UNI Universal Natural Intelligence

Coverage

Every subsystem, every runnable command, every document type and every published page is either covered or explicitly excluded with a reason. Covered plus declared-excluded is the whole; anything that is neither fails the build by name.

This is a gate, not a claim. The site once published 304correct documents and called itself a user guide — every page reachable, none of it findable, the largest section 85 documents in one alphabetical list. “100% coverage” means nothing without a denominator, so two of the four denominators below are taken from the world rather than from a list somebody wrote.
The scope of “discovered”, stated exactly — because it was overstated here until 2026-08-01. This box used to say “add a server to the estate and this table goes red naming it.” That is not true of the estate; it is true of four directories.

Discovery walks lib/mix/tasks, viewer and scripts in the colony repository, plus the package manifests of two others. A forensic review found a real server outside those roots — production/guest/pubgate/server.cjs, which binds a port on all interfaces — that is in neither the covered list nor the excluded one, and 52 runnable scripts under runs/ against the 16 under scripts/that form the denominator. The table below is honest about those four roots and says nothing about the rest of the estate.

The mutation that was supposed to prove the claim could not have: it appended a fabricated entry to the discovery result rather than putting a file on disk, so it proved the axis reacts to a longer list and not that discovery finds a new file. It now writes a real file into a real walk root. A proof-of-bite that cannot bite is the defect this whole site is organised against, and it was sitting inside the check that says so.

The four axes

AxisDenominatorTotalCoveredExcludedGap
Subsystemsdeclared8800
Entry pointsdiscovered10480240
Document typesdeclared8800
Published pagesdiscovered52452400
All axes644620240

Discovered means the denominator is measured from the source repositories or from the ingest, so it grows on its own. Declared means the list is editorial — a shorter list would score the same, and saying so is the only defence against that.

The ratchet: 100% cannot collapse and cannot reduce

A percentage is the most collapsible statistic there is. Covered ÷ total stays at exactly 1.0 under two completely different disasters, and the table above is blind to both because in each case the sum still closes and the gap is still zero:

Collapse — the denominator shrinks. A repository moves, a discovery rule is tidied away, a corpus stops resolving. 100% of a smaller world, reported identically.

Reduction — work moves from covered into excluded. Every exclusion carries a reason and the arithmetic still closes, so the gate is satisfied while the guides document less with each change.

So every measured quantity is also held against a committed floor, and the floor is compared against its own previous version in version control. Lowering one is allowed; lowering one silently is not — it requires a recorded amendment saying what moved and why. A floor that can be edited down in the same change that breaches it is not a floor.

81
committed bounds
81
held
0
breached
65
exactly at the line

Growth is free and needs no ceremony — this estate adds servers and documents constantly, and a gate that fights ordinary work gets switched off. Only going backwards costs anything.

Amendments

Every time a floor was deliberately lowered, with the reason given at the time.

