v0.1.3 — Interactive companion

Interactive companion + grounding harness. No changes to the importable Python API — analyze_pipeline() and friends are unchanged from 0.1.2.

Added

• Interactive companion widgets (web/): a dependency-free, embeddable suite that computes the paper's quantities live in the browser — a governance cockpit (build/edit a delegated workflow from a connector library; read feasibility, masking, motifs, delegation centrality, and risk; animate task tokens through the graph), plus focused explainers for feasibility, masking (M*=1.83), water-filling allocation, the Return Operator run on time, a stochastic-Petri-net token simulation, and a water-filling-vs-baseline benchmark.
• web/mso-core.js: a faithful browser port of minimal_oversight._formulae (Fisher info, return-operator step + fixed points, water-filling + solve_lambda, capacity propagation, masking, motif detection, delegation centrality, SOTA score, scope selection).
• web/mso-sim.js: a lightweight stochastic-Petri-net token simulator (tasks as tokens, review loops as real back-arcs).
• Grounding tests: tests/test_parity.py pins every ported equation to the package within 1e-6; tests/test_sim_grounding.py asserts the simulator's empirical end-to-end success matches the analytic C_op.

Documentation

• New "Interactive companion" guide, linked from the README and docs site.

Full changelog: https://github.com/crbazevedo/delegation-lab/blob/main/CHANGELOG.md

Companion video series (assets below)

An 8-part practitioner walkthrough of the widgets — each ~1.5–2 min, teaching one MSO concept live:

• 00-intro — the oversight problem + the MSO principle
• 01-feasibility — can the pipeline hit its target? (C_op vs p_min, the bottleneck)
• 02-masking — the trust trap: M* = sigma_corr/sigma_raw (reproduces M*=1.83)
• 03-waterfilling — where to place oversight (Euler-Lagrange alpha*)
• 04-return-operator — competence over time; the autonomy window
• 05-token-sim — run it: empirical end-to-end converges to the analytic C_op
• 06-benchmark — MSO vs uniform oversight; task allocation
• 07-cockpit — capstone: build a workflow and read its governance off the graph

Cockpit deep-dive series (ElevenLabs narration)

Five choreographed walkthroughs of the cockpit lessons — each reads the flagging metric, explains the theory behind the fix, applies it, and runs tokens to confirm:

• deep-01-reviewer-placement — delegation centrality; correct the shared source (diamond lesson)
• deep-02-reviewer-competence — catch rate vs skill; sigma_corr = sigma_raw + (1−sigma_raw)·c
• deep-03-merge-gate — aggregation as governance; product compounds errors
• deep-04-review-loop — loops compound catch: 1−(1−c)^k
• deep-05-chain-depth — depth as a tax; effective skill multiplies, masking compounds
• deep-06-concentrate — concentrate the fix: one node beats spreading upgrades; C_op vs centrality pick the node

Flagship — MSO masterclass (mso-concentrate-masterclass)

A ~5.5-min walkthrough: the MSO principle, every metric/term (C_op, σ_raw/σ_corr, catch rate, masking M*, B_eff, delegation centrality, motifs), then the core insight — concentrate, don't spread — across three differently-shaped workflows (upgrade the last-mile node; audit the hidden hub the ceiling can't see; cut chain depth).

v0.1.2

Summary

• Rename the preferred oversight terminology to Minimum Sufficient Oversight Principle (MSO).
• Add solve_mso() as the preferred allocation API while preserving solve_amo() as a backward-compatible alias.
• Update README, docs, notebook examples, visualization internals, and equation references to MSO terminology.
• Add regression coverage for solve_mso()/solve_amo() compatibility.

Validation

• PR #21 CI passed on Python 3.10, 3.11, and 3.12.
• main CI passed at 7063eb7.
• docs deployment passed at 7063eb7.
• local package build produced minimal_oversight-0.1.2.

v0.1.1

Summary

• Align equations and simulations with the revised arXiv paper, including prior-aware return dynamics via sigma_0.
• Update capacity, intervention, and channel-capacity helpers to match the revised assumptions.
• Reframe docs and package metadata around uncertainty-aware governed delegation for delegated AI systems.
• Add regression coverage for sigma_0, revealed/hidden channel capacity, recursive depth, and corrector-capacity thresholds.
• Fix docs deployment permissions for GitHub Pages.

Validation

• PR #19 CI passed on Python 3.10, 3.11, and 3.12.
• PR #20 CI passed on Python 3.10, 3.11, and 3.12.
• main CI passed at f22d895.
• docs deployment passed at f22d895.
• local package build produced minimal_oversight-0.1.1.

v0.1.0

Initial release