epic(dashboard): PM stack coherence - single-URL, fresh data, triage-able, hierarchical #1184

Summary

Epic for the PM stack coherence program, commissioned by Drew's full-stack review (2026-10-26 ~22:35 CDT). Drew owns the verdict this epic answers: the stack's beginnings exist, but it is "not easy to navigate or particularly coherent… the data is stale or only partially adopted." His requirements (the yardstick for every child):

• ONE single URL, mobile or desktop, to see ALL projects
• On it: quotas & usage; page health; issues that need his attention
• REMEDIATE / respond / basic triage from there — not read-only
• Queued + in-progress work clear: what would be picked up NEXT, what's being worked on, what's BLOCKED, what was DONE recently
• Hierarchy with rollups: roadmap item > epic > issue (not just issue level)
• The glued-together topology (services, ports, Kaneo vs the rest) documented ON the dashboard

Headline evidence from the review (2026-10-26 ~22:40 CDT, laptop + gh, cross-checked)

• /attention shows "Needs Drew (0)" while the published orchestration.json carries 11 NEEDS DREW items — tier source rejected (fix(aggregator/orchestrator): orchestration section rejected — provenance is an object, isOrchestrationSection requires a string #1110) with NO indication on the page
• Every footer falsely claims the orchestrator "has not published a section" (it publishes hourly; the aggregator rejects it)
• Board: 95 of 107 cards are Done — a completion log again
• Items layer: "fixtures for 172 ref(s) outside the pool" — fixture states wear live badges
• Pool: 139 items, 17 with no Area; Grist 23 rows vs 16 Project epics (spec wants 1:1)
• No health page, no staleness alerting, no triage write actions beyond comments (comments store still holds only the 3 demo comments from Oct 25)
• All services themselves were UP at review (aggregator/Kaneo/Postgres/Pad/Grist containers healthy; ready-for-agent, tunnel live) — the gap is coherence, not uptime

Children — information architecture & front door

• impl(aggregator): make / the stack front door - overview of all projects, attention, health, quotas (board moves to /board) #1134 front door: / becomes the stack overview
• fix(aggregator): navigation coherence - 10-item nav overflows on mobile, sub-pages undiscoverable #1135 navigation coherence + mobile nav
• impl(aggregator): deep-link every card/row out to its live service (Kaneo, Grist, Pad, ready-for-agent) #1136 deep-links out to Kaneo/Grist/Pad/ready-for-agent
• impl(aggregator): services & ports topology table on the dashboard - what runs where, bound how, consumed by whom #1137 services & ports topology table on the dashboard
• fix(aggregator): provenance footer is one jargon line and contradicts page content - replace with per-source freshness chips #1138 provenance footer -> per-source freshness chips

Children — data freshness, honesty, adoption

• fix(aggregator/orchestrator): orchestration section rejected — provenance is an object, isOrchestrationSection requires a string #1110 (pre-existing) orchestration section rejected: provenance shape
• fix(aggregator): /attention presents derived-only data as authoritative when the orchestration tier source is missing - banner + never a silent 0 #1139 /attention silent-0 honesty (banner, never derived-only as authoritative)
• fix(aggregator): health/footer misdiagnoses a rejected orchestration section as 'has not published' - distinguish absent vs invalid #1140 absent vs invalid diagnosis in /api/health
• impl(aggregator/collector): Project pool resolves a minority of board/roadmap refs (172 fixture fallbacks) - raise coverage and publish the ratio #1141 pool coverage for board/roadmap refs (172 fixture fallbacks)
• impl(orchestrator/project): Area-unset pool items (17) are invisible to per-project pages - triage pass + taxonomy check #1143 Area-unset triage (17 items)
• impl(pm-stack): reconcile Grist roadmap rows vs Project epics 1:1 and publish a drift check #1144 Grist <-> Project 1:1 reconciliation + drift check
• impl(dashboard): comments loop never exercised since the Oct 25 demo - verify triage/GitHub-post loop end-to-end and retire demo data #1145 comments loop end-to-end verification
• fix(aggregator): board is again a completion log - 95 of 107 cards are Done; age out / collapse done by default #1146 board Done composition (95/107)

Children — health & observability

• impl(aggregator): service & page health page - live probes for every stack service, feed ages, last runs #1147 service & page health page
• impl(fleet): staleness watchdog - surface dead/stale stack inputs as NEEDS ORCHESTRATOR instead of silent decay #1148 staleness watchdog -> NEEDS ORCHESTRATOR
• fix(pm-stack): aggregator install.sh does not restart a running service - deploys silently no-op #1150 install.sh deploy-restart fix

Children — triage & remediation (Drew acts from the dashboard)

• design+impl(aggregator): triage actions on /attention - acknowledge, snooze, escalate, answer - orchestrator-mediated #1152 triage actions on /attention (acknowledge/snooze/escalate/answer)
• impl(aggregator): 'request status change' action on project/board cards - orchestrator applies it as single writer #1154 request status change on cards (orchestrator applies as single writer)
• impl(aggregator): session actions from /sessions and /attention - poke, answer, stop - keeper/orchestrator-mediated #1156 session actions (poke/answer/stop, mediated)