DateWhatWhy
2026-08-01Lowered floors pages_total (292 -> 291) and corpora_pages.workbench (2 -> 1).A forensic pass held all 292 published pages against the commit each citation names and found ONE whose published bytes were an uncommitted working-tree edit published under a commit sha that does not contain them (workbench/living-science-walkthrough). The ingest now fences any document whose own bytes are uncommitted, per path, so it cannot assert a provenance it does not have. The page count falls by one because a page that was making a false claim stopped being published. This is a REDUCTION IN COUNT AND AN INCREASE IN CORRECTNESS, which is why the floor is lowered deliberately rather than the fence being weakened to preserve a number. The remaining four fenced documents return the moment the workbench worktree is committed.
2026-08-02Lowered floor article_chars.run-it (8678 -> 8667), a drop of 11 characters, as part of re-pinning the baseline for the 493-page expansion.No prose was removed. The run-it page body is byte-identical to its previous version. The 11-character drop is an earlier fix catching up: a dead internal link to a flagellar-motor route that does not exist was corrected to the live route, removing the 11 characters of the word laboratory. The floor was pinned before that fix and is only now re-emitted. This is a reduction in count and an increase in correctness -- a working link is shorter than a broken one -- so the floor article_chars.run-it is lowered deliberately and recorded here rather than the page being padded to preserve a number.
All 81 bounds, and the distance to each
BoundKindNowLimitMargin
entry points discovered · mix-taskfloor880
entry points discovered · serverfloor880
entry points discovered · powershellfloor42420
entry points discovered · elixir-scriptfloor16160
entry points discovered · npm-scriptfloor30300
entry points discovered · totalfloor1041040
pages published · encyclopediafloor45396
pages published · cookbookfloor34340
pages published · control-planefloor33330
pages published · flagellumfloor14140
pages published · workbenchfloor110
pages published · minecraftfloor87870
pages published · evidencefloor1051050
pages published · gpt-packfloor220
pages published · builderfloor550
pages published · cookbook-repofloor660
pages published · hierarchical-aiffloor64640
pages published · auditsfloor330
pages published · flagellum-repofloor330
pages published · science-labfloor17170
pages published · filmsfloor880
pages published · broadcastfloor56560
pages published · colony-builderfloor770
pages published · estatefloor990
pages published · lexiconfloor880
pages published · constantsfloor770
pages published · totalfloor52450816
pages in route · startfloor990
pages in route · install-runfloor660
pages in route · methodfloor11110
pages in route · colonyfloor440
pages in route · specsfloor17170
pages in route · broadcastfloor57570
pages in route · operatorfloor880
pages in route · architecturefloor33330
pages in route · flagellumfloor990
pages in route · workbenchfloor220
pages in route · reportsfloor23230
pages in route · lab-teamfloor770
pages in route · encyclopedia-methodfloor16160
pages in route · encyclopedia-naturefloor12120
pages in route · encyclopedia-stackfloor11110
pages in route · cookbook-kitchenfloor880
pages in route · cookbook-stackfloor14140
pages in route · cookbook-naturefloor12120
pages in route · plansfloor15150
pages in route · evidence-receiptsfloor62620
pages in route · evidence-preregfloor37370
pages in route · evidence-handoffsfloor12120
pages in route · pack-and-readerfloor770
pages in route · builderfloor12120
pages in route · haif-programfloor21210
pages in route · haif-preregfloor880
pages in route · haif-resultsfloor35350
pages in route · science-labfloor17170
pages in route · filmsfloor880
pages in route · lexiconfloor880
pages in route · constantsfloor770
RUNNABLE markersfloor16160
entry points documentedfloor80800
entry points excludedceiling24240
subsystemsfloor880
document typesfloor880
articlesfloor13130
resolved citationsfloor59590
quoted source blocksfloor15150
curated routesfloor33312
pages in no routeceiling000
substance · start-herefloor38062854952
substance · the-colonyfloor26571992665
substance · the-producerfloor28362127709
substance · the-control-planefloor33242493831
substance · the-brainfloor401130081003
substance · the-operator-planefloor35822686896
substance · the-flagellar-motorfloor549241191373
substance · the-broadcast-suitefloor1265994943165
substance · quick-startfloor36452733912
substance · installfloor749256191873
substance · run-itfloor866765002167
substance · maintenancefloor819061422048
substance · how-tofloor517038771293

A margin of zero is not a failure — it means the quantity is sitting exactly on its floor, which is where a freshly-measured baseline puts most things. It does mean the next reduction there fails the build.

Subsystems, in two layers each

Understanding without the ability to run something does not transfer; the ability to run something without understanding it does not survive the first problem. So each subsystem is documented twice.

SubsystemUnderstand itRun it yourself
The colonythe-colonyrun-it
The Producerthe-producerrun-it
The active-inference brainthe-brainrun-it
The scientific control planethe-control-planerun-it
The operator planethe-operator-planerun-it
The broadcast suitethe-broadcast-suitethe-broadcast-suite
The flagellar-motor laboratorythe-flagellar-motorrun-it
The math workbenchthe-flagellar-motorrun-it

The eight document types

TypeSatisfied byReachable from
Quick startquick-start/articles/
Install guideinstall/articles/
User guidesrun-it, the-broadcast-suite/articles/
Manualsthe-operator-plane, the-broadcast-suite/articles/
How-to guideshow-to/articles/
Maintenance guidesmaintenance/articles/
Specsthe specs route, at least 8 pages/wiki/
Code citationsat least 40 resolved source citations/articles/

Reachability is checked against the navigation component that actually renders the site, not against a copy of it kept here. A declared copy could certify a route the site no longer has.

What is deliberately excluded, and why

24 runnable entry points are documented nowhere on this site. Each is named with its reason, because an exclusion nobody can see is an omission with better manners.

powershell -File viewer/hud/_install_elevated_wrapper.ps1

Part of the native HUD's build-and-package chain: compiling the widget executable, fetching a service-wrapper binary, and an elevation shim. These build the artifact rather than run or maintain the system, and a reader who wants the HUD installs the built service rather than compiling it. Named here so their existence is not hidden.

powershell -File viewer/hud/build_exe.ps1

Part of the native HUD's build-and-package chain: compiling the widget executable, fetching a service-wrapper binary, and an elevation shim. These build the artifact rather than run or maintain the system, and a reader who wants the HUD installs the built service rather than compiling it. Named here so their existence is not hidden.

