No description
  • TeX 73.8%
  • Python 26.2%
Find a file
2026-08-12 16:04:10 +10:00
archive After welfare section and before I sent the version to Shachar 2026-08-12 16:04:10 +10:00
code/analysis After welfare section and before I sent the version to Shachar 2026-08-12 16:04:10 +10:00
figures After welfare section and before I sent the version to Shachar 2026-08-12 16:04:10 +10:00
paper After welfare section and before I sent the version to Shachar 2026-08-12 16:04:10 +10:00
sessions paper: sharpen temporary-cascade figure (sustained lambda=0.744 + zoom inset near -1); park information.tex literature-comparison draft with discussion notes 2026-08-09 22:11:28 +10:00
.gitignore paper: notation pass (§1 H_n/h_n^- consolidation, q→Pr(y|h)); §2 Prop 2 → "Limit cascade at m=1"; drop \Prob macro; restore potential + dissent 2026-08-08 01:02:44 +10:00
backtocowork.md paper: notation pass (§1 H_n/h_n^- consolidation, q→Pr(y|h)); §2 Prop 2 → "Limit cascade at m=1"; drop \Prob macro; restore potential + dissent 2026-08-08 01:02:44 +10:00
CLAUDE.md paper: sharpen temporary-cascade figure (sustained lambda=0.744 + zoom inset near -1); park information.tex literature-comparison draft with discussion notes 2026-08-09 22:11:28 +10:00
README.md Initial structure: memory files, paper skeleton, archived existing work 2026-07-24 14:01:38 +10:00

m-imperfect-information

Shared repository for the paper Social Learning under Imperfect Information with m Predecessors (Çelen and Kariv). This repo is the single source of truth for both authors. Work happens in Cowork sessions run on your computer, and Claude keeps the memory and does the git work.

Repo map

  • paper/ — the LaTeX manuscript. main.tex pulls in the files under sections/.
  • code/analysis/ — Python simulations and figure scripts.
  • figures/ — figure PDFs used by the paper.
  • sessions/ — one dated note per work session. The newest is the current state.
  • CLAUDE.md — the project's memory and the session protocol. Claude reads it automatically at the start of every session.
  • archive/ — the previous files (manuscript, referee material, old code), kept as-is until migrated.

How a work session goes

  1. Open a Cowork session on this folder (on your computer).
  2. Claude pulls the latest, reads CLAUDE.md and the newest note in sessions/, and tells you where things stand.
  3. You work on the paper.
  4. When you're done, say "wrap up." Claude compiles the paper, writes the session note, commits, and pushes. That is the whole ritual.

First-time setup (once per machine)

You need a local clone and a saved credential so Claude can push without you typing anything later.

# Save your login to the keychain so you enter it only once:
git config --global credential.helper osxkeychain
 
# Clone the repo (creates the folder):
git clone https://git.celen.cloud/bogac/m-imperfect.git "m-imperfect information"
# When prompted:  Username = your Forgejo user,  Password = a Forgejo access token

Create the token on git.celen.cloud: log in, then Settings → Applications → Generate New Token, with repository read/write scope. Paste it as the password once; the keychain remembers it after that.

For Shachar

You never type a git command. Do the first-time setup above one time (username shachar, and a token you generate on git.celen.cloud). After that, open the folder in a Cowork session, work, and say "wrap up." Claude compiles, writes the note, commits, and pushes for you.