A bench day and a website day at once. AFAL Studio gained a portrait layout and
reached V1_04, the fourth GAAD seat was named, Episode 6 was retitled, and two
website generators were corrected at source. Two publishes went out, 1_062 and
1_063.
The studio was unusable in portrait: the control panel stood taller than the
screen, so a slider sat below twenty species names. GLAS-web V1_04 gives it a
purpose-built narrow layout, measured at 789 pixels of panel at the start of
the day and 286 at the end. The desktop layout is untouched.
The phone gets native pickers for Species and Preset, which on a phone is the
system wheel; Points and Height as the two primary sliders with Radius, Taper,
Rotate and Twist behind a more control; the six buttons in two rows; the plan
and elevation drawings at 58 pixels; the band fixed over the 3D view so it
costs the stage nothing; and no inner scrollbar. What the phone does without,
all by ruling, is the twenty-species tree, the preset grid, the Generate box
chrome, the status line, the RNG box, the genome file box and the warning
block.
None of the reduction was taken out of the language. It came from the species
tree becoming a picker, the readout losing the block beneath it, and the
surviving boxes losing their frames.
The two pickers and all six buttons are the same elements in the document at
both widths. A mount function relocates them when the screen narrows and
returns them when it widens, verified in both directions. The obvious build is
a phone panel and a desktop panel each showing the same genes, which is the
shape that produced nine defects on 27 July: one truth written down twice, then
drifting. There is one Roll button in the document, so there cannot be two that
disagree.
Height was added as a derived control. The gene is Rise, per instance, and
height is rise multiplied by count less one. The cap is 300 metres by ruling,
and the implemented ceiling is the lesser of 300 and whatever Rise can reach at
the current point count. The count term is load bearing rather than tidy:
without it the control could write a Rise outside the slider's range, Blender's
socket would clamp it on import, and the same genome would build two different
forms in two hosts. That is the case the conformance test exists to prevent.
Also built at the bench: Species variety and Preset variety, which walk the
five built species and the six presets at two seconds each and name the current
one in gold, both reading their source tables so a new species joins the walk
with no edit; all three walks sharing one hold constant, with the second
constant deleted rather than set to match; mutual exclusion between the walks,
including the handover case; Reset returning to Spiral through a single default
constant; Points raised from 1200 to 2000, which is a gene limit and so applies
at both widths; the readout rebuilt as a fixed grid, four columns at desktop
and three on a phone, so the block no longer changes height as a number gains a
digit; and the viewport hint moved to the bottom right, dimmed, fading on first
orbit.
Removed with consequences noted: View PNG, which existed because the artifact
sandbox eats downloads silently; and the precedent drawings, whose six figure
generators are now unreferenced and were left in place deliberately, because a
deletion is an edit and this one is easy to want back.
Two generators were corrected at their source rather than at their pages.
build_studio.py went to v1_002. V1_04 ships a band whose first child is an
empty slot put there for the GAAD mark, and the injector was still trying the
panel first, which on the phone layout would have placed the mark below the
geometry because the panel now sits under the stage. The slot is tried first,
then the panel, then the legacy header, then a floating mark. One injection now
serves both widths, because the layout carries the band. The same version also
stamps the language version from the language file this site publishes, rather
than leaving it pinned in the fragment: the band prints that version, which
makes it a public claim, and the fragment cannot read a file at runtime. It is
the same cure as lifting the mark from index.html. The stamp states which
language file is published, not what the engine implements, and that gap is
A0019.
build_glasshouse.py went to v1_002. The frame page's closing note read that the
Glasshouse was in test at version V1_00 while V1_03 was the deployed build. The
version is now read out of the studio fragment, so the page cannot claim a
version again.
Episode 6 was retitled from How ready is a language? to How ready is a
project?, a broader door for a reader arriving from LinkedIn, with the opening
narrowed to the question every project eventually faces and the close extended
to the rung it gives them. The Code Technician seat was named Cody and the Team
page carries it, the diagram block reading CODE TECHNICIAN . CODY.
Publish 1_062 carried the retitle, the Cody name and the records resync.
Publish 1_063 carries the V1_04 studio and both generator corrections. The
Glasshouse remains hidden, noindexed and out of the sitemap by standing ruling.
gaad_web_check passed before each cut, and the regeneration was again proven
faithful: exactly five files differ from the tree as it stood, and every other
page came back byte-identical.
Searching the AFAL language file at v0.004 for the vocabulary the studio shows:
Organisation, no hits. Preset, no hits. Species, no hits. The six families,
twenty species and six presets on the page are bench vocabulary, drawn beneath
a band that reads AFAL 0.004, and every genome written there records that
pairing.
This is not a rule broken. F044 holds that nothing becomes language rule until
it has its own research behind it. The point is narrower and worth ruling on:
the studio states a language version beside a taxonomy that version does not
cover. Whether family is a language concept at all is open, since nothing
branches on it and it is not in the genome.