powershell -File viewer/hud/download_nssm.ps1

Part of the native HUD's build-and-package chain: compiling the widget executable, fetching a service-wrapper binary, and an elevation shim. These build the artifact rather than run or maintain the system, and a reader who wants the HUD installs the built service rather than compiling it. Named here so their existence is not hidden.

powershell -File viewer/hud/native/_cert_and_sign_elevated.ps1

Elevated steps of the native HUD's code-signing and service-registration chain, invoked while developing the widget itself. Documenting a code-signing procedure for a certificate the reader cannot possess would be a recipe that ends at a wall. The two elevated scripts a reader DOES need — the launcher install and the service swap — are covered in the install guide, and the two restart drills are covered in the maintenance guide.

powershell -File viewer/hud/native/_sign_and_reinstall_elevated.ps1

Elevated steps of the native HUD's code-signing and service-registration chain, invoked while developing the widget itself. Documenting a code-signing procedure for a certificate the reader cannot possess would be a recipe that ends at a wall. The two elevated scripts a reader DOES need — the launcher install and the service swap — are covered in the install guide, and the two restart drills are covered in the maintenance guide.

powershell -File viewer/hud/native/_sign_widget_elevated.ps1

Elevated steps of the native HUD's code-signing and service-registration chain, invoked while developing the widget itself. Documenting a code-signing procedure for a certificate the reader cannot possess would be a recipe that ends at a wall. The two elevated scripts a reader DOES need — the launcher install and the service swap — are covered in the install guide, and the two restart drills are covered in the maintenance guide.

powershell -File viewer/hud/native/_stop_service_elevated.ps1

Elevated steps of the native HUD's code-signing and service-registration chain, invoked while developing the widget itself. Documenting a code-signing procedure for a certificate the reader cannot possess would be a recipe that ends at a wall. The two elevated scripts a reader DOES need — the launcher install and the service swap — are covered in the install guide, and the two restart drills are covered in the maintenance guide.

powershell -File viewer/hud/native/_update_service_binaries_elevated.ps1

Elevated steps of the native HUD's code-signing and service-registration chain, invoked while developing the widget itself. Documenting a code-signing procedure for a certificate the reader cannot possess would be a recipe that ends at a wall. The two elevated scripts a reader DOES need — the launcher install and the service swap — are covered in the install guide, and the two restart drills are covered in the maintenance guide.

powershell -File viewer/hud/native/_urlacl_and_networkservice_elevated.ps1

Elevated steps of the native HUD's code-signing and service-registration chain, invoked while developing the widget itself. Documenting a code-signing procedure for a certificate the reader cannot possess would be a recipe that ends at a wall. The two elevated scripts a reader DOES need — the launcher install and the service swap — are covered in the install guide, and the two restart drills are covered in the maintenance guide.

npm run dev

The math workbench is a git WORKTREE of the flagellum repository on a different branch, so its package file is the same package file. Measured 2026-08-01: 14 of its 15 scripts are byte-identical to the flagellum ones already documented, and the 15th differs only by adding one test file to the same test command. Documenting them twice would pad the coverage count without telling a reader anything they do not already have. The workbench's own route is covered in the run-it guide.

npm run build

The math workbench is a git WORKTREE of the flagellum repository on a different branch, so its package file is the same package file. Measured 2026-08-01: 14 of its 15 scripts are byte-identical to the flagellum ones already documented, and the 15th differs only by adding one test file to the same test command. Documenting them twice would pad the coverage count without telling a reader anything they do not already have. The workbench's own route is covered in the run-it guide.

npm run start

The math workbench is a git WORKTREE of the flagellum repository on a different branch, so its package file is the same package file. Measured 2026-08-01: 14 of its 15 scripts are byte-identical to the flagellum ones already documented, and the 15th differs only by adding one test file to the same test command. Documenting them twice would pad the coverage count without telling a reader anything they do not already have. The workbench's own route is covered in the run-it guide.

npm run test

The math workbench is a git WORKTREE of the flagellum repository on a different branch, so its package file is the same package file. Measured 2026-08-01: 14 of its 15 scripts are byte-identical to the flagellum ones already documented, and the 15th differs only by adding one test file to the same test command. Documenting them twice would pad the coverage count without telling a reader anything they do not already have. The workbench's own route is covered in the run-it guide.

npm run experiment:ingest

The math workbench is a git WORKTREE of the flagellum repository on a different branch, so its package file is the same package file. Measured 2026-08-01: 14 of its 15 scripts are byte-identical to the flagellum ones already documented, and the 15th differs only by adding one test file to the same test command. Documenting them twice would pad the coverage count without telling a reader anything they do not already have. The workbench's own route is covered in the run-it guide.

