Establish CONTRACTS.md at the EVOLV root as the canonical map of where every
contract, rule, and standard lives. Surface it from CLAUDE.md so every fresh
agent or colleague lands there first.
Reshape .claude/refactor/ to reflect that the platform refactor is done:
live standards stay at the top level; the plan artifacts (CONTINUE_HERE.md,
TASKS.md) move into Archive/ with WARNING banners.
Drop content that drifted out of date or duplicated the new standards stack:
- docs/DEVELOPER_GUIDE.md (pre-refactor walkthrough; superseded by
wiki/Architecture, wiki/Getting-Started, .claude/rules/node-architecture,
.claude/refactor/MODULE_SPLIT + per-node CONTRACT.md + src/commands/).
- .agents/decisions/ (15 DECISION files): load-bearing decisions belong in
commit messages and PR descriptions; live open items in OPEN_QUESTIONS.md.
- .agents/improvements/TOP10_*.md: moved to Archive/.
Bump generalFunctions to 49c77f2 — adds CONTRACT.md inside the library:
different shape from per-node CONTRACT.md files (library API, not msg.topic),
with stability tags and pointers to .claude/refactor/CONTRACTS.md §N.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
MGC submodule lands the 2026-05-14 governance review fixes plus rolled-up
session work: _output-manifest.md per the new output-coverage rule,
computeEqualFlowDistribution extracted as a pure function (testable without
MGC), groupEfficiency degenerate-case fix, unit-self-describing set.demand,
eta = (Q*dP)/P formula correction, and dashboard fan-out hardening
(auto-init, NCog normalization, Q-H trim, null-trap closure). Suite 108/108.
Superproject adds:
- .claude/rules/output-coverage.md: every-output-every-state testing rule
prompted by the eta-null crash earlier in the session.
- CLAUDE.md: pointer to the new rule under Conventions.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
- package.json: remove @tensorflow/tfjs and @tensorflow/tfjs-node.
Monster's TF code was already stripped; the deps were stale and kept
pulling a heavy native binary back into every install.
- .gitignore: ignore .repo-mem/ regenerable indexes and per-session
.claude/*.lock runtime files.
- CLAUDE.md: prepend READ-FIRST pointer to .claude/rules/repo-mem.md;
collapse the 'three outputs' bullet to a pointer at node-architecture.
- .claude/rules/telemetry.md: drop Port 0/1/2 duplication; reference
node-architecture.md.
- .claude/rules/testing.md: stop requiring a separate test/edge tier and
the basic/integration/edge example flow trio. Reflects what nodes
actually do.
- .claude/rules/repo-mem.md (new): when-to-call-which guide for the
per-repo memory MCP, anti-patterns, refresh model.
- .mcp.json (new): wire repo-mem stdio server.
- docs/DEVELOPER_GUIDE.md (new): step-by-step guide for adding a new
EVOLV node under the three-layer pattern.
- Bump nodes/pumpingStation to 6ab585b (docs + simulations refresh,
spill-flow path renames consistent with d8490aa).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The demo was a single 96-node tab with everything wired directly. Now
4 tabs wired only through named link-out / link-in pairs, and a
permanent rule set for future Claude sessions to follow.
Tabs (by concern, not by data flow):
🏭 Process Plant only EVOLV nodes (3 pumps + MGC + PS + 6 measurements)
+ per-node output formatters
📊 Dashboard UI only ui-* widgets, button/setpoint wrappers, trend
splitters
🎛️ Demo Drivers random demand generator + state holder. Removable
in production
⚙️ Setup & Init one-shot deploy-time injects (mode, scaling,
auto-startup, random-on)
Cross-tab wiring uses a fixed named-channel contract (cmd:demand,
cmd:mode, cmd:setpoint-A, evt:pump-A, etc.) — multiple emitters can
target a single link-in for fan-in, e.g. both the slider and the random
generator feed cmd:demand.
Bug fixes folded in:
1. Trend chart was empty / scrambled. Root cause: the trend-feeder
function had ONE output that wired to BOTH flow and power charts,
so each chart received both flow and power msgs and the legend
garbled. Now: 2 outputs (flow → flow chart, power → power chart),
one msg per output.
2. Every ui-text and ui-chart fell on the (0, 0) corner of the editor
canvas. Root cause: the helper functions accepted x/y parameters
but never assigned them on the returned node dict — Node-RED
defaulted every widget to (0, 0) and they piled on top of each
other. The dashboard render was unaffected (it lays out by group/
order), but the editor was unreadable. Fixed both helpers and added
a verification step ("no node should be at (0, 0)") to the rule set.
Spacing convention (now codified):
- 6 lanes per tab at x = [120, 380, 640, 900, 1160, 1420]
- 80 px standard row pitch, 30-40 px for tight ui-text stacks
- 200 px gap between sections, with a comment header per section
New rule set: .claude/rules/node-red-flow-layout.md
- Tab boundaries by concern
- Link-channel naming convention (cmd:/evt:/setup: prefixes)
- Spacing constants
- Trend-split chart pattern
- Inject node payload typing pitfall (per-prop v/vt)
- Dashboard widget rules (every ui-* needs x/y!)
- Do/don't checklist
- Link-out/link-in JSON cheat sheet
- 5-step layout verification before declaring a flow done
CLAUDE.md updated to point at the new rule set.
Verified end-to-end on Dockerized Node-RED 2026-04-13: 168 nodes across
4 tabs, all wired via 22 link-out / 19 link-in pairs, no nodes at
(0, 0), pumps reach operational ~5 s after deploy, MGC distributes
random demand, trends populate per pump.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
