verify-shell-scroll.mjs was written and negative-tested, then left with no
package.json entry and no importer, so nothing could ever run it again. It
also defaulted to localhost:8899 -- an ad-hoc port from an earlier session --
so it failed with ERR_CONNECTION_REFUSED for anyone else and read as a broken
gate rather than a missing server. Now check:shell-scroll, defaulting to
BASE_URL or 127.0.0.1:4173 like every other browser gate, and listed in
verify-browser-all's GATES array.
check:no-orphan-gates fails any gate-shaped file in internal/ that nothing can
invoke. The repo carded this once already; it recurs because writing a gate
and registering one are separate acts and only the first feels like work.
Not every unreferenced file is a lost gate, so the rule is not "everything
must be wired": theme-flash-probe is a dated diagnostic for two fixed bugs,
worth keeping as a reproduction recipe. It now carries an @one-shot tag saying
so, which makes the author state which kind of thing they wrote.
Two self-referential defects in the first version of that gate, both from
reading evidence out of documentation:
- it cited theme-flash-probe.mjs in its own header as an example, then grepped
for importers, found itself, and declared the orphan reachable. The gate
written to catch orphans made an orphan pass by mentioning it. A gate must
never be its own witness, so it now excludes its own file.
- it matched @one-shot anywhere in prose, including its own paragraph
explaining the tag, and reported "1 tagged" when nothing had been tagged.
The tag must now be a marker at line start.
Negative-tested: removing check:shell-scroll from package.json fails by name.
The scroll fix itself is NOT done. contain:layout and .vtl-host overflow-y are
confirmed in the built bundle, and a real wheel gesture still moves the grid
0px with 2586px mounted and unreachable. The gate is what caught that; the
diff looked correct.
bun run appends arguments to the command string, so a script ending in an
inline # comment swallows them. Proven side by side:
"node argtest.mjs" bun run _t1 --self-test -> ["--self-test"]
"node argtest.mjs # a comment" bun run _t2 --self-test -> []
No error, no warning: the script runs in its default mode and exits 0.
Four scripts advertised a flag that could never arrive, and two of them are
--self-test. A --self-test IS the proof that a gate can fail, so running
check:facet-cascade --self-test printed the ordinary PASS and exited 0 --
which reads as "the negative test passed" when the negative test never ran.
The gate itself was fine: invoking node directly reverted the cascade and
failed 13 assertions correctly, including the real symptom (switching to
another language family leaves the map empty because the orphan still
filters). Only the invocation was broken, so the mechanism that certifies a
gate can fail was a gate certifying itself.
Each flag now gets a dedicated entry that hardcodes it, which also makes the
mode discoverable in bun run output instead of documented only in prose
nobody greps.
check:script-flags fails any script whose comment advertises a flag the
command cannot receive. Negative-tested: adding such a comment back fails by
name with exit 1.
Three docs still told a reader to start a server on every interface: the
reference recipe for the share round-trip, the quoted scripts block, and
app-wrappers.md. Agents followed those recipes faithfully, which is how twenty
publicly bound http.server processes accumulated on this box. All three now
bind 127.0.0.1. check:quoted-scripts caught the second one on its own.
Also adds the gate-authoring constraint that just broke a build. Plugin and
library source is written for Vite, not Node, and fails two ways: the @map
alias is unresolvable (hit first), and import.meta.glob is undefined in 18
files including 11 of the 20 plugin index.js. A new check:facet-cascade
imported the dashboard config, reached library/colors/_registry.js
transitively, and threw -- while already wired into build:unlocked, so the
build failed for a reason unrelated to the code being built.
The counts in that section were measured, not estimated. A first draft said
"every plugin index.js" and the real number is 11 of 20; it also blamed the
glob when the first error is actually the alias.
States the two rules that follow: build:unlocked is only for gates that run
clean in plain Node with no browser and no network, and a gate gets run before
it gets wired in -- a broken gate in the build chain is worse than no gate,
because it goes red for the wrong reason and trains people to skip the step.
Records the other box constraints that have each produced a wrong answer:
Node 20 (no fs.glob), bfs find rejecting relative -newermt, no hardware WebGL,
and bunx not existing.
serve used --bind 0.0.0.0, and every browser gate starts it. A live audit of
this box found nineteen publicly bound http.server processes on 4820-4970 plus
one on 8899, all rooted at public/, with curl to the box's public IP returning
200. Nobody noticed, because a listening socket produces no output and no
failure: the gate it served passed and the process outlived it.
The standing rule "keep the box closed" lived in prose. The behaviour lived in
a flag. Prose cannot constrain a flag, so the rule was never violated by a
decision -- only by a default nobody re-read.
Loopback is sufficient for the whole suite: verify-browser-all,
check:a11y-button-names, check:nav-reachable and smoke:share-all already
default to 127.0.0.1. The wide bind bought nothing.
check:no-public-binds fails any script with an explicit all-interface bind, a
bare vite --host, or an http.server with no --bind at all (its default is also
all interfaces, just less visible in review). It reads the repo rather than the
machine's live sockets, because a socket check would pass or fail on whatever
happens to be running and could not be fixed by editing anything here.
Negative-tested: restoring --bind 0.0.0.0 fails by name with exit 1.
The menu is position:fixed with viewport coordinates. That resolves against the
viewport only while nothing establishes a containing block for fixed
descendants. VerticalTabsLayout -- the cap-5 shell that hosts every plugin as a
tab -- sets contain:layout paint on .vtl-content, inside .vtl-main{left:256px}.
Measured at 1600x900, same component, two hosts:
standalone research-dashboard ...... dLeft=0 dTop=4 six of six
inside plugin-studio dashboard tab . dLeft=256 dTop=111 every one
256 is the rail width and 111 the header height. Reported as the menu landing
"in a random spot further down to the right"; it was never random.
An earlier probe measured six triggers at four viewport widths and passed, so
the card was closed while the bug was still there. Every number was accurate.
They were answers about a surface the user never looks at -- widening across
more widths could not have found it, because the variable was the host.
Fix measures the containing block instead of assuming it: a fixed probe at 0,0
inserted in the menu's own parent reports the origin the menu resolves against.
When nothing captures, that origin is 0,0 and the standalone numbers are
unchanged bit-for-bit.
