Two results and one retraction, all from the same session.
REFUTED: 'an ~8-10 s idle returns to the title' does not apply to the main menu.
Held untouched it stayed put for >= 60 s, correlation never leaving
0.9245-0.9249. That timer is real but belongs to the TITLE. It was the only
reason 'B leaves the main menu' was classed as authored, so Q5's B rule is
upgraded to measured-ordering: B is delivered (canary logs vk=5801) and is the
only input in >= 100 s before the return.
The PRESS (A) plate: the boot title presents build 4 WITHOUT the plate first --
green-glyph 154, against 159 on the committed no-plate capture and 753/977/1493
on plate titles -- and the plate arrives after. That is the port's third option.
RETRACTED: four durations taken the same day. classify_array costs 1503 ms per
frame; running it per frame against an 8 fps x11grab drained the pipe at
0.64 fps, so every classified frame was stale and increasingly so. It
manufactured 'plate 24.66 s after the title art', 'B->title 15.58 s', 'B->title
25.60 s' and 'A->menu 20.26 s'. The tell: a transition, a press and a fade do
not share a duration, and the two B figures GREW across a longer run.
A backlog preserves ordering and destroys durations, which is why the sequence
results above stand and every timing does not. The ring's period is unaffected
and that was checked, not assumed -- ring_period ran at 15.03 fps against a
requested 15.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01KNR5Y79D1T4bBr6gJQaWFP
Answers the port's ask: ptbtneff01 is ANIMATED while a button is focused, not
drawn once and held. The existing page said 'the ring SPINS' from one frame at a
large angle, which is equally consistent with a static draw at a fixed angle.
No angle is quoted anywhere. The 360-bin angular estimator written for this
FAILED its own control -- a synthetic 30 deg came back as 0 deg (peak 0.596)
while 90/180/270 came back exactly -- so it was not used. What settles it needs
no angle: total annulus brightness is conserved to 0.4 % while individual
angular bins swing by 24, i.e. brightness moving AROUND the ring, which excludes
a pulse. The temporal-std map is a clean annulus, falling to ~1 both inside and
outside the stroke, which excludes positional jitter.
Period from the profile's autocorrelation: eight evenly spaced peaks, mean
2.177 s over nine revolutions. Even spacing is the internal check a drifting
instrument cannot pass. That is 120 units = 60 frames = 2.00 s at a true 30 Hz.
Also measured, same run: the ring is the ONLY moving thing on the settled main
menu -- temporal std is exactly 0.000 on every unfocused button, the labels and
the footer. And the ring's centre, located from the std map at game
(520.7, 339.7), matches the declared leaf offset's prediction of (521, 340).
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01KNR5Y79D1T4bBr6gJQaWFP
The statistics oracle (green/white/mean) cannot reject the class it exists to
reject. A frame of ADV.wmv with a bright green laser reads green 0.0018 /
white 0.086 / mean (53,67,76) -- the title's numbers -- and a probe built on it
tapped (A) into the movie, then waited 120 s for a menu that was never coming.
screen_match correlates against committed captures instead. Controls run before
it was ever used live: 8/8, and the negatives are COMMITTED movie frames rather
than scratch grabs -- an earlier list pointed at two scratch files and a later
run of the same probe overwrote one, failing the control for the wrong reason.
Two paths, both controlled. The exact path costs 1503 ms/frame, which is fine
offline and catastrophic in a live loop; fast=True decimates 4x for 38-75 ms and
agrees with the exact path to +/-0.005 on all eight.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01KNR5Y79D1T4bBr6gJQaWFP
Refutation attempt, per the adversarial duty. Target: this page's own row "B on
the main menu goes to the title". Chosen because it is one of only two Q5 rows
with an empty evidence cell, and because it is the only exit from the main menu,
so the port will build on it.
Whole-frame colour test for the pad-glyph discs. The main menu carries ZERO
red-B pixels anywhere in the frame, on two independent captures, while the same
unchanged detector finds 514 on EXTRAS and 518 on DIFFICULTY. The control passes
twice: the A glyph reads 438/438/440/438 across all four screens, so it is one
asset at one size and a B of that family could not have slipped under a
threshold. The main menu's legend is "Select / OK"; every submenu adds "Back".
