**Top engineering priorities (from importance):** the matrix puts **F2 (connect, 105)**, **F6 (polarity, 99)** and **F9 (undo, 98)** above the headline detector **F5 (81)**. The lesson: loop insight is only as good as its *inputs* and *ergonomics* — getting wiring (F2) and polarity capture (F6) effortless matters as much as the detection itself, and a reliable safety net (F9) underwrites fearless building. F4 (58) is real but lowest-leverage — keep it lightweight (guide, don't nag).
**Top engineering priorities (from importance):** the matrix still puts **F2 (connect, 105)**, **F6 (polarity, 103)** and **F9 (undo, 98)** at the top, with the headline detector **F5** climbing to **93** — G4 reinforces it, because _seeing_ a loop classified R/B is itself the lesson. The lesson holds: loop insight is only as good as its _inputs_ and _ergonomics_ — effortless wiring (F2) and polarity capture (F6) matter as much as detection, and a reliable safety net (F9) underwrites fearless building. **F10 (samples, 45)** is the lowest-leverage function, exactly right for a supporting aid: it earns its place by being nearly free — static Models reusing the whole pipeline, no new model code — not by priority. F4 (58) stays low: keep it lightweight (guide, don't nag).
- **F5 ⊗ F3** — live loop enumeration on every change vs the 60 fps frame budget. The design-shaping conflict. Mitigated by the **cap** (≤200 cycles) + running detection **off the render frame** (debounce / idle callback / worker). Owned by ADR-0001 + the §7 budget.
- **F2 × F4** — guardrails can add friction to connecting. Mitigated by *guiding* the gesture (invalid handles refuse the drop) instead of erroring afterwards.
- **F2 × F4** — guardrails can add friction to connecting. Mitigated by _guiding_ the gesture (invalid handles refuse the drop) instead of erroring afterwards.
- **F3 × F9** — snapshot cloning on every change costs memory/time. Bounded by small models + a fixed-size ring buffer; revisit with structural sharing only if profiling demands.
- **F2 ◎ F6, F5 ◎ F6** — polarity is captured *in* the connect gesture and *is* the detector's input; these reinforce, so build them together.
- **F2 ◎ F6, F5 ◎ F6** — polarity is captured _in_ the connect gesture and _is_ the detector's input; these reinforce, so build them together.
- **F10 × F7** — loading a sample calls `setModel`, which _replaces the whole document and clears undo history_. Against autosave (F7), one click could wipe real work. Mitigated by confirming the replace only when the canvas is non-empty (frictionless on the empty start). Owned by T6. (F7 and F10 sit outside the core roof table above; this is the new tension this release.)
| T1 | Vue Flow over hand-rolled SVG | fast F1–F3, batteries-included editor mechanics | a dependency owning the view layer + a Model↔VueFlow projection to maintain | ADR-0002 |
| T2 | Flow-as-node + materialised Clouds | uniform graph for loops/render; info-links can target flows; sim-clean | Cloud lifecycle management; a node that renders as a pipe | ADR-0003 |
| T3 | Enumerate loops with a cap | per-loop R/B insight (G2) within the frame budget | capped completeness on pathological graphs ("+N more") | ADR-0001 |
| T4 | IndexedDB(`idb`) behind `ModelRepository` | size headroom + swappable engine (sync later) | async wrapper + a small dependency; no sync today | — |
| T5 | Snapshot undo over command pattern | simplicity and correctness | memory per snapshot (bounded by ring buffer + small models) | — |
| T1 | Vue Flow over hand-rolled SVG | fast F1–F3, batteries-included editor mechanics | a dependency owning the view layer + a Model↔VueFlow projection to maintain | ADR-0002 |
| T2 | Flow-as-node + materialised Clouds | uniform graph for loops/render; info-links can target flows; sim-clean | Cloud lifecycle management; a node that renders as a pipe | ADR-0003 |
| T3 | Enumerate loops with a cap | per-loop R/B insight (G2) within the frame budget | capped completeness on pathological graphs ("+N more") | ADR-0001 |
| T4 | IndexedDB(`idb`) behind `ModelRepository` | size headroom + swappable engine (sync later) | async wrapper + a small dependency; no sync today | — |
| T5 | Snapshot undo over command pattern | simplicity and correctness | memory per snapshot (bounded by ring buffer + small models) | — |
| T6 | Samples load by replacing the whole Model | a sample is _just a Model_ — no special load path; reuses the whole pipeline; one click | the load is destructive (`setModel` clears undo history); guarded by a confirm on a non-empty canvas | — |
- **Brief vocabulary vs glossary.** The original request said "functions," "arrows," "faucets"; DESIGN uses the resolved CONTEXT.md terms verbatim — **Converter**, **Flow**/**Information Link**, faucet dropped.
- **G3 ambiguity.** "Safe and portable" initially read as possible cloud sync; clarified — G3 is satisfied by **local persistence + JSON export**, not sync (PouchDB deferred, see Tensions).
- **Storage default drift.** Initial recommendation was localStorage; resolved to **IndexedDB via `idb`**, behind a `ModelRepository` interface so the engine choice is no longer hard-to-reverse.
- **Teaching intent unrecorded.** The project is framed as Meadows-inspired (README) and the whole vocabulary is pedagogical (CONTEXT.md), yet no Goal captured "teach the language" until now — the intent lived only in the vision. Surfaced and recorded as **G4** (weight 4), answered by the sample gallery (**F10** / **C13**).
_Part of [meadows](../../README.md) · see [DESIGN.md](../../DESIGN.md)._
In the domain `Model`, a **Flow** is a *node* carrying `source` and `target`
references (each to a Stock or Cloud), not an edge. The pipe you see is *rendered*
In the domain `Model`, a **Flow** is a _node_ carrying `source` and `target`
references (each to a Stock or Cloud), not an edge. The pipe you see is _rendered_
from those references. The **Source/Sink** boundary is a **materialised Cloud
node**, so every Flow connects node→node uniformly — no `null` ends.
Reference in New Issue
Block a user
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
Julien Calixte