Not fixed by dropping contain from the layout: it is load-bearing there,
stopping a rail collapse from reflowing a heavy mounted map.
check:fixed-anchoring measures both hosts and treats an unmeasured host as a
failure. Negative-tested: reverting the subtraction fails the studio host with
dLeft=256 while standalone still passes -- it fails exactly where the old probe
passed.
Also cards a second symptom of the same cause, not yet fixed: fixed-position
modal scrims written inset:0 stretch to the content pane instead of the
viewport, leaving the rail clickable behind an open modal.
A card was closed on an agent measurement while the bug was still visible to
the Driver, and the card had no field in which that disagreement could live.
The board said done and the product said broken, and nothing in the data knew.
Adds fields.verification (agent|human|both) plus verification_kind
(visual|state) and a conflict flag, back-filled across 179 closed cards, and
a gate that refuses an anonymous close.
The gate does not require human sign-off. Gating the board on human review was
tried and it stalled. What it forbids is losing the distinction, so the
question "which of these did a human actually look at?" stays answerable.
The visual/state split encodes the asymmetry the Driver named: human
verification is more reliable on visual bugs, agent verification on non-visual
ones. A headless probe reads geometry, not perception -- it can measure a menu
at dTop=4 and still be blind to it feeling wrong. So a visual card closed on
agent evidence is a hypothesis; a state card closed on agent evidence is a
result. The counts are reported separately.
Negative-tested three ways: stripping a tag, claiming human verification with
no quote, and marking a conflict card done each exit 1 by name.
Expanding a node that had just been selected pushed its children to the bottom of the
tree instead of under the node. The Driver isolated it precisely: clicking only the
arrow always worked, and clicking the row first and then the arrow did not — which
pointed at the filtered path rather than at the expand logic.
The re-admission of filtered-out rows walked the unfiltered baseline and appended
anything the baseline did not recognise. A child revealed by expanding a selected node
is exactly that: it never existed in the unfiltered snapshot, so it had nowhere to go
but the end. With no filter active that path is never taken, which is why the arrow
alone behaved.
It now walks the current rows and splices the missing ones back at their anchors, so
live order is preserved and the baseline is only consulted for what is absent.
Verified three ways: expanding alone puts the first child at index one; selecting then
expanding also puts it at index one, one level below its parent; and a filtered rail
still keeps its shape, going from sixty-six rows to sixty-seven with thirty-five dimmed
in place and none lost. A ten-step random walk left one hundred and fifteen selectable
rows and no page errors, and Clear All returned the rail to sixty-five.
The export carried its provenance comment above the header, and every spreadsheet
importer treats line one as the header — so the scope sentence became the column names
and the real header became the first data row. A single-column line above a sixty-eight
column table is what "glitched" looks like on screen.
Three other explanations were plausible and all three are wrong, measured before
anything changed: the file is comma-separated, not tab-separated, so the tabs in the
report were terminal rendering rather than bytes; the escaper is genuinely RFC-4180, and
embedded commas, quotes and newlines all survive a strict parser round trip; and
injection-prefixed cells were already neutralised. Only the position was wrong, and
chasing the other three would have produced churn and no fix.
The provenance line moves to the end rather than away. It records which filters produced
the file, which is the difference between a dataset and an anecdote, and a trailing
single-column row lands as one populated cell instead of destroying the column mapping.
The field counter that read line two moved with it — left alone it would have counted
the first data row, usually the same number, so it would have looked right and been
wrong.
A new check parses the export the way a spreadsheet would, which no existing gate did:
forty-six checks and not one read the file. It extracts the shipped escaper rather than
restating the rule, so it cannot agree with itself while the product disagrees.
Negative-tested by putting the comment back on top.
Cluster and People Group changed the dashboard correctly and could never move the grid,
because the fields behind them were not in the grid's fetch. They are real fields, unlike
the two family axes, which are derived and have no row field to filter against at all —
those stay unmapped rather than being aimed at a lookalike and quietly showing the wrong
rows.
Verified against the live data before mapping, since a wrong mapping is worse than no
mapping: both fields are present on 2,106 of 2,107 rows in the same value-and-label shape
as the axes already wired, with 194 and 1,736 distinct values, no collisions once
normalised, and no case where one entry's label matches another's code. Codes and names
never overlap, so either vocabulary can be accepted safely.
Measured end to end: selecting the Telugu cluster narrows the grid to 93, which is
exactly the number of rows the API holds for it.
The selections row left the DOM whenever nothing was selected, so it appeared on every
select and vanished on every clear — and the map, the table and the grid moved each
time. That is what made the dashboard hard to use: not where the chips sat, but that
everything below them shifted while you were looking at it.
The row now always renders and only its contents change. Empty it reads "Selected
World", which is also true rather than merely a placeholder: nothing filtered really
does mean the whole world is in scope. Clear-all is simply absent when there is nothing
to clear, instead of the row collapsing around it.
Height is pinned and the row no longer wraps, because wrapping was the other way it
could grow; a long selection scrolls sideways instead of taking a second line. Slim
enough to read as a status strip rather than a tray.
Measured across all three states — empty, selected, cleared — the row height stays 28px
and the map top stays at 186px. Nothing below the row moves.
The smoke check reads a data attribute to confirm a deep link selected the row it asked
for. That attribute now exists in the built page, so the gate measures the contract it
was written against instead of reading undefined.
The people-groups grid under the dashboard could not be reached by scrolling. Every
earlier investigation looked at section heights and overflow, because that is what the
documented fix addressed — but the cause was the map itself: Mapbox binds the wheel for
zoom, so the document never saw the event. Measured before: fourteen wheel ticks left
scrollY at zero and the first grid card exactly where it started.
Scroll-zoom is now disabled only while the page has somewhere to scroll, so a map that
is the whole page still zooms on wheel and a map stacked above a grid gets out of the
way. Measured after: scrollY 1549 and the first card moved from 897 to -652.
Pins keep their navy but gain a white ring and a radius that shrinks as you zoom out, so
neighbouring dots separate instead of merging into one blue mass. The Driver asked for
rings rather than clustering, explicitly to avoid more logic.
The top header no longer carries selection state. That belongs under the filter toggles,
in the second header, and nowhere else.
The filter menu is anchored to its trigger again. It had been moved to fixed positioning
to escape a scroller that was clipping it out of existence, and viewport coordinates go
stale in ways that put the menu somewhere unrelated to the box that opened it.