Three, all recorded by the bench rather than found later. A merge regressed
against an approved mockup, by compressing the desktop panel instead of
reproducing the approved shape. A highlight was put on the button rather than
on the readout line, from misreading which element changes during a walk.
Third, a cached iframe produced a false defect report that was caught before it
was reported, with a second instance the same day where probing a
frame-driven canvas in the same tick made three species look identical. The
rule taken from both: when a test says something failed, check the test before
checking the code.
Touch is untested. A narrow iframe gives the right viewport and no fingers, so
the scroll-instead-of-orbit fault seen on a real phone is unverified. The phone
has no pinch zoom. The collapse and bloom warning went off the phone with the
block below the status line, so a degenerate form now gives no notice there.
The harmonised gene limits are still unruled, with only Points moved today.
Preset genomes still record the last roll's seed, which never made that form,
and the two hosts' preset values still differ. Blender is untouched at v3_042
and has seen none of today's rulings.
Publish 1_063 went to the server so the layout could be tested on a handset,
and it was, within minutes. The device found a fault no bench test had. The
viewport was a fixed 58dvh while the panel was a fixed height, so on a 932 pixel
screen the sum of a 58 percent viewport and a 286 pixel panel fell short and the
remainder showed as two empty strips, one above the controls and one below the
buttons. It did not reproduce at 390 by 780, which is why it reached the server.
The lesson is one line: a fraction cannot know how tall a panel is. V1_05 gives
the panel its content height and hands every remaining pixel to the view, and at
430 by 932 that is 932 equals 714 plus 218, with nothing left over.
Two further faults from the same screenshot were fixed in V1_05. The line
stating the genome format version and the one law float64 claim was invisible at
phone width, and it now sits at the foot of the panel, since it cannot go under
the band, which is fixed over the geometry. The warning line now stays on a
phone while the heading, buttons and status still go.
V1_06 followed and closed the day. The drawings under the stage became a view
switcher of four tiles, 3D, TOP, SIDE and DIMS, each taking the main view to
that camera with the active tile in gold. There is no second renderer behind it:
top and side are the same projection at elevation PI/2 and 0, which is what the
thumbnails already drew, so a tile and the big view cannot disagree about the
form. DIMS is the side view carrying height and diameter, one drawing routine
serving both the big view and its own thumbnail, both reading the same values
the readout prints. The dimensions come off on the first hand orbit, which is
D220, and spin now resumes only on the oblique, because it turns the azimuth and
would otherwise rotate the reader off the square-on view they just asked for.
One bench defect surfaced while that was built. The active tile was synced only
from the thumbnail redraw, which a hand drag never calls, so an orbit left TOP
or SIDE still lit. Cosmetic with three tiles; with DIMS it would have left
dimension arrows drawn on a foreshortened form. It now hangs off the one
function every drag, wheel and release already passes through. The bench
recorded that it had been stated as verified when it was not.
T038 was filed saying a collapse warning had been lost at phone width, on the
evidence of a flat Phyllotaxis disc. The bench challenged it and was right on
both counts, and the file was read rather than the challenge taken on trust.
There is exactly one warning path in the web host, a Phyllotaxis check
comparing the bloom's outer radius against the reach of its own roll window, and
no collapse warning has ever existed there. That is a Blender v3_042 feature.
The wording had been lifted from a stale comment inside the bench's own file
rather than from the code beneath it.
The form was not degenerate either. Phyllotaxis at Rise 0 is a sunflower head,
which is what the Sunflower preset is by design. Computed here from the file's
own roll window, the threshold is 5.0 times the square root of 799, which is
141.3 m, against that form's 2.0 times the square root of 499, which is 44.7 m.
The warning was right not to fire. T038 is reworded to the real fault, the warn
line living inside a box the phone hid, and marked done in V1_05.
Miles ruled the house rule retrospective, on the condition that no record be
damaged. Measured first: 1366 em and en dashes across 81 files, 132 of them on
the Records Hub, which is the page every homepage card and every graph dot
points into. Fixed at source in stages and never in the copies, because a sweep
of generated files regresses on the next build: five record JSONs, eleven record
generators which were emitting a dash as the empty-cell placeholder and in every
page title, the hand-maintained pages, then nineteen back-filled diary pages
regenerated from the cleaned JSON. Every stage was proved position by position,
the file length unchanged and every differing character a dash becoming a
hyphen. The count fell to 8, all inside an archived bench handover, which is
history and the bench's own file under D212. One exemption stands, the marker
drawn before an open disclosure block, which is a glyph rather than text.
Five website cuts went out on 29 July. 1_062 carried the Episode 6 retitle, the
Cody name and the records resync. 1_063 put V1_04 on the server for the phone
test. 1_064 carried the dash sweep. 1_065 carried V1_05 and the T038 correction.
1_066 closes the day with V1_06 and the day's records. V1_04 and V1_05 are both
superseded and neither should stand on a server. gaad_web_check passed before
every cut, and the Glasshouse stayed hidden, noindexed and out of the sitemap
throughout.