Children — queue visibility & hierarchy

• impl(aggregator+orchestrator): Next-up view - what would be picked up next, in ladder order, across projects #1158 Next-up data contract (ladder order, cross-project)
• impl(aggregator): queue visibility views - in-progress, blocked, recently-done as first-class cross-project lists #1160 queue views: in-progress / blocked / recently-done
• impl(aggregator): one reconciled in-progress list - board lane, Project status, and session claims currently disagree #1162 one reconciled in-progress list
• impl(aggregator): epic-level rollups - pool has 16 Kind=Epic items but every page renders flat lists #1164 epic-level rollups (16 Kind=Epic items)
• impl(fleet/project): adopt native GitHub sub-issues + blocked-by for epic children (16 epics) instead of text-derived links #1166 native sub-issues + blocked-by adoption
• impl(aggregator): /roadmap drill-down - roadmap row expands to its epic, progress, and issues #1168 /roadmap drill-down to epic + progress

Children — quotas & usage

• fix(aggregator): quota pools carry no visible per-pool read age - stale manual reads look live #1170 per-pool read age / stale badges
• impl(aggregator): /quota covers 8 SaaS services; the tracked estate is dozens - list the uninstrumented with reasons #1172 quota coverage: the uninstrumented estate listed
• impl(aggregator): quota page shows numbers, not actions - surface expiring/cancelled-pool actions into /today + /attention #1174 quota actions surfaced into /today + /attention

Children — contracts, performance, access, hygiene

• impl(pm-stack): contract test suite for publisher -> collector -> aggregator schemas (the #1110 failure class) #1176 publisher->collector->aggregator contract tests
• impl(collector): feed sections fail silently (omitted on error) - publish per-section status in feed meta #1178 collector per-section status in feed meta
• fix(aggregator): heavy pages + per-request live fan-out - /projects 109KB, /sessions 100KB, / renders in ~0.7s, /today ~1.8s #1179 page weight / fan-out performance
• fix(aggregator): per-page mobile audit - tables, timelines, chips at 390px across all 10 pages #1180 per-page mobile audit (390px)
• fix(pm-stack): ready-for-agent binds 0.0.0.0:6056 while the rest of the stack binds loopback+tailnet - align bind scopes #1181 ready-for-agent bind scope alignment
• design(pm-stack): one public URL for the dashboard behind Cloudflare Access (settled access model, not yet implemented) #1182 one public URL behind Cloudflare Access (design; Drew's go required)
• chore(pm-stack): tracker hygiene - stale design issues #851/#852 still open though shipped; #781 predates the stack review #1183 tracker hygiene (design(aggregator): quota + throughput view — pooled quotas, resets, burn rates, model-choice support #851/design(aggregator): 'needs you' attention view — session state taxonomy, machine-first #852 zombies, pm-stack: Pad/Grist compose files hard-code one host's tailnet IP (100.99.113.103), breaking the documented move-hosts procedure #781)

Related programs (distinct epics, don't blur)

• epic(dashboard): per-project status pages for all projects — generalize /program beyond pm-dashboard #1129 — multi-project documentation pages (/program hub + per-project). This epic is the stack's own coherence/UX/health; epic(dashboard): per-project status pages for all projects — generalize /program beyond pm-dashboard #1129 is project documentation content.
• impl(aggregator+collector): complexity-weighted project visualization (rollup, hands queue, blocked weight, depth graph) #1122 — weighted visualization (consumes this epic's epic-rollup data work in impl(aggregator): epic-level rollups - pool has 16 Kind=Epic items but every page renders flat lists #1164/impl(fleet/project): adopt native GitHub sub-issues + blocked-by for epic children (16 epics) instead of text-derived links #1166)
• impl(fleet): work-picking jobs consult the orchestrator published priorities before choosing work #911 — work-picking jobs consult the orchestrator (consumes impl(aggregator+orchestrator): Next-up view - what would be picked up next, in ladder order, across projects #1158's published next-up)

Roadmap representation

This issue carries epic + now labels so the orchestrator's label-gated auto-add (#910 rules) picks it into org Project #4 as Kind=Epic; the Area must be pm-dashboard (set by whoever reconciles first pass if auto-add doesn't carry Area — tracked as part of #1143's taxonomy work). Grist "Program roadmap" rows 1-5 cover the ORIGINAL PM phases (complete); a row for this coherence phase is requested in #1144's reconciliation scope rather than written ad hoc.

Done when

All children closed: Drew opens one URL and sees every project, fresh data with honest ages, working triage, a true queue picture, epic rollups, and a health view — and staleness anywhere in the stack announces itself instead of decaying silently.

Acceptance criteria

• A single URL (mobile or desktop) shows all projects with quotas & usage, page health, and the issues that need Drew's attention — with no silent-zero or fixture-backed state presented as live.
• Triage is possible from that URL (remediate / respond / basic triage), not read-only, through the same audited action path as the rest of the stack.
• Queued + in-progress work is legible on the dashboard: what would be picked up next, what is being worked on, what is blocked, and what was done recently.
• Hierarchy renders with rollups (roadmap item > epic > issue), and the service/port topology (including Kaneo vs the rest) is documented on the dashboard.
• Every child issue listed in this epic is either closed with evidence or explicitly carried as a named remainder; the epic closes only when Drew's verdict — easy to navigate, coherent, fresh — can be answered yes from the dashboard itself.