A gallery row now carries its identity as an attribute. A Vue key is a reconciliation
hint and is never rendered, so the smoke check read undefined and reported a broken deep
link on a page that works — measured zero occurrences of that attribute in both the
generator and the built page. Asserting on the visible label instead would have coupled
a machine check to display text.
That repaired gate immediately earned its keep: it caught the gallery asking for assets
one directory too deep. The relative asset base was correct while the gallery was the
root page and silently wrong once it moved aside for the studio, and the failure surfaced
as a JSON parse error rather than a 404, because an unknown path answers with HTML.
stack-demo rendered two empty sections in dev. Its sibling bundles were addressed with
a tree-relative path, which the browser resolves against the document URL, so no host
and no attribute can redirect it — correct in the built tree, a 404 in dev. Measured:
the dev entry answers 200 while the path the relative form resolves to answers 404. The
sources now resolve at runtime, so the same file is right in both trees, and the
production branch compiles away to the string it replaces.
That was the third place this defect appeared. Both capability-5 shells carried the same
hardcoded paths and were cured the same way, and each instance was found separately
after the previous one was fixed — because a literal parent-directory path reads as
correct in review and is invisible to a gate that only ever checks the built tree.
Removing the 404 was not enough: the sections were still empty with zero failed
requests. The shared layout injected each bundle with a classic script, and a classic
script pointed at a dev ES module fires its load handler successfully and only then
throws, so the promise resolved while the tag never registered and the wait for
definition quietly timed out. A load handler cannot detect that.
The module rule now lives in the shared layout rather than in a fourth private copy of
the same loader, so the two other composites that inject through it are covered too.
Verified after: both elements define, sixty-six facet rows and twenty-four cards on
stack-demo and on dashboard-plus-grid, zero failed requests on each.
public/ is committed so a developer can run the dev server immediately without a
four-minute build first, which means stale build output ships to every clone. This
rebuild carries the heading-prop rename in both capability-5 shells and relocates the
archived gallery to its own path.
A bundle can be compiled and present on disk while its profiles never reach the
manifest, and nothing noticed: the leaf floor counts index.html recursively, so one
bundle losing every leaf while another keeps extras stays above the floor; the
embed-target check iterates what the manifest lists, so an omission is invisible to it;
and a bundle count passes too, because one bundle can be swapped for another and the
totals still match. Counting is not covering. The only way to see an omission is to
compare the manifest against the scanner rather than against itself.
The first version of this check cried wolf on its first real run. It reported key-cities
as missing — discovered with four profiles, absent from the manifest, and still absent
after a full rebuild, which made it look persistent and real. It is neither: the build
config gates that bundle by name and generate-tree filters on it, so the bundle is
legitimately discovered and legitimately unpublished. The scanner answers what exists;
the config answers what ships. Asserting the first against the manifest without
consulting the second asserts an invariant that was never true.
That failure mattered more than the bug it was chasing. A gate that reports a deliberate
configuration as a defect gets switched off, and a switched-off gate protects nothing —
so the check now reads the same config the build reads, and names the gated bundles in
its passing output so the exclusion stays visible rather than becoming folklore.
A manifest entry the scanner does not return is reported as a note rather than a
failure: stale build output heals on the next full build, while a missing plugin is
silently absent from the studio for everyone reviewing the product.
Negative-tested: removing any published bundle from the manifest fails with exit 1
naming it; restoring passes; the gated bundle never trips it.
Both shells declared a `title` prop. defineCustomElement maps every prop to an
attribute, and `title` already exists on every HTMLElement as the tooltip, so Vue warned
on every single mount that it would overwrite the native property. Harmless in effect —
nothing here wants a tooltip — but a warning repeated on every mount is where a real
error goes to hide, and this repo has already paid for that once when a thrown setup()
silently removed three buttons behind noise that looked ignorable.
Renaming the prop was safe because nothing outside ever set it: no host passes title=,
no profile-config reads it, and the value is only forwarded to the shared layout, which
is an ordinary Vue component rather than a custom element and so has no collision of its
own. Both shells carried the identical declaration — the same duplicate-implementation
pattern that earlier turned one resolver assumption into two separate outages — so both
were fixed together.
The studio smoke check now samples several tabs instead of only the first, and measures
node depth inside the mounted element rather than counting the slot's children. The old
check read one child for every pane whether the element rendered a full dashboard or
nothing at all, which is the "mounts is not works" trap its own header warns about. It
polls for content rather than sleeping a fixed interval, because a fixed wait either runs
too slow to use on every edit or marks a slow tab broken, and a gate that cries wolf gets
switched off.
Measured after: zero Vue warnings in the browser console.
check:embed-targets was already wired into package.json while the file it runs sat
untracked — so a fresh clone got a script that could not execute. It asserts that every
embed.iframe src in the manifest resolves to a file on disk, which is the load-bearing
check for the iframe deployment method: one of only two of the five that are proven.
That assertion matters more now than it did. The build stopped emitting a chooser page
per bundle, which took the recursive leaf count from sixty-three to forty-four, so the
coarse leaf floor no longer distinguishes "nineteen pages deliberately removed" from
"a partial wipe deleted real leaves". This gate does: it names the missing file and the
manifest path that pointed at it.
Negative-tested: hiding one leaf fails with exit 1 naming both the file and its JSON
path; restoring returns to a clean forty-three.
Checkpointing thirteen files of agent work that were sitting uncommitted while three
agents wrote to a live tree the Driver was watching through hot reload. Uncommitted work
is work at risk, and a tree being edited under an observer oscillates between states that
look like regressions and are really just partial saves.
The second capability-5 shell now mounts its tabs as real elements too, so no iframe
remains in either shell. It was the last holdout: eight tabs, all iframe mode, each
pointing at a hardcoded built-tree leaf that does not exist under the dev server.
The facet rail keeps its shape when a filter narrows it. Selecting one place took the
rail from sixty-six rows to twenty-one — forty-six rows deleted, none dimmed — so the row
a user was about to click vanished under the cursor, which is what "the buttons went
dead" actually was. Rows whose count falls to zero are now re-admitted at their original
slot and depth, dimmed and inert, which also restores the component's own promise to lose
nothing against the tree it replaced.
Expanding a branch and filtering by it are now separate, visible affordances: the whole
row filters, and expand is a real button with its own hit area and label. Rows that were
never clickable are unmistakably inert rather than silently ignored.
