MainsKeeper · Round 2 / Round notesFictional demonstration data — not real equipment

Round 2 · notes · nonfinal

MainsKeeper, round two

Three new directions answer the same brief against a corrected, frozen scenario (Wed 30 Sep 2026, 15:22). These notes explain what changed, how the six differ, what was found wrong in round one, which states are covered, and what we need you to decide. No direction is selected.

Fictional demonstration data — not real equipment

What changed this round

  • Three new concepts: 4 Porchlight (site-first), 5 Meridian (time-first) and 6 Docket (exception-first), each with an identity board, a desktop dashboard with a calm / incident switch and a working interaction demo, a 390px mobile view and a mainskeeper status CLI preview with a NO_COLOR rendition.
  • Typography is now portable: 4–6 self-host SIL OFL fonts (WOFF2 with licence files inside each concept folder) instead of relying on macOS-resident faces.
  • The scenario was corrected for 4–6 (below). Concepts 1–3, the round-one comparison and the round-one notes are preserved unchanged, including their known inconsistencies.
  • A new round 2 comparison canvas shows all six marks together, links every interface, reviews 4–6 in detail and compares 1–6 in one table. The launcher gained a Round 2 section; round-one links still work.

The three new ideas

4 · Porchlight

site-first everyday tool

Starts from places. A site switcher and a plain-English fleet home (“Three places need a look. Broadwater first.”) lead into one site at a time: its UPSs, collectors or probes with role badges, incidents, and a 24-hour strip.

Answers “which place needs me?” first. Warm, light, friendly type; built for glancing, not auditing.

Detailed review · Dashboard

5 · Meridian

time-first investigation

Starts from time. Four sites × four lanes on one axis, a “Now” rail that never goes stale, and an Evidence panel for every event with event time, received time, source and certainty.

Answers “what happened, and how sure are we?” first. Cool, exact, instrument-like; investigation is the home.

Detailed review · Dashboard

6 · Docket

exception-first workbench

Starts from exceptions. A queue of dockets, most severe first, filtered with counted chips and worked by keyboard; one evidence pane; acknowledge never resolves.

Answers “what needs me now?” first. Bold flat graphics; the calm state is a single sentence: “Nothing needs you.”

Detailed review · Dashboard

How they differ

Between 4, 5 and 6

  • Home screen: Porchlight opens on places; Meridian on a timeline; Docket on a queue of open incidents.
  • Where history lives: Porchlight keeps it per site (24 h strip, Duplex timeline on desktop); Meridian makes it the main surface with explicit history-vs-now banners; Docket keeps it inside each docket’s evidence and behind Show resolved.
  • Input style: Porchlight is pointer-and-tab friendly; Meridian adds brushing, presets and nudging; Docket is keyboard-first with letter shortcuts.
  • Tone: warm and conversational · cool and exact · bold and graphic.
  • Calm state: “all normal” places · quiet lanes with the Now rail · “Nothing needs you.”

Relative to 1–3

  • Porchlight shares Vigil’s plain sentences and light ground but trades the editorial report for a site-by-site workspace.
  • Meridian inherits Feeder Board’s rigour about event time vs received time, but lays it on a shared time axis instead of a board and ledger.
  • Docket inherits Feeder Board’s ticket mechanics (acknowledge, notification delivery) in a light, flat, keyboard-driven inbox.
  • Brownout remains the only theatrical, dark-first direction; all three new concepts are light-first (Meridian also specifies a dark set).
  • 1–3 rely on macOS-resident fonts; 4–6 self-host their fonts and should render the same on any platform.

Mark vs interface

You may prefer one concept’s mark and another’s interface. They were built to be judged separately, but some separate more cleanly than others.

Marks that travel well

  • 5 Meridian — a single open line with no enclosure or palette dependence; it works in one colour on any ground. The most portable.
  • 6 Docket — three flat blocks; the silhouette holds in monochrome and at 16px, though its pink/butter accents pull the Docket palette along.
  • 4 Porchlight and 1 Vigil — simple line constructions that survive recolouring; Porchlight reads as an “m”, Vigil as a ring.
  • Least separable: 3 Brownout (gradients and lacquer are the mark) and 2 Feeder Board (drawn light-on-dark for its board).

Interfaces that travel well

  • 5 Meridian’s lanes-and-evidence model and 6 Docket’s queue-and-evidence model are structural; either could be re-skinned with another concept’s palette and type.
  • 4 Porchlight’s site workspace depends more on its warm, conversational voice, but the layout itself is neutral.
  • 2 Feeder Board’s board is portable in structure, less so in its dark dispatch styling; 3 Brownout’s interface is inseparable from its theme.

Round-two scenario changes

Concepts 4–6 use a corrected frozen scenario. Concepts 1–3 were not rebuilt and still show the round-one version.

