Agent of Somebody. Where the assets come from.

Every surface, sky and scanned prop under assets/ is from Poly Haven (polyhaven.com) and is CC0. They can be used, changed and shipped with no attribution. They are listed here so each file can be found again.

Surfaces, in assets/tex
asphalt_04, bark_willow, brown_leather, brown_mud, brown_planks_09, cobblestone_05, cobblestone_pavement, concrete_pavement, denim_fabric, granite_wall, green_metal_rust, metal_plate, rusty_metal_04

Skies, in assets/sky
belfast_sunset_puresky, kloppenheim_07_puresky, qwantani_dawn_puresky, qwantani_moonrise_puresky

Props, in assets/models
street_lamp_01, concrete_road_barrier, metal_trash_can, trashbag, painted_wooden_bench, fire_hydrant

The scripts that fetch and convert them are tools/game-assets.py and tools/game-models.mjs in the site's repository.

Figures, in assets/chars
The bodies are built on MakeHuman's base mesh, skeleton, skin weights and morph targets (makehumancommunity.org), which its authors released as CC0 in 2020: Data Collection AB, Joel Palmius and Jonas Hauquier. They can be used, changed and shipped with no attribution.
agent: the android. The base mesh shaped by the macro targets, the default skeleton reduced to 39 bones, and the shape of shoes05 from MakeHuman's CC0 system asset pack for the boots. Its shell panels, helmet and textures are made by tools/game-characters.py.
clerk, archivist, stamper: the three people of the Paris street (the Commis, the Archiviste and the Tamponneur). The base mesh shaped by the macro and detail targets, with the eyes shut by the eye closure targets. From the same pack: the skin old_caucasian_male, greyed and stained with ink by the tool. The clothes male_elegantsuit01 and male_worksuit01 and the shoes shoes02, darkened and dirtied by the tool. The red tape, the typed form, the iron band and collar and the stamp are made by tools/game-characters.py, the first four grown off the surface of the same mesh, and so is the one small atlas they share (mob_kit). The far copy of each figure is the same mesh with its vertices gathered into cells by the tool.

Motion, in assets/chars/clips.bin and assets/chars/mobclips.bin
The motion is captured human movement from the Carnegie Mellon University Graphics Lab Motion Capture Database, in the BVH conversion by B. Hahne, retargeted to the game's skeleton by tools/game-characters.py. Trials used: 139_02 (idle), 143_32 (walk), 127_08 (run), 127_23 (roll), 90_18 (fall), with 127_03, 143_03, 139_02 and 90_18 read for each subject's neutral stance.
The creatures' clips, in mobclips.bin: 137_15 (standing, swaying), 137_32 (standing bent), 120_04 (both arms raised, for a cast), 79_01 (the overhead blow), 91_59 (the stumble), with 143_32, 127_08 and 90_18 again for the walk, the run and the fall. Each subject's upright stance is measured from one more trial of the same subject: 137_01, 120_04, 79_64, 91_16.
The data used in this project was obtained from mocap.cs.cmu.edu. The database was created with funding from NSF EIA-0196217.

Voices, in assets/voice
What the three bosses say and what their creatures mutter is speech made by open models and then processed until little of the reader is left: lowered, doubled an octave or a fifth below, and mixed with breath, paper, fluid and static made in code. The models are Piper voices from huggingface.co/rhasspy/piper-voices. Only voices whose recordings are public domain or CC BY, and which were trained from nothing or from another such voice, were used. No voice imitates anybody.
Paris, in French: fr_FR-siwis-medium. Trained on The SIWIS French Speech Synthesis Database by Junichi Yamagishi, Pierre-Edouard Honnet, Philip Garner and Alexandros Lazaridis (University of Edinburgh, 2017, doi.org/10.7488/ds/1705), CC BY 4.0, from a base model trained on Hi-Fi TTS.
London, in British English: en_GB-cori-medium. Trained on reader 92 of the Hi-Fi Multi-Speaker English TTS Dataset by Evelina Bakhturina, Vitaly Lavrukhin, Boris Ginsburg and Yang Zhang (openslr.org/109), CC BY 4.0, which is built from LibriVox recordings in the public domain.
New York, in American English: en_US-john-medium for The Open Question and the words under the creatures' breath, en_US-norman-medium for the Old Guard, en_US-kristin-medium for Crosstalk and the Second Reviewer. All three were trained by Bryce Beattie on LibriVox recordings in the public domain and released as public domain (brycebeattie.com/files/tts).
The speech was changed: that is the notice CC BY asks for. The sounds with no words in them (coughs, breath, paper, stone, tones, static) are made by the same script from nothing.
The script that makes every clip is tools/game-voices.py in the site's repository, and the creatures' words are in tools/game-voices-lines.json beside it.

Everything else in the game is drawn by its own code. The logos of listings are loaded from the site as the game runs and belong to their owners. Listing names, grades and review lines are Anchor Terminal's, CC BY 4.0.