Three more facet axes reach the grid — affinity bloc, engagement status and primary
language — each verified to share a field with an existing prop rather than being mapped
on name resemblance. Two derived facets remain unmapped on purpose: they have no matching
row field, and mapping them would filter confidently to the wrong rows.
One commit shipped the requested change and a regression together: a base prefix applied
to an already-rooted URL 404'd all 29 studio tabs, and finding it took 32 minutes. The
root-cause analysis attributes the largest single share of that to there being no gate
that exercises the dev server at all — of 46 gate-ish scripts and 13 browser gates, none
did. The built manifest hands relative refs and the dev manifest hands rooted ones, so a
base-join bug is invisible to every build-time gate and fatal to every developer.
The check finishes in about four seconds against an already-running dev server, which is
the point: the full sweep costs about seventy-five, so it only ever gets run at the end
of a change, which is exactly how a regression rides along unnoticed. It activates one
tab rather than twenty-nine because a single mount already exercises the whole resolver
path — manifest fetch, script resolve, module load, define, mount. If the resolver is
wrong, the first tab fails like all of them.
It asserts the five things that have actually broken: that the root serves the studio and
not a staging page, that the element both mounts and registers, that the tab list is not
empty, that no request failed, and that the activated pane holds real content — because
an element can register and attach while its content never arrives, which is what a
missing API base looks like.
An unreachable dev server exits 2 rather than 1. Conflating "no server" with "broken
product" is how a gate earns a reputation for flakiness and gets switched off.
Negative-tested: removing the tree base makes it fail on the empty tab list, and an
unreachable URL exits 2; restoring returns to green.
The built leaf pages set window.MAP_APP_API_URL; the dev pages set only the Mapbox
token. Every dashboard in dev therefore fell back to a relative /api/people-groups/list,
which Vite answers with SPA-fallback HTML at HTTP 200 — so the fetch "succeeds", the
parse yields nothing, and the facet rail renders zero rows with no console error and no
error card. One studio sweep produced eight such requests. It reads as a broken
component and is really a missing global.
Both dev routes now read the value from the same build config the built leaves use, so
the two surfaces agree by construction rather than through a second hard-coded copy, and
each route resolves the token and the API base from a single config read so they cannot
disagree about their source. Measured after: the research dashboard renders 66 facet
rows and 216 table cells with zero 4xx.
There were THREE staging surfaces, not two. The repo-root index.html was Vite's default
dev entry, and its title was the line that kept reappearing in front of the Driver:
"doxa-maps · Embeddable JAMstack Maps · staging", above a bundle strip that loaded each
chosen map into the page itself. It is archived with its internal/staging.js, not
deleted, and git recorded both as renames so history follows.
Archiving is safe because it was never a build input: rollupOptions.input is built from
per-bundle JS entries, and the root HTML is referenced only on the dev branch. It was
also already unreachable — the dev plugin intercepts /, /index.html, a query-string
root and /__gallery, all measured returning the studio with no staging title before the
move, and measured again after. What remained was a file that would resurface the moment
the middleware changed.
The settled-tree leaf floor drops from 50 to 40 because the architecture changed, not
because the tree got sicker. That floor was set when the build also emitted a chooser
page per bundle; those 19 pages are deliberately gone, which takes a correct tree from
63 leaves to 44. Left at 50 the gate would fail on every correct build, and the cheapest
way to make it green again would be to restore the choosers — so a stale floor does not
protect the tree, it defends the old architecture. The floor still catches the
partial-wipe it was built for, and the load-bearing assertion is now that every
embed.iframe target resolves: 43 checked, 0 missing.
Prefixing a base onto an already-rooted URL 404'd all 29 studio tabs, and no existing
gate could have seen it: the built manifest hands relative refs, so concatenation is
correct there and the assumption is never exercised. Only the dev server hands rooted
refs, so the entire build-time suite passes green while every developer's studio is
broken. A product that has to work on ten laptops cannot be verified by the build alone.
The check discovers every shell that defines a resolve(ref) rather than naming them,
because the two capability-5 shells carried byte-identical implementations and one
latent assumption therefore produced two separate outages that had to be found
separately. A third shell copied from either is now caught on its first run.
It evaluates the shipped function against a stubbed host instead of re-implementing the
rule, so the gate cannot quietly agree with itself while the product disagrees.
Negative-tested: removing the rooted-ref guard fails 4 of 8 cases and reproduces the
exact double-slash path from the outage; restoring it returns to green.
A capability-5 shell mounted inside the studio carries its own tab list and resolves
it against its own tree base, which it derives from its script URL when nobody supplies
one. Nested here that guess was empty, so the resolver took its parent-directory
fallback and requested paths above the leaf at the server root — three bundles 404'd
that way. The requests were non-fatal, because each tab still rendered from its
fallback list, which is exactly why they survived: a silent 404 behind a working
screen. The studio already knows the correct base, since it is the one it resolved
against, so it now passes it down through the existing profile-config path rather than
letting the child re-derive it.
EVERY TAB FAILED TO LOAD. Introducing tree-base for the dev server prefixed it onto
URLs that were already rooted, producing /__plugin-gallery//plugin-designs/<b>/index.js
and a 404 on all 29 tabs. Both capability-5 shells had byte-identical resolve()
implementations, so both carried the same defect; both now return an already-absolute
ref untouched and join through exactly one slash. The built tree hands relative refs
and keeps its base, so this is one rule for both trees rather than a dev-only branch.
THE STUDIO NO LONGER FRAMES ANYTHING. Map elements were nested in iframes pointing at
their standalone leaf pages, to dodge a one-shot customElements.define collision. That
collision cannot occur here: 19 bundles claim 19 distinct tags, none shared, and two map
elements were proven to coexist in one document with the first unaffected by the second.
Every pane now mounts the plugin's real custom element, which is what makes a row both
reviewable and nestable.
Panes keep two lifetimes, and the difference is the WebGL context budget rather than
taste. Light rows stay mounted so switching back is instant and state survives. Map rows
are destroyed on deactivate, because a hidden map holds its GPU context: the browser
caps live contexts, then silently reclaims the OLDEST, so a map visited ten tabs ago
would go blank with no error anywhere. getContext never returns null and contextlost
never fires on the new canvas, which is what makes that failure so hard to attribute.
The dev server serves a leaf route for one element with no chrome, and passes the tree
base to nested shells so they discover the live tree instead of a baked fallback.
