docs: organize research into publication pipelines and recurring Agent protocol

[LightChainr]

Purpose

Convert the current research DAG into a publication-oriented portfolio without changing scientific priority, execution authorization, frozen protocols, raw evidence, or claim levels.

This PR is documentation only. It evaluates the current mature/near-mature research lines as candidate papers and gives a recurring Agent protocol whose unit of progress is a paper claim rather than a PR/Issue/count.

Added

docs/PUBLICATION-PORTFOLIO.md

Defines two separate rankings — manuscript readiness and ultimate scientific upside — and records explicit publication gates for:

1. cut-network / predictive-state nonclosure;
2. certified bounded algebraic threshold exclusion;
3. global matching-odd finite-size response + scalar-closure falsification;
4. finite positive dynamics / operator de-identification;
5. original-U observable identifiability flagship;
6. reserve terminal-algebra, capillary-effective-model, and proof/statistics method tracks.

Each track states the central claim, existing nucleus, highest-value tasks, nonclaims, and a stop rule. The portfolio deliberately prevents cheap adjacent census/size/descriptor expansion from being mistaken for publication progress.

docs/AGENT-PUBLICATION-PROMPT.md

A state-refreshing recurring research prompt. Each run must:

• re-read live repository state, recent comments, open/branch-only/closed-unmerged assets;
• score 3–5 candidate tasks by manuscript delta, information gain, cost, reusability and leakage risk;
• choose exactly one PRIMARY task;
• prefer theorem/existing-data/counterexample/identifiability work over new acquisition;
• preserve dependency/chronology/observable semantics;
• return a typed verdict and update one paper dossier;
• move mature tracks into manuscript mode rather than indefinitely extending them.

Publication-planning judgment encoded here

• Promote now: cut-network/nonclosure, certified threshold exclusion, global matching-odd/scalar-closure synthesis, P398 realization/de-identification.
• Highest-impact flagship but not forced to a positive operator identity: original-U identifiability/asymptotics.
• Reserve until a structural gate is crossed: terminal algebra must reach probability comparison/obstruction; capillary must remain reduced/effective-model scoped until physical bridges close; proof-carrying elimination needs genuine cross-case generalization before becoming a methods paper.

Boundaries

• no code/data/scorer changes;
• no Issue priority or lifecycle changes;
• no production authorization;
• no merge of branch-only science into main by implication;
• publication readiness is explicitly separate from Issue priority and scientific integration state.

Branch is two commits ahead of current main, touching only the two new documentation files.