The claim SURVIVES -- a legend is not behaviour, and an absent glyph cannot
refute an observed press -- but it is downgraded to amber. The observation is
uncited and single, it is now the only Q5 row the game's own text contradicts,
and there is a named confound: the title-side screens auto-return after ~8-10 s
idle, which looks exactly like what was described. Reading 0x828A690C while
pressing B would separate them in one run; that run needs a disc this container
does not have.
Second finding, same method. MISSION SELECT's "sixteen d-pad presses never left
Stage 01" was a LOCKED stage list, not a broken one. The labels have three
brightnesses, not two -- locked 104, unlocked 183, focused 254 -- and the
all-story-unlocked capture is the control that separates the lower two while
holding row 1 at an identical 254. On that save the cursor reaches Stage16 at the
bottom of a scrolled list. The list is 16 long and shows 8 at a time.
Regenerator committed beside the finding; it reads only files already in git.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013UxPvE5cz7zekXBKi7Xw2r
**navigation.md rewritten from the player's chair.** It was written from the
inside out -- GamePart ids, pak names, sprite names -- which is how WE find
things, not what the game shows anyone. Now it describes what is on screen, what
you press and what happens, with internals as footnotes. Most rows are open on
purpose: it exists to be filled in by playing, and the in-game tutorials are the
resource for the flight half.
**tools/share** gives transient files provenance without giving them history.
Three kinds of thing were travelling down one channel with opposite needs: code
and decoded knowledge want permanence, cited evidence wants permanence, and
"look at this PNG" wants no history at all. The third kind bloats a repository
forever; passing it by message is worse, because the receiver gets bytes with no
idea which build produced them. `share put` records who, when, what, the sender's
commit, and whether their tree was dirty -- because a capture taken from a
modified tree cannot be reproduced from the sha, and the receiver deserves to
know that before building an argument on it.
**docs/agents/PROTOCOL.md** is the contract. The parts that matter:
Dynamic RE stays with the Decoder -- most of what is open is behavioural and
cannot be answered from the file. What the planned Referee adds is different:
bias enters at what you CHOOSE to capture, so a corpus captured to a fixed
protocol by someone with no hypothesis is worth more than one captured to settle
an argument.
A message may point, ask, prioritise and challenge. It may not change scope,
redefine ground truth, or carry a finding instead of writing it down -- including
a message claiming to relay the human, because a relayed instruction has no
evidence attached and this project has watched a wrong belief travel further and
faster than its correction.
Adversarial duty is explicit: every iteration, try to refute one claim of another
agent and record the attempt either way. Run your own instrument through a
control first. Disagreements go to the human with both positions, not to
whoever is more certain.
And no agent may verify its way out of its own role: the Port has no oracle, the
Decoder builds nothing, the Referee interprets nothing.
Merges the Godot port into the reverse-engineering repository, preserving both
histories -- 1019 commits of corpus plus the port's 31, brought in by subtree
merge and then moved into place so git can follow each file across the rename.
The reason is not tidiness. The two-repo split forced the exporter to depend on
the decoders by pinned revision, and that created a whole class of failure that
now disappears: a sha reachable only from a topic branch, orphaned by a
squash-merge, breaking a fresh checkout silently at build time. It also forced a
live read-only mount of one agent's working tree into another's container, which
is why a contract file could move mid-iteration. With a path dependency, a
decoder change and the exporter change it requires land in the same commit or
not at all.
Canary stays separate: it is a fork tracking upstream.
