The example is hand-authored to show the interface. Its numbers are not measured results.
The report
Part
What it shows
Call locations
Every indirect call that has a debug location, with its number of targets before and after the analysis.
Call graph
The listed call locations and the functions they can reach, drawn as one graph. Each edge is marked as kept, added or removed.
Targets
For the selected call, each function that was added, removed or kept.
Evidence
For each target, the check that supports it and where the analyzer recorded it.
Source
The lines around the call, read from the source directory you supply.
Profiles
The sound, checked and precise results of one run, chosen from a menu.
Pending decisions
Changes the analyzer proposed and did not adopt, each with its reason.
Method
Type-based indirect-call analyses resolve a call through a structure field by matching the field's type against the functions assigned to fields of that type. Since LLVM 17, pointers in the IR are opaque. The IR no longer says which field a call loads its function pointer from, so that matching key is gone.
Facet reconstructs the relation itself. For each indirect call it recovers the dispatched field from debug information. For each field it recovers the set of functions the program assigns to it, counting initializers, stores and aggregate copies. The two sides are joined by field identity.
Every change to the starting call graph is one fact: one edge added or one edge pruned. A fact is accepted only when a rule verifies it from program evidence, and it keeps a record of the rule, the evidence it used and the premise it holds under. A language model may be asked about the facts that remain undecided, among candidates the symbolic stages have already bounded, and its answers pass through the same rules.
One run yields two call graphs: one that preserves recall and one that is refined.
Results across language models
The language model answers only the questions the symbolic stages leave open. To see how much the result depends on which model answers them, the same pipeline ran on 14 open-source C programs three times: with Gemini 3.7 Flash, with Claude Opus 5.5 at high reasoning effort, and with GLM-5.3, a model from a third vendor, at maximum reasoning effort. The recall-preserving graph accepts no model decision, so it is the same in all three runs.
Mean F1 over the 14 programs
Recall-preserving graph, no model decision0.448
Refined graph, Gemini 3.7 Flash0.709
Refined graph, Claude Opus 5.5, high reasoning effort0.717
Refined graph, GLM-5.3, maximum reasoning effort0.714
00.20.40.60.81
F1 for each program
Recall-preserving graph, no model decision
Refined graph, Gemini 3.7 Flash
Refined graph, Claude Opus 5.5, high reasoning effort
Refined graph, GLM-5.3, maximum reasoning effort
nginx0.255 Recall-preserving graph0.457 Gemini 3.7 Flash0.510 Claude Opus 5.50.500 GLM-5.3
sqlite0.106 Recall-preserving graph0.509 Gemini 3.7 Flash0.515 Claude Opus 5.50.515 GLM-5.3
dovecot0.358 Recall-preserving graph0.481 Gemini 3.7 Flash0.482 Claude Opus 5.50.471 GLM-5.3
libarchive0.267 Recall-preserving graph0.710 Gemini 3.7 Flash0.718 Claude Opus 5.50.740 GLM-5.3
tmux0.337 Recall-preserving graph0.655 Gemini 3.7 Flash0.655 Claude Opus 5.50.655 GLM-5.3
redis0.806 Recall-preserving graph0.904 Gemini 3.7 Flash0.930 Claude Opus 5.50.912 GLM-5.3
libjpeg-turbo0.101 Recall-preserving graph0.805 Gemini 3.7 Flash0.811 Claude Opus 5.50.788 GLM-5.3
lua0.828 Recall-preserving graph0.913 Gemini 3.7 Flash0.914 Claude Opus 5.50.914 GLM-5.3
gdbm0.583 Recall-preserving graph0.693 Gemini 3.7 Flash0.693 Claude Opus 5.50.693 GLM-5.3
oniguruma0.578 Recall-preserving graph0.989 Gemini 3.7 Flash0.989 Claude Opus 5.50.989 GLM-5.3
libxml20.199 Recall-preserving graph0.474 Gemini 3.7 Flash0.485 Claude Opus 5.50.477 GLM-5.3
libdwarf0.593 Recall-preserving graph0.631 Gemini 3.7 Flash0.631 Claude Opus 5.50.631 GLM-5.3
md4c0.744 Recall-preserving graph0.744 Gemini 3.7 Flash0.744 Claude Opus 5.50.744 GLM-5.3
sudo0.511 Recall-preserving graph0.962 Gemini 3.7 Flash0.962 Claude Opus 5.50.962 GLM-5.3
00.20.40.60.81
Values as a table
Program
Recall-preserving graph
Refined graph, Gemini 3.7 Flash
Refined graph, Claude Opus 5.5
Refined graph, GLM-5.3
Mean of 14
0.448
0.709
0.717
0.714
nginx
0.255
0.457
0.510
0.500
sqlite
0.106
0.509
0.515
0.515
dovecot
0.358
0.481
0.482
0.471
libarchive
0.267
0.710
0.718
0.740
tmux
0.337
0.655
0.655
0.655
redis
0.806
0.904
0.930
0.912
libjpeg-turbo
0.101
0.805
0.811
0.788
lua
0.828
0.913
0.914
0.914
gdbm
0.583
0.693
0.693
0.693
oniguruma
0.578
0.989
0.989
0.989
libxml2
0.199
0.474
0.485
0.477
libdwarf
0.593
0.631
0.631
0.631
md4c
0.744
0.744
0.744
0.744
sudo
0.511
0.962
0.962
0.962
F1 compares the targets a graph gives each indirect call with the targets observed at that call while the program ran a workload, averaged over the calls. Observation is a lower bound: a target that was never observed counts against precision even when the call can reach it.
Earlier run, reported in the paper
The paper repeats the analysis under Gemini 2.5 Flash, Gemini 3.7 Flash and GLM-5.3. These are the mean F1 values it reports for the refined graph. Its Gemini 3.7 Flash value comes from an earlier run and differs slightly from the one above.
Refined graph, Gemini 2.5 Flash0.689
Refined graph, Gemini 3.7 Flash0.708
Refined graph, GLM-5.30.714
00.20.40.60.81
Requirements
Input
Whole-program LLVM bitcode, preferably built with debug information.
Where it runs
On your own machine. The browser workspace listens only on the loopback address.
API key
None for the symbolic analysis. The language model is optional.