Start with the panels

Three self-contained pages. Nothing to install, no server, no build step — they run entirely in your browser.

The panel interface is in Vietnamese. Variable names, Greek symbols and numbers are not language-specific, and each sheet group is labelled Σ₀ … Σ₅.

The result worth arguing about

Three quantities that are all exactly zero in the original model — not small, zero, and zero for a structural reason.

The original model's layer E9 is a function of the current state and carries no history, so two paths reaching the same state must produce the same output. It is not missing a term — it sits in a function class that cannot express "the hole that stays behind."

That turns into an empirical question, and it does not need a long sample:

Does the order of two shocks change the permanent level, when the final state is identical?
Yes → fatigue. No → hysteresis.

Note the hysteresis commutator is zero by the definition of the operator: it takes a running maximum over history, and two orderings traverse the same state set. Holonomy and residue are non-zero in both. A closed loop returns to the starting state; the scar remains.

Three packages, three different things

The relationship between them is the interesting part.

amr2 stands apart because it is a rival model and has to be comparable. amr3 nests amr because it is the same model in different coordinates, and the way you prove a re-reading is faithful is to run the original inside it:

τ ≡ 0  →  no overrides  →  ln DXY identical to amr, difference 0.000e+00

amr3 adds exactly one parameter and no new edges — the edge set is taken unchanged from the gamma field of amr/registry.py.

Run it yourself

Python 3.9+. No dependencies at all — standard library only.

git clone https://github.com/vinhatson/amr-engine.git
cd amr-engine
python -m amr check            # structural constraints, with the slack on each

python -m amr list             # the scenarios, and what the book says about each
python -m amr run narrative    # a full qualitative reading
python -m amr audit            # every scenario against the book's own numbers
python -m amr2                 # book model vs corrected model, same exam
python -m amr3                 # the atlas, three measurements, the τ sweep

python tests/test_engine.py    # amr
python tests/test_amr2.py      # amr2
python tests/test_amr3.py      # amr3

All three suites pass on Python 3.12 across Windows, macOS and Linux. They check structure, not just output: test_atlas_covers_every_amr_variable_exactly_once asserts 78 of 78, nothing added and nothing dropped.

Honest limits

• The rendered narrative is Vietnamese only. The method is narrative_vi(); there is no narrative_en() yet. Everything structured — numbers, arrow scale, block attribution, evidence_mix — is language neutral, and the registry carries a name_en field. An English renderer is the most useful contribution anyone could make.
• On Σ₅ the tautness rule is exact; on Σ₁ it is an approximation, because σ_sys is a linear composite rather than a logistic. The code says so rather than passing over it.
• τ defaults to 0, which is the null. Quote no τ > 0 result without stating τ and the memory model.
• One data defect is carried, not hidden. CPI for 2025-10 does not exist — the BLS did not publish during the shutdown that began 2025-10-01 — so that month is interpolated and row.cpi_imputed is True there.
• The geometric framing is a reading, not a dimension count. A two-sheet model that scars is already a puncture; a ten-sheet model that does not is still a slide. The invariant that matters is whether the layer operators commute.

Documentation