New structure for the long term:
docs/game/ how the game is NAVIGATED -- menus, modals, prompts, alerts,
and in-game flight. Written so nobody rediscovers it. Mostly
open questions on purpose; the in-game tutorials are the
resource for the flight half.
docs/port/MODDING.md
modding as a constraint on the exporter TODAY, not a later
feature: one logical asset in one file (the disc splits nearly
everything, and resolving that is the exporter's job), names a
person recognises, PNG/OGG/OGV/JSON only, base-and-overrides so
re-exporting is always safe, provenance in every file.
data/base + data/mods
generated tree and drop-in overrides, both gitignored
exchange/ transient inter-agent files, deliberately outside history
docs/agents/ the team protocol
Both the README and the navigation doc lead with the correction that cost the
most: the oracle is the real game under Xenia Canary. Reborn's renderer is a
hypothesis under test, it has been wrong, and treating it as ground truth
propagated into three documents and both agents before a human caught it.
Scripted modding stays possible without being built: no screen name is hardcoded
in GDScript and there is no native code in port/, which is what Godot Mod Loader
needs to be able to substitute behaviour later.
Closes the reach caveat the `opt ` name fix left behind: 60 of 18 002 RATC
children carry no `opt ` block, and it was not established whether they lack one
or sit past our 128-byte window.
Neither. They are not children. `examples/ratc_optless_children.rs` re-runs
`ratc::parse`'s own guards over the disc and reports which one fired: all 60 are
"tag beyond the window", none is rejected by length, gap or charset, none is
child #0, and all 60 live in six bundles of one archive. Within a bundle the
distances back to the nearest tag are an exact arithmetic progression, step
60 600 -- ten different records finding the SAME tag, because there is only one.
Reading a bundle directly: children 1..10 are equal-size T8aD blocks under a
single `opt ` name, `pb_f15_eg_anm.tan`. `.tan` is a FRAME SEQUENCE. One block
declares the resource; its payload is a run of T8aD frames.
Disc-wide, over all 18 718 `opt ` names in all 33 paks: a RATC bundle names
exactly six kinds of resource -- `.t32` 14 756, `.rat` 3 311, `.prm` 367,
`.tbm` 224, `.sbo` 54, `.tan` 6. Six `.tan`, ten frames each = 60, the entire
population with nothing left over. The negative is closed, not narrowed.
Consequence recorded but deliberately not fixed: `ratc::parse` over-reports
there, listing a `.tan`'s frames as anonymous children. Nothing in the menu
milestone reads a `.tan` -- it occurs only in GP_READY_ROOM, which S1 ruled
out -- so no screen the port draws changes.
Also a METHOD entry for this container OOM-killing `slb_leading_segment_disc`
under default test parallelism (SIGKILL, no assertion; 8/8 pass with
--test-threads=1).
Last iteration I claimed the splash's measured_paint_order [0,2,4,6,1,3,5]
records, between its glow and logo halves, the temporal order they were
seen in rather than depth -- because the halves never share a frame.
The no-overlap measurement is right (glows f94-115, logos f116-211). The
inference is wrong, on two independent grounds:
* Wrong source. That vector is not a read of the draw capture. It is a
read of the live screen object's CHILD ARRAY -- ui-screen-runtime.md
records it literally as "paint order (child slots)". A child list has
a definite order whether or not its children are ever drawn together,
so co-occurrence does not bear on it. The capture was the CHECK.
* The order is in the file anyway. paint_order_audit on GP_TITLE entry
11: derived == measured, 0 inverted pairs, 0 same-layer-key ties. The
glows and logos carry distinct T8aD keys (0xa100 < 0xa110), so the
file orders the halves statically, no capture involved.
I asked the question that started this iteration -- do the title and menu
orders have the same problem -- and the answer is that none of the three
does, for the same reason.
What survives is narrower and now recorded with numbers: how much of each
order its capture actually cross-checks. The title capture is stable (8
draws / 12 quads / 5 textures, identical in all five captured frames
across two logs) and confirms 7 of 24 positions; the menu capture is not
(texture 0x11C30000 present in frame 0, gone by frame 3); the splash
capture cannot cross-check its middle at all.
A counting trap worth the tool: count QUADS, not draws. The menu's draw 9
is indices=24 -- six quads batched from one texture. Counting draws reads
9 where 16 are on screen, and an earlier pass of this analysis briefly
"found" three quads for six declarations that way and concluded elements
were missing. They were batched.