THE DEV SERVER LANDED ON THE WRONG PAGE. `bun run dev` served only the staging
gallery at /, and had no route for a bundle leaf, so there was no URL at which the
capability-5 plugin studio could be seen at all. Opening / showed plugins wrapped in
per-bundle frames behind a wall of explanatory prose, which reads as "none of the
elements are working" when in fact the elements were fine and the page was a chooser.
/ now mounts <plugin-studio> directly: a 605-byte host page with no chrome and no
prose, the element filling the viewport, and the Mapbox token baked as the token
ladder's first rung so maps work without anyone appending ?tk=. The gallery is not
deleted — it still generates the built index and is load-bearing for the manifest —
it simply moves to an explicit /__gallery URL and stops being the landing surface.
tree-base is passed explicitly, without which the studio mounts EMPTY. It builds its
tab list by fetching <base>manifest.json and falls back to a path relative to its own
script URL, which in dev resolves to a file that does not exist. The dev server serves
the same manifest-shaped model on its own route, so dev now shows the same tab list as
production: 19 bundles, 30 profiles, five capability groups.
PLUGIN FRAMES SHOWED THE GALLERY. The studio nests map profiles in a lazy iframe
pointed at that profile's leaf page, and dev had no such page, so each frame fell
through to the SPA fallback and rendered / — the gallery, recursively, once per map
tab. A /__leaf/<bundle>/<profile> route now serves the dev equivalent of the built
tree's leaf: one element, its bundle entry, the token, and nothing else.
FILTER DROPDOWNS WERE INVISIBLE. Opening a filter rendered its options underneath
the surrounding content, making the dashboard's primary control unusable.
The cause was CLIPPING, not stacking order, and that distinction is why an earlier
attempt failed. The filter band is a one-row horizontal scroller authored as
overflow-x:auto with overflow-y:visible — but CSS Overflow 3 coerces a visible axis
to auto when the other axis is not visible, so it computes to auto/auto and clips
vertically. Clipping happens during painting, before stacking order is consulted, so
raising the band to z-index 55 changed nothing: it won a stacking fight that was never
the problem. You cannot re-order pixels that were never generated.
The menu is now position:fixed, anchored to the trigger's viewport rect, which takes it
out of the clip chain entirely. Fixed coordinates go stale whenever anything scrolls or
resizes, so they are recomputed on open and while open, listening in the capture phase
so the dashboard's inner scroll regions are seen too.
PEOPLE-GROUP IMAGES WERE LETTERBOXED. A fixed 130px height against a 220px column
cropped each person to a band. The image is now a square via aspect-ratio, and the
column floor rises to 240px, roughly doubling the visible area. object-fit:cover is
kept deliberately — stretching to a square would distort faces, which is worse than
the letterbox it replaces.
Four fixes to the research dashboard and the cap-4 composite, all Driver-reported.
GRID SHOWED ZERO ON EVERY SELECTION. Picking a region showed "1,309" in the chart
above an empty grid below. The two halves spell the same value differently: the
dashboard emits the human LABEL ("Asia", "North America & non-Spanish Caribbean")
while the API stores the machine VALUE ("asia", "north_america_&_non-spanish_caribbean"),
and the comparison was a strict ===, so it never matched. Measured over the 2,107
live records. Adds a shared valueMatches() normaliser used by BOTH applyDropdownFilters
and applyWith, so facet counts and shown rows cannot disagree. Compares value OR label,
case- and separator-insensitively, chosen over a label-to-value lookup table that would
need maintaining per axis and would rot when the API adds a region.
The grid's own dropdowns were never affected, because they build their options from the
data itself; only the host-prop path could inject a label into the comparison. That is
why the standalone grid is fine and the gate passes 13/13 on the same records.
COUNTRY NOW REACHES THE GRID. Clicking India in Top Countries did nothing because
country_code had no matching prop and was deliberately dropped rather than guessed at.
Adds a real countryCode prop and maps it. Country is prop-only, not a fifth dropdown.
LONG FACET LABELS HID THE COUNTS. white-space:nowrap let a long label eat the row, and
the rail scrolls vertically only, so there was no way to recover the number. Labels now
wrap; the bar yields; the count never shrinks.
CAMERA FLEW TO THE NORTH POLE on globally dispersed selections. Fits now trim outliers,
reject invalid and null-island coordinates, and unwrap longitudes across the antimeridian,
so the view lands on the data instead of a pole.
The provider option landed last change but nothing a host passed could reach it:
the dashboard's config is a static import, so the setting existed in source and
was unreachable by an embedder. Real and unusable is the same trap as a nav
control that exists off-canvas. Hot-swappable has to mean a partner can choose
without a rebuild.
profileConfig is the channel every other plugin already uses for this, so this
adds no new mechanism — it stops the dashboard being the one plugin that ignores
it. The merge is shallow and the host wins. Shallow is deliberate: the static
config's nested objects are a coherent set that a partial override would corrupt,
and a deep merge would invite half-replaced facet maps that fail far from the
edit. What a host legitimately sets is top-level scalars.
With no host config the behaviour is byte-for-byte what it was, verified against
the baseline captured before any of this work: same node count, same canvas, same
row count, same engine, same ten tile responses.
The Driver wants MapLibre available without breaking Mapbox: keep every Mapbox
component working, add the option beside it, start with this dashboard.
The change is deliberately additive. provider defaults to mapbox, so with no
config the behaviour is what it was and nothing Mapbox-shaped can regress. A host
opts in through dashboardConfig, which is how every other dashboard setting
already arrives, so the cap-4 composites can pass it down with no new mechanism.
The plumbing already existed and was simply unreachable from here: useMapInstance
takes a provider and loads the matching engine, and the style registry maps every
named style to both engines with the MapLibre side pointing at keyless URLs. This
call passed neither provider nor style, so it silently took the default.
The token stays wired when MapLibre is selected. MapLibre ignores it, and keeping
it means switching back is one word with nothing else to restore, which is what
hot-swappable has to mean to be worth anything.
The Driver's report was exact: the pins load and the underlying map does not —
a grey field with data floating on it. That is the URL-restriction signature. The
baked token was scoped to doxa.life, so tiles returned 403 from nuxtbase.app,
from localhost, and from a bare origin, while the style document returned 200
everywhere. The map therefore booted, drew every dot, and showed no basemap.
The Driver supplied a token with no URL restriction. Measured side by side: it
serves tiles 200 with no Referer, from nuxtbase.app and from localhost, where the
previous one returns 403 in all three cases.