D216 rules that a host may offer a control that is not a gene, provided it
converts on the way in and cannot ask for a value a host would clamp. D217 rules
that a control needed in two layouts is moved rather than copied and
synchronised. D218 rules that the studio page states nothing this site cannot
show you, which is what both generator corrections implement. D219 makes the em
dash rule retrospective across the estate. D220 rules that a dimension is drawn
only where it is true, which is the scale bar reasoning of 27 July generalised.
F057 records the phone layout and closed the day at built, with the real device
test written into it, which retires both caveats it was filed with. T038 stands
corrected and done. P077 records the panel reaching 218 pixels with nothing
taken out of the language, and now carries the device verdict rather than the
measurement alone. Q052 keeps Miles's line about the rig having regressed from
the approved mockup.
Four of the bench's questions became actions rather than decisions, because they
are the Chief Architect's to rule and not this seat's to invent. A0032 asks
whether species, family and preset belong to the AFAL language, sharpened by the
bench into two problems with opposite consequences: family is displayed and
means nothing, while species means everything and is undeclared. A0033 asks
whether gene limits are language or per host, and whether a host may clamp on
import. A0034 asks whether Height should hold when Points changes. A0035 books
the repository session for Monday 3 August. A0036 books Blender coming up to the
day's rulings tomorrow.
Releases filed for bundle 0_147 and for website publishes 1_062 through 1_066,
clearing the lag between a cut and its record.
| Action | Category | Priority | Owner | Status | Title |
|---|---|---|---|---|---|
| A0019 | AFAL | high | Claude | open | The language file is behind the engine |
| A0024 | AFAL | high | Claude Code | part done | Validate the Shell range slice, tune the emergence window, roll to five |
| A0036 | AFAL | high | Miles + Cody | open | Bring Blender up to the day's rulings |
| A0020 | AFAL | medium | Claude | open | T1 research for the bench species |
| A0021 | AFAL | medium | Claude | open | Surface topology: the tier that makes a panel system possible |
| A0027 | AFAL | medium | Miles + Code Technician | open | Itemise the v3_029 issues Miles saw in Blender |
| A0034 | AFAL | medium | Miles | open | Rule whether Height holds when Points changes |
| A0025 | AFAL | low | Claude Code | open | Geometric NATURE row on the Singularity panel |
| A0017 | FILMS | medium | Miles | part done | Film 001 - The Rebirth: storyboard v3, real GAAD footage |
| A0011 | GALF | high | Miles | open | Rule the fourth element: does a relationship carry a rationale |
| A0012 | GALF | high | Miles | open | TWAL s6: map the three retired types |
| A0013 | GALF | medium | Miles | open | ARAL s6 declares types but states no triples |
| A0033 | Genome | high | Miles | open | Rule the gene limits, and whether a host may clamp on import |
| A0029 | Languages | high | Miles | open | Rule T1 readiness for AFAL, and whether the float64 engine rebuild is the prerequisite |
| A0032 | Languages | high | Miles | open | Rule whether species, family and preset belong to the AFAL language |
| A0030 | Languages | medium | Miles | open | Rule the waisted phyllotactic species - the pine cone question |
| A0031 | Languages | medium | Miles | open | Rule the preset faults: seed provenance and the two hosts' differing values |
| A0001 | MHAL | high | Miles | open | Midtown East recount under the D181 tax block rule |
| A0015 | MHAL | medium | Miles | open | The prototype carries MHAL verbatim at v0.003 |
| A0014 | Platform | high | Claude | part done | The language-file generator: JSON to HTML, in the pipeline |
| A0022 | Platform | high | Claude | open | AFAL has no assert contract and no Validate steps |
| A0035 | Platform | high | Miles + Claude | open | Git setup session, Monday 3 August 2026 |
| A0016 | PLATFORM | medium | Miles | part done | Fly Past on generator strings |
| A0008 | Platform | low | Miles | open | Prototype/ carries stray ARAL versions |
| A0007 | Website | medium | Claude | part done | Canonicals for generated record and language pages |
| A0023 | Website | medium | Claude | open | Episode 6: AFAL and the singularity |
| A0006 | Website | low | Miles | open | Latest episode first - name the surface |
| A0010 | Website | low | Miles | part done | Website drifts from the bundle - accepted, not overlooked |
Five actions were added today, A0032 to A0036. A0023 still names Episode 6 as
AFAL and the singularity, which is not the episode that shipped, so that subject
remains unwritten.
Touch is now proven, since Miles drove the layout on his own phone twice, but
the 14 pixel slider thumbs are still unconfirmed against Apple's 44 pixel
guidance, and the phone has no pinch zoom. The Phyllotaxis bloom warning is
still recommended for retirement. The harmonised gene limits are unruled with
only Points moved. Preset genomes still record the last roll's seed, which never
made that form, and the two hosts' preset values still differ. Blender stands at
v3_042 and has seen none of the day's rulings, which is A0036 and tomorrow.