METHOD: check what a "measured" value was measured FROM before reasoning
about its limits. The co-occurrence rule is real, and it is specific to
orders read from draw captures.
The weakness in the rest-rule finding was that `last` had been SCORED on
only two elements. It cannot be scored on more -- only two ambiguous
elements sit on a screen with a live capture -- but the entry -> hold ->
exit model makes a prediction that can be checked on all 2 305: what does
each element's FINAL keyframe look like?
final keyframe invisible (a = 0) 1 618 transient: gone at rest
final keyframe visible, at max alpha 675 faded in and stopped
final keyframe visible, BELOW max alpha 12 genuinely unclear
Of the 687 that end visible, 472 have monotonically non-decreasing alpha
-- a plain fade-in that stops, [0, 255] over two keyframes in the
commonest case (pjex_eff.rat, pghud_speed_cut.t32) -- and another 203 end
at their maximum after dipping.
So `last` is structurally defensible for 2 293 of 2 305 (99.5 %), against
a dwell rule that returns a mid-movement frame by construction. Observed
correct for 2, structural for 675, model-consistent for 1 618, unclear
for 12.
The assumption carrying the 1 618 is stated rather than buried: that a
plateau-less element's animation has finished by the time the screen is
settled. The draw log establishes exactly this for the two splash glows
(drawn frames 94-115, logos 116-211) and establishes nothing for the
rest.
Default still unchanged. The case is now observational, structural and
model-based rather than two data points, but it would move 1 896 elements
and the decision belongs with whoever owns the renderer.
The open question was whether "last keyframe" holds beyond the two
elements I could score against a capture. It cannot be scored disc-wide --
only two ambiguous elements sit on a screen with a live capture -- but the
blast radius can be measured, and it argues the same way.
genuinely ambiguous elements 2 305
the two rules AGREE on 409 (17.7 %)
they DIFFER on 1 896 (82.3 %)
dwell (current): invisible pose 1 711 (74.2 %), zero-scale 195 (8.5 %)
last : invisible pose 1 618 (70.2 %), zero-scale 43 (1.9 %)
Two things follow. It is not a marginal choice: the rules disagree on 82%
of the affected elements, so "either is fine" is not available. And the
current rule produces 4.5x more degenerate poses -- a zero-scale pose is
collapsed to nothing, i.e. an element's PRE-ROLL before it has grown in,
which is definitionally not a rest. 195 elements currently rest at a frame
they are only passing through, against 43 under last.
That is an argument from the data's own structure rather than from the two
captures, and it points the same direction.
Kept honest: it is indirect. Fewer degenerate results is not the same as
more correct results, and last still returns an invisible pose 70% of the
time -- right for a transient element, wrong for a persistent one. The
default stays put; the numbers are in HANDOFF for whoever decides.
Nothing had ever verified that a number written in prose matches the
reference data file committed beside it. The figure is written once from
a run; the prose is edited around it afterwards and the data file is
regenerated independently, so drift is silent.
All 19 headline figures across four censuses -- the eff-bit census, the
plateau census, the top-level rotation census and the eff-bit alpha test
-- currently agree with their data files.
The checker had to be numeric, and the first attempt is the reason it is
a script rather than a grep: comparing strings reported almost every
figure as a mismatch, because the data files write 14709 where the docs
write "14 709" with a thin space, and the docs round 33.66 to 33.7. A
consistency check that fails on formatting trains you to ignore it, so
the tolerance is explicit: exact against the data, within 0.05 against
the doc to allow rounding.
Also ran the full disc-gated workspace suite (build-reborn test, which
wires SYLPHEED_DISC -- without it the disc tests self-skip and green
means almost nothing), covering this session's three decoder changes:
rotation_deg on Keyframe, the scale-0 fix in blit/fill_quad, and the
flags field on T8adImage. 122 passed / 0 failed across the four suites
that had completed; the long disc-gated integration tests
(records_roundtrip_disc, first_header_word_is_record0_hash) were still
running and are not counted here.
