Begin IB 0.2 as a shell-first Pensieve

[isomorphisms]

Purpose

Start IB 0.2 from the persistent information substrate rather than from the old Idriç browser core or Android renderer work.

This line is deliberately arXiv-only and shell-first.

0.2 model

• Cauldron: acquisition/intake. Keeps requested URL, arXiv abstract HTML, paper HTML when available, PDF, fetched image resources, acquisition time, and a figure manifest.
• Pensieve: distilled/findable material. Keeps source linkage, title, local text representations, links, and the figure manifest.
• Indexes: rebuildable views over the Pensieve. The first one is only an exact-text file list used by local grep-like search.
• Hooks: executable after-distill.d process boundary for later vector spaces, hyperplanes, other indexes, and model-context adapters.

A tab or renderer is not part of this slice.

Commands

bin/ib fetch ARXIV...
bin/ib distill ARXIV...
bin/ib add ARXIV...
bin/ib search TEXT
bin/ib reindex
bin/ib paths

ICU is the HTTP boundary. The 0.2 path does not use curl.

Initial corpus

• 1107.0595
• 2203.11355
• 1901.09021
• 1606.05336
• 2305.00241

Acceptance

The branch has a deterministic fake-ICU round trip proving:

• abstract/HTML/PDF acquisition into the Cauldron;
• figure discovery and acquisition;
• Cauldron -> Pensieve text/link/figure distillation;
• local exact-text retrieval without returning to the network;
• after-distill hooks receive both Pensieve and Cauldron paths;
• re-adding an already complete Cauldron item performs no second acquisition.

GitHub Actions run 34287017806 is green on exact head fc3975efc141ceda8a0491062a7333e1f0169845.

Deliberate limits

This is not general Web compatibility, a renderer, Android work, or a new typed browser core. Existing pre-0.2 material remains present as reference while the smaller architecture proves itself.