npm run experiment:run

The math workbench is a git WORKTREE of the flagellum repository on a different branch, so its package file is the same package file. Measured 2026-08-01: 14 of its 15 scripts are byte-identical to the flagellum ones already documented, and the 15th differs only by adding one test file to the same test command. Documenting them twice would pad the coverage count without telling a reader anything they do not already have. The workbench's own route is covered in the run-it guide.

npm run experiment:verify

The math workbench is a git WORKTREE of the flagellum repository on a different branch, so its package file is the same package file. Measured 2026-08-01: 14 of its 15 scripts are byte-identical to the flagellum ones already documented, and the 15th differs only by adding one test file to the same test command. Documenting them twice would pad the coverage count without telling a reader anything they do not already have. The workbench's own route is covered in the run-it guide.

npm run science:run

The math workbench is a git WORKTREE of the flagellum repository on a different branch, so its package file is the same package file. Measured 2026-08-01: 14 of its 15 scripts are byte-identical to the flagellum ones already documented, and the 15th differs only by adding one test file to the same test command. Documenting them twice would pad the coverage count without telling a reader anything they do not already have. The workbench's own route is covered in the run-it guide.

npm run science:verify

The math workbench is a git WORKTREE of the flagellum repository on a different branch, so its package file is the same package file. Measured 2026-08-01: 14 of its 15 scripts are byte-identical to the flagellum ones already documented, and the 15th differs only by adding one test file to the same test command. Documenting them twice would pad the coverage count without telling a reader anything they do not already have. The workbench's own route is covered in the run-it guide.

npm run cross-study:ingest

The math workbench is a git WORKTREE of the flagellum repository on a different branch, so its package file is the same package file. Measured 2026-08-01: 14 of its 15 scripts are byte-identical to the flagellum ones already documented, and the 15th differs only by adding one test file to the same test command. Documenting them twice would pad the coverage count without telling a reader anything they do not already have. The workbench's own route is covered in the run-it guide.

npm run cross-study:verify-raw

The math workbench is a git WORKTREE of the flagellum repository on a different branch, so its package file is the same package file. Measured 2026-08-01: 14 of its 15 scripts are byte-identical to the flagellum ones already documented, and the 15th differs only by adding one test file to the same test command. Documenting them twice would pad the coverage count without telling a reader anything they do not already have. The workbench's own route is covered in the run-it guide.

npm run cross-study:run

The math workbench is a git WORKTREE of the flagellum repository on a different branch, so its package file is the same package file. Measured 2026-08-01: 14 of its 15 scripts are byte-identical to the flagellum ones already documented, and the 15th differs only by adding one test file to the same test command. Documenting them twice would pad the coverage count without telling a reader anything they do not already have. The workbench's own route is covered in the run-it guide.

npm run cross-study:verify

The math workbench is a git WORKTREE of the flagellum repository on a different branch, so its package file is the same package file. Measured 2026-08-01: 14 of its 15 scripts are byte-identical to the flagellum ones already documented, and the 15th differs only by adding one test file to the same test command. Documenting them twice would pad the coverage count without telling a reader anything they do not already have. The workbench's own route is covered in the run-it guide.

npm run lint

The math workbench is a git WORKTREE of the flagellum repository on a different branch, so its package file is the same package file. Measured 2026-08-01: 14 of its 15 scripts are byte-identical to the flagellum ones already documented, and the 15th differs only by adding one test file to the same test command. Documenting them twice would pad the coverage count without telling a reader anything they do not already have. The workbench's own route is covered in the run-it guide.

npm run db:generate

The math workbench is a git WORKTREE of the flagellum repository on a different branch, so its package file is the same package file. Measured 2026-08-01: 14 of its 15 scripts are byte-identical to the flagellum ones already documented, and the 15th differs only by adding one test file to the same test command. Documenting them twice would pad the coverage count without telling a reader anything they do not already have. The workbench's own route is covered in the run-it guide.

Routes through the wiki

33 routes covering 524 published pages. Every page belongs to a route; there is no default bucket, and a document that matches none of them fails the build rather than landing nowhere.

What this table cannot tell you

It cannot tell you a guide is followable.Coverage proves nothing is unaccounted for. It says nothing about whether the quick start works on a machine that is not the author’s, whether the ordering makes sense to somebody meeting this for the first time, or whether an explanation actually explains. Only a stranger following it can answer that, and no gate substitutes for one.

It also cannot see sensitive narrative, which is the other thing no scanner here can do. Both limits are printed rather than left to be discovered.