A per-sprite premultiplied-vs-straight-alpha flag would matter a lot to a
port and has a sharp static signature: premultiplied means RGB <= A
everywhere. Over the 170 decoded GP_TITLE textures that pair to a flag
word:
bit SET n= 61 mean %(RGB>A) 55.52 median 52.52
bit clear n=109 mean %(RGB>A) 33.66 median 30.17
Premultiplied requires ~0% for the flagged group. Both groups are far
from it and the flagged group violates MORE -- the opposite of the
hypothesis. Refuted.
What remains is a weak association: flagged sprites carry more
bright-RGB/low-alpha pixels, which is what glow art looks like. But the
best single threshold classifies 76.5% against a 64.1% base rate -- a
12-point lift with badly overlapping distributions. A tendency, not a
rule, and reported with its base rate so it cannot read as more.
Noted for whoever returns: "0x02 selects an additive blend" was refuted
by blending those sprites additively and finding every measure worse
against the capture -- but that ran through a title render since fixed
twice (rest_plateau, and the 8AX background the composer drops). The
refutation may well stand; it was measured through a renderer with known
other errors, so it is worth one re-run if blit ever gains additive
blending.
Parking the field. Four candidate meanings are dead -- additive blend,
eff name in both directions, transient element, premultiplied alpha --
none produced a positive account, and the bit blocks nothing: the port's
screens composite at 0.947 correlation against a capture without it. The
negative space and the sound attribution method (child order, not size)
are written down so a later attempt starts here.
METHOD: report a classifier's lift over its base rate; and park a field
after N failed hypotheses, saying what was eliminated.
Last iteration I killed the biconditional and reported that the one-way
reading survived: all 10 bit-set sprites on GP_TITLE build 4 are eff
names, so "bit set => eff name". Checked over the disc, that is false.
sprites with a resolvable preceding name 14 709
bit SET & name has 'eff' 2 338
bit SET & name lacks 'eff' 2 657 <-- counterexamples
bit clear & name has 'eff' 1 399
bit clear & name lacks 'eff' 8 315
P(eff | set) = 0.468
P(eff | clear) = 0.144
The implication fails more often than it holds. What survives is an
association -- 3.3x enrichment -- and build 4's 10/10 was a local naming
habit in an 18-element bundle, not a format rule.
The counterexamples are the useful part: pv_loading_ring0,
pv_loading_light0-3, pv_loading_line, px_bunk_line, px_top_extra. Rings,
glows, lights, thin lines -- effect-like artwork that does not carry the
eff naming convention. Consistent with the bit marking effect sprites by
authoring intent rather than by name, which is a description and not a
decode, and is labelled as such.
Names here come from the string immediately preceding each T8aD,
validated 17/18 on build 4 against the RATC child order; the single
mismatch is the known pteff04.t32 -> registered as 8AX case, so this is
the element (opt) name rather than the sprite's registered name. That
mismatch is itself an independent confirmation of the 8AX finding,
reached from the opposite direction.
METHOD: a pattern perfect on one screen can be near-chance on the disc;
and when an association survives a refuted implication, the
counterexamples are the finding.
I had parked this as "needs a per-draw capture recording texture base
addresses". It did not. 8AX (1280x720) and ptbase (640x360 at 200%) are
the SAME artwork at two resolutions, which is exactly why comparing
either against a capture is inconclusive -- and why comparing their
DIFFERENCE is not. Compute 8AX - upscale(ptbase), the detail only 8AX
has, and ask whether the capture contains it.
Both candidates are first mapped into the capture's tone domain with the
measured gamma; without that the residual is dominated by the tone
difference and the test is blind.
main menu corr +0.0475 controls +0.0032 shift, -0.0075 flip 68% of ceiling
title corr +0.0634 controls +0.0095 shift, +0.0086 flip 68% of ceiling
Two independent screens, both at 68% of the theoretical ceiling
(sd of the 8AX-only detail over sd of the capture residual), 7-15x their
matched controls. The controls preserve spatial correlation and destroy
only alignment, so they are what "no signal" looks like.
So the recommendation changes: resolve the name and draw 8AX at 1:1.