Also empties tokenUrl. It pointed at the endpoint that serves the restricted
token, and that endpoint sits above the baked token in the ladder — leaving it
set would fetch the broken token at runtime and overwrite the good one.
All 43 leaves rebuilt and carry the new token; none carries the old one. Verified
in a browser with no Referer sent, the way a real visitor arrives: nine of nine
tile requests returned 200.
This unblocks the interoperability work, which could not be judged while the map
underneath was blank.
A backtick in live shadow CSS breaks the build exactly as one in a CSS comment
does, and check:template-literal only reads the comments. Proven both directions
last tick: the planted defect left that gate at pass while the build died with a
Babel stack trace pointing far from the edit.
The card concluded this was not achievable, on good evidence — three lookahead
heuristics were tried and reverted, scoring 98 and then 136 false positives on a
healthy tree, because the real terminator of a template literal is also a
backtick and no window of surrounding characters separates the two. The evidence
was right and the inference one step too strong. No heuristic works is not
nothing works, and the card named the answer in its last line: needs a tokenizer.
A parser does not guess where a literal ends; ending it is what parsing means.
Measured against the same tree the heuristics failed on: zero false positives
across 83 files. Splitting the SFC first means it sees exactly the script text
the build compiles, and using the parser the build reports errors from means its
verdict matches what actually breaks rather than approximating it.
Negative-tested with the exact case the other gate misses: a backtick planted in
live CSS passes check:template-literal and fails this one, naming the file and
the line that was edited rather than a stack trace somewhere else. That line
number is the point — the build was always the backstop, just an unhelpful one.
This gate's own header names the failure it exists to prevent — two artifacts
encoding one fact with nothing forcing agreement, where fixing one copy is not
fixing the fact. It then guarded exactly one artifact.
Measured: planting a claim that doxa-vision-card is the capability 1 exemplar
into documentation/reference.md passed both this gate and the exemplars gate.
That is precisely the error this file was written to catch, sitting one directory
outside its scan. doxa-vision-card ships an instances file with two entries, so
the classifier types it parameterized — capability 2, not 1.
The documentation folder states the tier correctly today, so this is not fixing a
live defect; it closes the hole that would let the same one recur unnoticed in a
second place. Negative-tested: the planted claim that used to pass now fails with
the file named, and removing it returns to green.
The same stale citation also lives in the plugin-designer skill, which is outside
this repo and so beyond any gate here. That half stays a Driver edit; the repo
now holds the checkable half.
I excluded smoke:share from the aggregate two ticks ago and the reason was wrong.
Every clause of it was true — the script takes required arguments, and its
gallery mode does need the vite dev server — but the conclusion did not follow. I
had tried exactly two invocations, bare and --plugin-gallery, and never --url,
which the script's own header documents first. It runs fine against the static
tree: 8 of 8 pass, each Share button reopening the correct custom element with a
live canvas.
That is the same shape I logged as an anti-pattern last tick, now found in my own
work: a not-applicable verdict is more durable than a wrong bug report, because
it closes the question. It cost the suite its highest-consequence surface for two
ticks. A share URL is the one artefact a partner pastes into an email, and it
fails silently.
It runs as smoke:share-all, carrying its own eight --url flags and --browser
chromium, since Firefox headless has no WebGL on this box and the default engine
stops. The eight leaves are the set derived by grepping for ShareButton, not a
guess. The simple-map case is the discriminating one: it shares to a different
leaf than the one opened, so the deep link is really being followed rather than
an SPA fallback rendering something plausible.
Two shell traps on the way. A VAR=x prefix does not apply to $VAR on the same
line, because the shell expands the argument before the assignment scopes to the
command, so the URLs came out with an empty base; deferring through sh -c fixes
it. And a glob in a block comment closed the comment early.
Every map image shipped data floating on an empty navy void — no coastlines, no
borders, no labels. Two cards had this blocked on the Driver adding origins to
the Mapbox token's allow-list, for four ticks.
It did not need the Driver. Measured both directions: a tile request with no
Referer returns 403, the same request with the doxa.life Referer returns 200,
and the style document returns 200 either way, which is why the map still booted
and every gate passed. The shot harness serves from 127.0.0.1 and so sent no
Referer at all. These are doxa.life's own documentation images, generated by its
own repo from its own token; the restriction exists to stop third parties burning
the Driver's quota, and it was blocking the owner.
Route interception looked like the careful way to scope it and silently does not
work — Chromium ignores a referer supplied through route.continue, so tiles still
403'd. Only the context-wide header applies, and that breaks same-origin iframes:
the iframe-host shot went from 187 elements to 15.
So it is opt-in per shot, and the predicate is architectural rather than
anecdotal: any surface that nests content in an iframe opts out. I set it for the
one shot that failed first, then watched both cap-5 shells collapse on the next
run, and generalised. Map leaves get their basemap; shells keep their content.
Verified by reading the images, not the file sizes — a size jump had already
misled me once this tick.
It showed an empty pane reading "Choose an element from the list." The README's
headline image for the highest capability level demonstrated nothing, and every
shot gate passed it, because freshness, distinctness and element counts are all
satisfied by an empty pane. The shot now selects the first rail row before
capturing, so the picture shows what the studio is for: a nested SPA rendering
inside the shell that lists every element by capability. The file went from 167KB
to just over a megabyte, in line with its sibling.
Found by reading the image, the same way the mobile error card was found
yesterday. Reading one picture has now caught two defects that eleven gates
passed.
Add check:shot-fonts. The studio labels each row with a capability emoji, and
this machine has no emoji-capable font, so all 28 rendered as tofu boxes in the
committed PNG. The markup is right and the product is fine in a real browser —
the shot machine could not draw it — so the fix is not to strip the emoji. The
gate asks fontconfig whether this system can render the glyphs the UI actually
uses, and names the font install rather than the markup.
It is deliberately not in build:unlocked: it reports an environment condition
that is currently true here, and a gate that always fails in the build makes the
build useless. Its first version also flagged the star this repo uses as a
comment marker, which every text font renders; the range is now narrowed to the
pictographic planes.
Two composites nested a child by tag and URL without a profile-config. Those
children do not render a component directly — their bundle entry mounts
ProfileLoader, which reads profile-config.profile to choose which view to load.
With none, the element still mounts and still reports its tag, so every
structural check said both children were present, while the loader rendered
"missing or invalid profile-config prop" where the map should be. The grid half
of each page worked, which is how it survived.
