Reset the engine after a template builds, not before

[zntznt]

The bug

Loading a starter template left the step-0 history entry empty, so every chart opened on a hole. Run the Predator & Prey template straight after loading it and the Rabbits series came back:

[null, 89, 94, 93, 87, 76]

The starting population of 80 is nowhere in it and the curve begins at step 1.

Cause

Purely ordering. Both loaders called _clearAll(), which resets the engine, and only then ran the template's load(). reset() ends by recording a step-0 baseline of the current nodes, so it captured an empty diagram; the nodes arrived afterwards and nothing reset again.

The saved-diagram loader already had this right: clear, load, then reset.

Measured before and after, same template:

before   snapEntries: 0   for 3 nodes   series[0] = null
after    snapEntries: 3   for 3 nodes   series[0] = 80

Why it matters beyond the chart

That baseline is also what scrub shows at step 0, and what spike attribution reads as the "from" value for the first real step. All three were working from nothing.

The fix

Reset once the template has built its nodes.

Both entry points (Library template, welcome demo) carried their own copy of the same nine lines, which is how one bug came to live in two places. They now share _installTemplate, and each keeps only its own last step: the Library hides its modal, the welcome demo re-renders the properties panel.

Files

• js/app-library.js — _installTemplate helper with the corrected reset order; _loadTemplate reduced to the guard plus its modal
• js/app.js — _loadDemo reuses the helper
• test/smoke.js — regression test

UI layer only. No change to model.js or engine.js, and no serialized fields added.

Testing

• node test/run.js → 235 passed, 0 failed (unchanged; UI-layer change)
• npm run smoke → SMOKE PASSED, no console or page errors
• The new smoke test drives both entry points, asserting every node appears in the step-0 baseline and that a stepped series has no hole at index 0. Without the fix it reports step0Entries: 0, covered: false, firstIsNull: true for both paths.

The test calls _loadTemplate and _loadDemo rather than the new shared helper deliberately: an earlier version called _installTemplate, which meant that without the fix it threw a ReferenceError instead of detecting the bug, so it would have "failed" for the wrong reason. Driving the real entry points measures behaviour rather than the presence of a function.

Worth noting the existing P3 CSV: history export (header + step-0 baseline + 4 steps) test passes both with and without this change: its diagram is not template-loaded, which corroborates that the step-0 baseline works normally and this was specific to the template path.

---

Generated by Claude Code