Shop
No computer on site. Two standalone probes: shop-esp32-ups (ESP32, UPS role, USB to Shop Bench UPS) and shop-esp32-net (ESP32, connectivity role).
Duplex
No computer on site. One standalone probe: duplex-pi-probe (Raspberry Pi 4 running mainskeeperd, UPS + connectivity).
Fleet
6 collectors (bw-pi-01, home-studio, home-esp32-garage, shop-esp32-ups, shop-esp32-net, duplex-pi-probe) and 6 UPSs.
Incident
4 of 6 collectors reporting (both Shop probes silent since 15:10); 5 of 6 UPSs communicating (Shop Bench UPS last known 15:10).
Calm
6 of 6 collectors reporting; 6 of 6 UPSs communicating.

Inconsistencies found in round one (recorded, not fixed — concepts 1–3 are preserved)

  1. Collector count. Round-one dashboards say 3 of 5 collectors reporting in the incident, while the round-one inventory lists six collectors (bw-pi-01, home-studio, home-esp32-garage, shop-bench-win, shop-esp32-probe, duplex-nuc). With both Shop collectors silent the count should read 4 of 6.

    concepts/a/dashboard.html:422 · concepts/a/mobile.html:224 · concepts/b/dashboard.html:367 · concepts/b/mobile.html:284 · concepts/c/dashboard.html:813 · concepts/c/mobile.html:299

  2. Computers at Shop and Duplex. Round-one Shop monitoring ran on a Windows 11 bench PC (shop-bench-win) and Duplex on a Fedora Linux 42 NUC (duplex-nuc), which conflicts with those sites having no computers. Round two replaces them with standalone probes.

    e.g. concepts/a/dashboard.html:731 (shop-bench-win, Windows 11) · concepts/a/cli.html, concepts/b/cli.html, concepts/c/cli.html (duplex-nuc, Fedora Linux 42)

  3. macOS-resident fonts. Round-one concepts rely on Iowan Old Style / Charter (Vigil), DIN Alternate (Feeder Board) and Didot / Futura (Brownout), so they render with fallbacks on Windows, Linux and Android and look different from what was reviewed.
  4. Found on reading the round-one sources:
    • The calm state carries the same error: “5 of 5 expected collectors reporting” (should be 6 of 6), while the Home site alone lists two collectors. concepts/a/dashboard.html:428 · concepts/a/mobile.html:228 · concepts/b/dashboard.html:368 · concepts/b/mobile.html:285 · concepts/c/dashboard.html:813 (data-calm="5" of 5) · concepts/c/mobile.html:299
    • Vigil’s report kicker says “four sites, six UPSs, five collectors”, contradicting its own site sections, which name six collectors. concepts/a/dashboard.html:392
    • No other disagreement in counts, times or device names between concepts 1–3 was found; incident times, backlog figures (1,248 pending; 382 of 1,630; batch 16 of 64), the 14:30–14:52 gap and the 14:03 → 15:17 delayed record agree across all three.

Builder-invented fixture values in 4–6 (disclosed)

  • Porchlight: calm-state Duplex runtime estimate of 46 min (dashboard and mobile) and the calm CLI probe “up 12 days” — neither is in the frozen scenario. concepts/d/dashboard.html · concepts/d/mobile.html · concepts/d/cli.html
  • Meridian: received times other than the scenario’s 14:03 → 15:17 are invented for the evidence panel (for example 15:02:04, 15:02:06, 15:09:04, 15:10:30, 15:21:56, 15:21:57, 06:42:11). concepts/e/dashboard.html
  • Docket: Duplex notification times (13:44 opened, 15:02 recovered) are inferred; the scenario says only “notifications delivered”. concepts/f/dashboard.html

State inventory

Demonstrated

  • Healthy reporting (calm fleet)
  • Partial hardware capabilities (“Not reported”)
  • On battery with live readings (Broadwater)
  • Contact lost, cause unknown (Shop)
  • Recovery with history uploads draining (Duplex)
  • History gap (14:30–14:52, no records)
  • Uncertain timestamps (14:52 ±90 s)
  • Acknowledged but still open
  • Notification delivery failure (Shop push, retrying)
  • Resolved history (Docket Show resolved; Meridian yesterday’s Duplex outage)
  • Historical vs current distinction (Meridian “Viewing history” banner and Return to now)

Still to design

  • First use, no devices and initial loading
  • Low estimated runtime
  • Utility restoration detail
  • UPS communication lost while the collector still reports
  • Suspected vs confirmed connectivity failure
  • Local storage limits
  • Enrollment states (waiting, approved, expired, rejected, offline)
  • Authentication expiry and revoked credentials
  • Maintenance windows
  • The full resolution flow

Questions for you

  1. Which mark or marks should carry forward, independent of which interface you prefer?
  2. What should the home screen be: places (site-first), a timeline (time-first) or a queue (exception-first)?
  3. For daily use, which feel suits you: warm and light, cool and exact, or bold and graphic?
  4. Should investigation be a separate mode you enter, or the home itself?
  5. How much does mobile matter — glance-and-acknowledge only, or full investigation on a phone?
  6. Do the round-one directions (1–3) stay in the running, now that 4–6 exist?

No winner is selected and no concepts are combined this round.