Trucible · Knowledge OS
Wiki and Explorer modes for an enterprise context management system.
Trucible started with a direction from Sid: "this is a context management system, not a document management system." That sentence rewrote the product. We weren't building Notion. We were building the memory layer of the organisation.
The dual-mode design (Wiki and Explorer) emerged from that distinction. Wiki mode presents Trucible-generated articles, synthesised from connected sources, maintained by Tango, readable by humans. Explorer mode is the database view: folder trees, graph visualisations, list tables. The same knowledge, two interfaces for two mental models.
The hardest problem wasn't the interface, it was trust. An AI-synthesised article is only useful if the reader believes it, and belief needs evidence, not just fluent prose.
Hover a tier. Each one exists because a percentage score can't tell a Demand Manager whether to trust what Tango just told them.
The Context-Starved Operator
Not a project manager. Not a delivery lead. A governance role, accountable for asking "are we doing the right things?" not "are we doing things right?"
Get enough organisational context to act credibly in the next meeting, without reading six months of Slack and email history first.
Re-opening a decision that was already made and reversed, because nobody could find the thread that explained why.
An answer with a source attached, in under a minute, that they can defend if a colleague pushes back on it.
Will trust Tango's synthesis exactly as far as its citations hold up under a click-through, and no further.
01 · Replace percentage scores with named tiers
A percentage answers "how much" when the real question is "can I trust this." Naming the tiers by their implication for reliability turns a number nobody could interpret into a decision aid.
02 · Wiki mode as a written destination, not a search result
Enterprise search optimises for retrieval, not comprehension. A synthesised article is faster to read than three source documents stitched together by hand, and it's the thing you actually forward to a teammate.
03 · Explorer mode kept, not replaced
Some jobs need the synthesis; some jobs need to audit the raw structure it came from. Removing Explorer mode would have solved a reading problem by breaking a verification one.
Context, not documents
Trucible doesn't store files. It maintains a living, Tango-synthesised understanding of what the organisation knows, with the source documents underneath as evidence.
Two mental models, one dataset
Wiki mode is for reading a synthesised answer; Explorer mode is for browsing the raw structure. Same knowledge graph, never forked between them.
Provenance is not optional
Every Wiki article traces back to its connected sources. An unsupported claim is treated as a bug in the synthesis, not an acceptable shortcut.
Freshness decays visibly
A context tier isn't set once and forgotten. Sources age and connections go stale, and the tier badge reflects that in real time, not just at creation.
Qualitative context strength tiers replaced percentage scores.
A 67% context strength score is meaningless. What does 67% mean? Do I need to add one document or fifty? The tiers (Deep Root, Warming Up, Seedling, Blank Slate) communicate actionability: what state is this knowledge in, and what does that mean for how much I can trust Tango's outputs from it.
Trucible Trivia as a preview panel before commit.
Before a user commits to opening a full article, a preview panel surfaces the three most load-bearing facts, borrowed from search-engine result snippets. In an enterprise context, time is the scarcest resource, so the preview has to earn the click rather than assume it.
"I joined mid-project and Trucible got me to "not embarrassing myself on the client call" in twenty minutes, not two weeks."
"Deep Root or Seedling tells me in one glance whether I can quote this to a client or whether I need to go verify it first."
Learnings that apply beyond this project
Name the trust, don't just score it
Any confidence metric shown to a user should answer "what should I do with this," not just "how much." A number without a stated implication gets either ignored or over-trusted.
Don't force one interface onto two jobs
Reading-for-understanding and browsing-for-verification are different tasks with different correct interfaces. Serving both from one view under-serves both.
It would have been simpler to ship Wiki mode only. Keeping Explorer alive for auditors and analysts turned out to be the reason compliance-heavy teams adopted Trucible at all.
The percentage-score version shipped first and had to be pulled back after users told us it was actively misleading them. Testing the tier concept before the percentage version ever shipped would have avoided a full rework cycle.