Home / Work / Case study

Beanfield · Sep 2026 to now

IPTV knowledge vault

One engineer, one platform, dozens of systems from three vendors. I built a linked knowledge vault of the entire IPTV platform: every system, symptom, log, incident and ruling on its own page, every fact pointing at the file it came from, and a weekly check that flags any page whose source has moved on.

My roleDesigner, author and maintainer
Size373 pages, about 677,000 words
Evidence2,060 cited sources on 368 pages
Linked6,716 page-to-page links

Why it exists

The IPTV platform spans set-top boxes, home Wi-Fi, fibre, the access network, the headend, recorders, storage, middleware, billing and three vendors' portals. The knowledge lived in vendor PDFs, tickets, chat threads, logs and people's heads. A customer ticket could need five systems and a dozen facts before the first useful step.

Before
  • Facts scattered across vendor guides, tickets, mirrors and chat
  • No single map from a customer symptom to the system at fault
  • Documents drifted out of date silently
  • New staff had no starting point
After
  • One linked vault, one page per system, symptom and procedure
  • Symptom decision trees and fault-path maps from the TV to the vendor
  • Every page names its sources; a weekly check flags drift
  • Read hot, then index, then the page in front of you

Design decisions

  1. The project stays the source of truth. The vault is a reading layer. A fact is fixed in the project file first, then the page is refreshed. Every page carries a project_sources list in its front matter.
  2. One page per thing. Each system page has the same shape: access, what it is used for, traps, links and sources. Big systems (the monitoring host, the middleware, telemetry, billing, the service catalogue) are hubs with component and tool pages beneath them.
  3. Start from the customer. The CX Troubleshooting area opens with symptoms (freezing, no channels, recording missing, wrong guide) and walks down decision trees to the layer at fault.
  4. A hot cache. hot.md holds the newest context in a few screens, so anyone (or any tool) reads the latest truth before the deep pages.
  5. Plain text, local first, under version control. Markdown files in git: diffable, greppable, portable, no lock-in, readable without the app.
  6. Two stores, two names. This vault is the IPTV Obsidian Vault Wiki; the published, team-facing store is the ServiceNow IPTV Knowledge Base. Naming them apart stopped every "which KB?" question.

What is inside

AreaPagesWhat it holds
Systems118Every platform system, hubs with component and tool pages, the server map
CX Troubleshooting7618 symptom decision trees, 21 anonymised case studies, flows, patterns, ticket taxonomy
How-tos22Step-by-step procedures, one per job
Network21IPTV path, BNG sessions, multicast and QoS, rate caps, edges
STB20Box models, firmware, boot, freezing deep dive, on-box commands
nPVR16Cloud recording end to end: booking, scheduler, recorder, storage, offload, quota
Home Network15Mesh Wi-Fi, routers, ONTs, GPON and XGS-PON
Vendors and People14Who supplies and decides what, and how to engage them
EPG13Guide structure, identifiers, feed, import job, guide versus picture
Decisions12169 current rulings by topic, newest first
Incidents119 platform outages, each with cause and fix, plus the Outage Playbook
Logs10Log atlas, retention table, grep recipes, the client API call catalogue
Glossary10Hundreds of terms and acronyms (353 at the 2 Oct count), each pointing to its page
Projects10Wall, outage bot, baseline, remote box access, tunnel tool and more

Twelve live database views (Obsidian Bases) sit on top: one per area, an Open Items view and a Wiki Freshness view that shows days since each page's update and its source count.

Fault-path maps

Four canvases turn the vault into maps: the IPTV Customer Path, and the fault paths for freezing, recordings (nPVR) and the guide (EPG). Each box is a real vault page; colour shows the layer or stage. In October 2026 I redrew them with straight right-angled links in lanes so a path reads left to right.

IPTV Customer Path canvas: customer symptoms on the left linked through home, access network, headend, middleware, data and tools, and billing
IPTV Customer Path: symptoms on the left, then the six layers from the home to billing (118 boxes, 84 links). Click to open full size.
nPVR Fault Path canvas: six-hop recording chain, decision trees, verdicts, detail pages and case studies
nPVR Fault Path: the six-hop recording chain, two decision trees, 21 verdicts and their detail pages (297 boxes, 280 links). Click to open full size.
CanvasBoxesLinks
IPTV Customer Path11884
nPVR Fault Path297280
EPG Fault Path187163
Freeze Fault Path168150

Staying true: the freshness check

A knowledge base is only useful while it is right. I wrote a read-only checker that runs weekly and reports three things: dead sources (a cited file moved or was removed), stale pages (a source changed more than 7 days after the page was last updated) and lint findings. It never edits a page; refreshing is a separate, reviewed step.

Refresh roundPages checkedResult
2 Oct 2026109Re-checked against the project; 57 dead sources repointed
9 Oct 2026114Two passes of 57 pages matched to project sources
10 Oct 20265822 changed, 36 verified; 0 stale, 0 dead
19 Oct 2026205154 changed, 51 verified, 65 new decision rows, a new incident page; 0 stale, 0 dead, 0 lint
The rule that makes it work: a page is never "probably right". It is either matched to its sources this week, or the checker names it.

Timeline

  1. 23 Sep 2026: vault set up; Systems, major nodes, Vendors and People, Incidents, How-tos, Decisions, Glossary, nPVR, EPG, Network, Home Network, STB and Logs written the same day.
  2. 24 to 27 Sep: recording symptom trees, the IPTV Customer Path canvas, the customer troubleshooting hub, gap passes on freezing, data stores and vendor facts.
  3. 2 Oct: Bases (freshness, per area, open items) and the Freeze, nPVR and EPG fault-path canvases; 109-page refresh.
  4. 9 to 10 Oct: two refresh rounds; ServiceNow capability map; the two-store naming ruling.
  5. 19 Oct: full refresh of 205 pages, canvases redrawn, monitoring-host GUI and provisioning pages; a cleaned snapshot shared with the senior director.
  6. Since: weekly freshness checks, updates as incidents and rulings land; 87 commits in all.

How it is used

  • Every investigation starts here: the hot cache, then the page for the system or symptom in front of me, then its sources.
  • The project's routing command points at the vault by name, so a lookup for "what is X" lands on the right page.
  • Shared with leadership as a cleaned snapshot (no history, raw sources or personal area), as the platform's handover record.
  • Feeds the published ServiceNow IPTV Knowledge Base: the vault is where facts are proven, the KB is where the team reads them.

The four roles in this project

  • Engineer: every fact traced to logs, live reads or vendor documents; 21 real case studies and 9 incident write-ups.
  • Designer: one page shape for every system, customer-first entry points, colour-coded maps that read left to right.
  • System engineer: the whole platform modelled as layers and hops, from the TV to billing, with the links between them.
  • Innovator: documentation that checks itself: source lists in the page, a weekly drift report, database views over the pages.

Numbers

Pages373 in 13 areas
Wordsabout 677,000
Cited project sources2,060 on 368 pages
Page-to-page links6,716
Canvases4 (770 boxes, 677 links)
Database views12
Decision rows169
Commits87 since 23 Sep 2026

Stack

ObsidianMarkdown with YAML front matterJSON CanvasObsidian BasesgitPython freshness and lint checksClaim and source ledgers