Found by reading a regenerated README screenshot, not by any gate. The mobile
screenshot showed an error card and an empty pane; the fixed one shows the real
research map with its data, controls and progress figures.
research-map-plus-grid was mine, shipped that way. stack-demo predates it and had
the same defect against research-dashboard. Both now pass the shape their own
generated leaf pages use, so nested and standalone mounts ask for the same thing.
check:composite-srcs now requires a profile-config on any child whose bundle
mounts ProfileLoader, derived from which entries import it rather than a list.
The first version of that check tested for the word and passed on its own
explanatory comment; it now matches the quoted attribute key, and a negative test
that mutates only the key fails as it should.
Also regenerates three derived artifacts that had drifted behind this flight's
changes: the library index, the capability exemplars table, and all twelve README
screenshots.
All seven browser gates were orphans: each had a script, and not one was invoked
by anything else. They were reachable only by someone who already knew the name
to type, which is exactly the person who does not need the list. This repo has
been bitten by this before — the Gate-1 acceptance test existed, worked, and was
invoked by nothing, so it could never report anything. An orphan gate is worse
than a missing one because it looks like coverage.
The data guards already solved this; verify-all.mjs says it plainly — a guard
nobody runs is documentation. That file has no reference to playwright or to the
built tree. It is the data half. This is the browser half.
Every gate listed here was run before it was listed, following the rule the
earlier audit set: listing a program you have not executed puts unvalidated
entries in the hand-off surface. All six pass, and the aggregate now runs them
end to end: capabilities, embeds, map paint, the Gate-1 chain, accessible names,
and phone navigation.
smoke:share is excluded on purpose and says so. It takes required arguments and
its gallery mode reads a registry only the vite dev server serves, so listing it
would make the suite fail for something that is not a defect.
Preconditions are asserted up front and exit 2, so a missing server reads as a
setup problem rather than six mystery failures.
The file header said tab 1 was bp-big-picture, mounted directly via its real tag.
Both halves are false: tab 1 is dashboard-plus-grid, and every tab is an iframe —
there is no direct-tag mount left in this shell. A reader trusting that text would
look for a mount path that does not exist and maintain the wrong bundle believing
it was the one users see. The big-picture bundle is still a valid element; it is
simply one of the parts already composed into the tab-1 composite.
The locale switcher also poked slot zero by hand, writing a profile-config named
big-picture onto what is now an iframe, which ignores it. Harmless, and that is
why it survived: the effectiveLocale watcher was doing the real work all along,
branching on the tab's mode and re-pointing the iframe src that carries the
locale. Removed rather than repaired, since a second hardcoded locale path is one
more thing to keep in sync for nothing. Verified by driving the switcher: the
active frame moves from locale=en to locale=es.
The direct-tag branch in the template is kept. It is generic, keyed on the tab's
mode rather than a bundle name, and remains the right way to nest a future
token-free element. Only the comment claiming tab 1 uses it was wrong.
Yesterday the cap-5 shell shipped unusable on a phone: the rail was an off-canvas
drawer, the toggle lived inside a header the shell does not render, and the only
other control sat inside the drawer itself at x:-244. Every tab but the one that
happened to open was unreachable, with no error and a page that painted fine.
That survived because it is an intersection bug. Header-off is correct, and an
off-canvas mobile drawer is correct; the defect exists only where they overlap.
So this checks the intersection directly — each cap-5 shell as it ships, at a
phone viewport — and asks whether any on-screen control opens the rail.
It then presses that control and requires the rail to arrive. Presence is not
function: a present-but-inert opener is the same dead end wearing a disguise, and
asserting only presence is how the original bug survived a card marked verified.
This is not covered by the accessible-name gate. Neither property implies the
other: the off-canvas button had a perfectly good name while being unreachable,
and reachability is geometry, not markup, so no source grep can answer it.
Negative-tested by re-creating the real bug — removing the floating opener makes
it fail naming that shell, while the header-bearing studio still passes.
The Driver's ordering: desktop opens on the Big Picture dashboard; mobile opens
on the research map and does not show Big Picture at all. Both capability-4
composites already existed, so this is the shell choosing between them. A row in
the nav model can now declare onlyWhen desktop or mobile; everything below the
first rung is identical on both, which is why only the first two rows carry it.
One nav model rather than two, so the teaching order lives in one place. Two
parallel models would be two things to keep in sync and the failure would be
silent: a row added to one and forgotten in the other simply never appears.
Crossing the breakpoint can remove the row that is currently active, which would
leave the user on a mounted but unlisted tab with no highlight. The shell now
snaps to the first surviving row, and establishes the viewport class before
opening the first tab so a phone never opens Big Picture for a frame.
Also fixes a pre-existing bug this exposed: with no header there was no way to
open the rail on mobile. The drawer sits off-canvas, the only toggle lives inside
the header, and the collapse control is inside the drawer itself — measured
off-screen at x:-244. Every tab except the one that happened to open was
unreachable. A floating button now appears only when narrow, headerless and shut,
so no existing consumer can see a change.
A cap-5 shell nests its children as iframes, and window.MAP_TOKEN does not cross
an iframe boundary — each frame is its own window. A nested map cannot inherit
the parent's token, so it works only because the child leaf carries its own baked
copy. Proven in a browser: all thirteen nested frames in the cap-5 shell hold a
real token, including a map two levels deep. That was a one-off run; this makes
it standing.
The existing smokes cannot catch this. Both inject window.MAP_TOKEN with
addInitScript before the page runs, which is right for what they test but means
the test supplies the thing under test. A user opening a shipped leaf has nothing
injecting a token for them.
Also asserts the reverse: no token may be compiled into a .js bundle. Bundles are
cached, SRI-pinned and reused across hosts, so a token baked into one cannot be
rotated without a rebuild and leaks into every embed.
Reads the built manifest rather than walking directories, so it checks the set
that actually ships. Counted 43 leaves, matching an independent filesystem count.
Negative-tested both directions: stripping a leaf's token and planting one in a
bundle each fail with the artifact named.
posterUrl() said .png. There are 378 poster .webp files and no matching .png, so
the helper could only ever have produced a 404 — but it was never called, because
the export and print paths each built the same URL inline. A dead helper carrying
a wrong constant is worse than no helper: it is primed to break whoever tidies up
by routing a call through it. Point it at .webp and route both callers through
it, so the format lives in one place.
