approach?
|
[]
|
||
6e74320a
|
05a10ed8
|
What decisions were affected by our budget ($2,000)?
|
[]
|
||
6e74320a
|
ac49ee6c
|
What decisions were affected by our deadline (in 2 weeks)?
|
[]
|
||
0747e010
|
b532ab09
|
Why did we choose go with MVP only for approach?
|
[]
|
||
0747e010
|
79f5575b
|
Why did we choose Rails for framework?
|
[]
|
||
0747e010
|
b14004b6
|
What decisions were affected by our deadline (in 2 weeks)?
|
[]
|
||
0747e010
|
fe1acb2a
|
What decisions were affected by our team_size (15 people)?
|
[]
|
||
c59a28d1
|
39e57f7f
|
Why did we choose AWS for hosting?
|
[]
|
||
c59a28d1
|
7b012c6c
|
What decisions were affected by our budget ($10,000)?
|
[]
|
||
66718b13
|
8d031c2d
|
Why did we choose prioritize core features for approach?
|
[]
|
||
66718b13
|
857d7689
|
Why did we choose DigitalOcean for hosting?
|
[]
|
||
66718b13
|
dc074a7d
|
What decisions were affected by our deadline (next month)?
|
[]
|
||
66718b13
|
0755258f
|
What decisions were affected by our budget ($500)?
|
[]
|
||
9b2c3828
|
b2e2cd78
|
Why did we choose Heroku for hosting?
|
[]
|
||
9b2c3828
|
1bc28860
|
What decisions were affected by our budget ($500)?
|
[]
|
||
3da31ded
|
4f9e4566
|
Why did we choose Django for framework?
|
[]
|
||
3da31ded
|
6d636ccf
|
What decisions were affected by our team_size (a team of 10)?
|
[]
|
||
1b44ab9b
|
472dd38c
|
Why did we choose Django for framework?
|
[]
|
||
1b44ab9b
|
2d227e4b
|
Why did we choose follow best practices for approach?
|
[]
|
||
1b44ab9b
|
30c8b0dd
|
What decisions were affected by our team_size (5 engineers)?
|
[]
|
||
1b44ab9b
|
4249c112
|
What decisions were affected by our deadline (end of Q2)?
|
[]
|
||
59def12e
|
5c718a84
|
Why did we choose prioritize core features for approach?
|
[]
|
||
59def12e
|
27bab82e
|
Why did we choose Google Cloud for hosting?
|
[]
|
||
59def12e
|
90398c45
|
What decisions were affected by our deadline (next month)?
|
[]
|
||
59def12e
|
7e64820c
|
What decisions were affected by our budget ($50,000)?
|
[]
|
||
3efe2745
|
c1f311ed
|
Why did we choose Spring Boot for framework?
|
[]
|
||
3efe2745
|
568245fb
|
Why did we choose build a robust solution for approach?
|
[]
|
||
3efe2745
|
21c6d090
|
What decisions were affected by our team_size (15 people)?
|
[]
|
||
3efe2745
|
947605b5
|
What decisions were affected by our deadline (by year end)?
|
[]
|
||
995e37cf
|
0b67c497
|
Why did we choose FastAPI for framework?
|
[]
|
||
995e37cf
|
0099913c
|
Why did we choose Google Cloud for hosting?
|
[]
|
||
995e37cf
|
ae2a21e5
|
What decisions were affected by our team_size (2 developers)?
|
[]
|
||
995e37cf
|
03dcc41a
|
What decisions were affected by our budget ($2,000)?
|
[]
|
||
222de20b
|
28b567aa
|
Why did we choose Spring Boot for framework?
|
[]
|
||
222de20b
|
99f9ebb9
|
Why did we choose Google Cloud for hosting?
|
[]
|
||
222de20b
|
afc51d3c
|
What decisions were affected by our team_size (a team of 10)?
|
[]
|
||
222de20b
|
b44c9d6e
|
What decisions were affected by our budget ($50,000)?
|
[]
|
||
4c9f1cce
|
0c32b3a7
|
Why did we choose Azure for hosting?
|
[]
|
||
4c9f1cce
|
adfa946c
|
Why did we choose follow best practices for approach?
|
[]
|
||
4c9f1cce
|
ccb98d0a
|
What decisions were affected by our budget ($10,000)?
|
[]
|
||
4c9f1cce
|
52997b9f
|
What decisions were affected by our deadline (end of Q2)?
|
[]
|
||
5fafd808
|
14934bcd
|
Why did we choose Heroku for hosting?
|
[]
|
||
5fafd808
|
ea471367
|
Why did we choose follow best practices for approach?
|
[]
|
||
5fafd808
|
9eec505c
|
What decisions were affected by our budget ($500)?
|
[]
|
||
5fafd808
|
0ef71939
|
What decisions were affected by our deadline (end of Q2)?
|
[]
|
||
b21fb4e0
|
058033f2
|
Why did we choose build a robust solution for approach?
|
[]
|
||
b21fb4e0
|
b0be7b6a
|
Why did we choose Google Cloud for hosting?
|
[]
|
||
b21fb4e0
|
f74526cc
|
What decisions were affected by our deadline (by year end)?
|
[]
|
||
b21fb4e0
|
aa730520
|
What decisions were affected by our budget ($50,000)?
|
[]
|
||
ec234fa1
|
91f4d8a3
|
Why did we choose use existing templates for approach?
|
[]
|
||
ec234fa1
|
669f92ba
|
Why did we choose Azure for hosting?
|
[]
|
CogCanvas is a training-free framework for maintaining long-term memory in LLM conversations. This repository contains the Controlled Benchmark used to evaluate CogCanvas's ability to extract verbatim-grounded cognitive artifacts and perform temporal and multi-hop reasoning.
The CogCanvas benchmark consists of two primary evaluation sets:
Available configurations: standard_turns, standard_facts, multihop_turns, multihop_questions.
from datasets import load_dataset
# Standard benchmark (5000 turns, 400 planted facts)
standard = load_dataset("tao-hpu/cog-canvas-benchmark", "standard_turns")
# Multi-hop reasoning benchmark (2500 turns, 176 questions)
multihop = load_dataset("tao-hpu/cog-canvas-benchmark", "multihop_turns")
@article{an2026cogcanvas,
title={CogCanvas: Verbatim-Grounded Artifact Extraction for Long LLM Conversations},
author={An, Tao},
journal={arXiv preprint arXiv:2601.00821},
year={2026},
url={https://arxiv.org/abs/2601.00821}
}