The file header still said posters "DO NOT EXIST YET. Nothing has been generated."
378 previews are generated, shipped and served live, and the wall paints from
them. A reader following that line would go build a generator that already ran.
Replace it with what actually ships: previews by default, full-resolution posters
excluded because bundling them pushed the tree past what the catalog would accept.
The desktop composite pairs the research dashboard with the grid. This is its
mobile sibling: the monolithic 7-tab research map on top, the same grid beneath,
stacked to scroll. Which one a shell shows is an aspect-ratio choice made at the
SPA level, so neither plugin knows about the other and both stay embeddable.
The grid's own search and filters are ON here. The desktop composite turns them
off because the dashboard header owns those controls and drives the grid through
props; the research map has no such header, so switching them off would leave the
grid with no way to be searched at all.
Composes its children by tag and bundle URL rather than importing their .vue
files. Importing the research map directly would bypass its bundle entry, which
is where its three bundle-private colour strategies, its profile registry, its
Pinia and i18n pair and the RTL text plugin are installed — and losing those is a
runtime defect in a green build, because a thrown setup() removes the subtree
with an empty error.
Add check:composite-srcs, which resolves every nested src against the built tree
and asserts every nested tag is registered. A wrong relative depth is a 404, and
the stack layout swallows a failed section on purpose so one broken child cannot
take the whole stack down — which means the cost of that resilience is a silently
empty band. Negative-tested both ways: breaking the depth and renaming a tag each
fail with the file and path named; restoring returns to pass.
The header claimed, in the present tense, that this workflow serializes builds
at cluster scope. The forge reports 0 recorded runs across 288 commits, so that
serialization is not in force: the local flock is doing all of it. State the
measurement and the caveat rather than let an inheritor rely on a gate that has
never fired.
Ran the browser-free gate suite inside a fresh clone -- the inheritor's
conditions. Five passed; check:dxlife went red with "clone not found at
/root-dx-life - nothing to check".
That is not a defect. The gate deliberately uses three values (0 unchanged,
1 drift, 2 no clone) because it exists after a check that returned the PASS
value BECAUSE IT FAILED: `cd root-dx-life && git status --porcelain | wc -l`
printed 0 for a directory that is not a git repo. Collapsing "cannot judge"
into "pass" is exactly the bug it was written to prevent.
But `bun run` and any && chain see only zero/non-zero, so the middle value
evaporates precisely where a reader consumes it. The fix belongs in the
documentation, not the gate: added a preconditions row stating what it needs,
what exit 2 means, and that it is NOT in build:unlocked, so a clone still
builds.
Verified the whole inheritor path again after the Cargo.lock change: clone
(2358 files, lockfile present) -> bun install exit 0 -> bun run build exit 0,
18 bundles / 27 profiles, with check:dep-names and check:no-abs-paths both
running inside the clone's own build.
deployment/src-tauri/.gitignore listed /target, /gen and Cargo.lock together
with reasoning that fits the first two. /target is 2GB of compiled output and
/gen is emitted by tauri-build on every compile -- both genuinely
regenerated. A lockfile is not build output: it is the RESOLUTION RECORD, the
only artifact that makes cargo build reproducible.
Cargo's rule splits on crate type -- commit for a binary, ignore for a
library -- and cargo metadata reports this crate as kinds=['bin']
(src/main.rs, no lib.rs). Checked with cargo rather than inferred from the
file layout.
The stakes are concrete: Cargo.toml pins nothing tighter than
tauri = { version = "2" }, while the lockfile pins 430 packages. Without it
an inheritor building next month resolves a different dependency tree than
the one that produced the shipped 64 MB binary, and method 5 is already
"built, not shot", so its build must at least be repeatable.
Verified before committing 109KB: cargo metadata --locked --offline succeeds,
so the lockfile is consistent with the manifest, and it contains zero local
paths or credentials.
check:tauri gains a fifth assertion -- Cargo.lock present AND TRACKED.
Tracked-ness is the point: the file sat on disk the whole time and was still
absent from every clone. Negative-tested against the real state, not a
simulation: the gate FAILED naming the file while it was untracked, and
passed once staged.
Makes the cheap half of last tick's P0 automatic. A folder rename had
rewritten three npm package names, so bun install failed in a fresh clone
while everything here passed -- the real packages were already in
node_modules from before the rename, making the breakage invisible to every
build and every gate.
Checks two things, offline, in milliseconds: the declared name resolves under
node_modules/, AND the installed package's own name field equals the declared
key. The first alone catches the capacitor case; the second catches a name
that resolves to a DIFFERENT package, which a path-only check would miss.
Together they detect the corrupted-name class without a network round-trip,
which is what lets it run on every build rather than in a nightly job nobody
watches.
It does not replace the clone test. A gate can only see what is IN the repo;
only a clone sees what the repo fails to do on a clean machine.
Negative-tested four ways: recreating the EXACT P0 (the three
@deployment/capacitor/* names) FAILS naming all three; a name that resolves
to a package whose own name differs FAILS; an absent node_modules SKIPs
honestly rather than passing or failing; an empty dependency list FAILS via
the vacuous guard. Wired into build:unlocked and confirmed running inside a
full build (17 declared, 17 verified, exit 0).
Tested the north-star path as an inheritor would, by cloning the repo from
Forgejo and running the documented first command. It FAILED:
error: GET https://registry.npmjs.org/@deployment%2fcapacitor%2fcli - 405
error: @deployment/capacitor/cli@7 failed to resolve
The rename of capacitor/ to deployment/capacitor/ did a blind string replace
that rewrote the NPM PACKAGE NAMES too. @capacitor/android, @capacitor/cli
and @capacitor/core became @deployment/capacitor/*, which do not exist on
npm. Restored the real names, pinned to ^7.6.9 -- the version actually
installed -- and fixed the two places that repeated the wrong name in prose.
This is the inheritor-only failure shape, in a form the new absolute-path
gate cannot see. The local tree works perfectly because node_modules/
@capacitor/* was installed BEFORE the rename, so every build and every gate
passes here. The breakage lived only in the DECLARED names, visible solely to
someone installing from scratch.
Verified by doing it: a fresh clone now runs bun install to exit 0 (206
packages, all three @capacitor/* resolved) and bun run build to exit 0,
producing 18 bundles / 27 profiles -- identical to the working tree.