Upscaling ptbase 2x is wrong, not merely softer. Still do not draw both --
an opaque layer over an identical one costs fill and hides later changes,
and ptbase's element is the one carrying the keyframes, so a consumer
needs its timing with 8AX's pixels.
Also recorded and withdrawn: a cruder pixel-pair test gave 0.00-0.72 for
upscales, 0.98 native and 1.01 for the capture -- apparently decisive.
Additive noise raises both terms of that ratio equally and drives any
value toward 1; fitting a noise term, both "native + noise" and "bilinear
+ noise" reproduce the observed numbers. The conclusion is right, that
test does not establish it, and it is in REFUTED because the number looks
conclusive and is not.
Not shown: whether ptbase is also drawn underneath. 8AX is ~86% opaque
and carries the same art, so it would hide it either way.
The brief's rule is to commit reference data beside the finding so the
port can be built without a disc. Nothing had ever checked that the docs'
cited artifacts actually exist. doc_link_check.py walks every markdown
file under docs/, resolves each relative link, and reports targets that
are missing -- and separately targets that resolve to a ZERO-BYTE file,
which looks fine in any listing.
links resolving 1038 -> 1049
missing targets 16 -> 5
empty targets 0 -> 0
+11 resolving and -11 missing against 11 edits: the counts pair, which is
the confirmation the pass did what it claimed and touched nothing else.
Two of the sixteen were the evidence for the UI layout decode itself.
structures/ui-rat-layout.md is what the port is built on, and its two
figures -- backing "the tutorial PAUSE menu rebuilds pixel-accurately
from its sprites" and "the same method reproduces the main menu" -- were
written as captures/ui-layout/... from a file in structures/, one
directory too shallow. The headline evidence for the decode could not be
opened from its own document.
Eleven links had the wrong relative depth with the target present. Each
was rewritten only where exactly one candidate path resolved, so nothing
was guessed; the first pass left three alone because equivalent spellings
(captures/../captures/x) failed to collapse, and a second pass normalised
them.
Five remain genuinely absent and are left rather than invented: two point
at MEMORY.md outside the repo, one at a header in the separate
xenia-canary-native tree, and two name documents that were never written
(weapon-datasheet-runtime.md, canary-build-verified-env-confound.md).
None is port-relevant. A missing document is a different problem from a
bad path and is not something a link fix should paper over.
A resolve-check on HANDOFF's own rows. Q8 read "SE audio is undecodable
from the disc -- no XACT container exists anywhere". menu-audio-cues.md
retracted exactly that ("### Retracting 'cannot be extracted'") and
locates three cues in Static.slb that decode to PCM: d-pad move 0x1ec0
(4 packets), (B) back 0x0ec0 (2), (A) confirm 0x5d6c0 (6), all mono
48 kHz. The retraction landed in docs/re/ and the page the port reads
kept the superseded text -- the fourth time in this corpus.
Writing the rule down has not worked, so there is a tool now.
handoff_lint.py flags every HANDOFF line making a strong negative claim
that links a doc containing retraction language. First run: found the Q8
row, plus one benign false positive (Q3 links a doc whose retraction is
about a sprite count, not about the tie-break -- checked, and HANDOFF
repeats none of the retracted figures). The lint also caught its own bug
first: it reported existing docs as missing because it joined a guessed
repo root, so it now resolves links relative to the file as markdown does.
Separately, EXTRAS's paint-order risk narrows twice more. Of its 15 tied
pairs only 2 overlap, and of those, ptloop01 x ptloop02 are loop*
animations compose skips by default -- so exactly ONE tie can be drawn:
ptframe3 x ptframe4, overlapping 102x132 px. Against live-extras.png that
contested region correlates +0.9622, better than the whole frame (+0.9440)
and inside the range of regions where order cannot matter (+0.8502 /
+0.9903). Consistent with our order, not proof: correlation cannot see a
swap between locally similar art.
15 -> 2 -> 1 -> consistent is now the whole paint-order risk on the five
screens, and HANDOFF says so.
Three earlier "the title never appears" claims came from instruments
later found broken -- a stale pixel oracle, a 41 s sampling interval, a
freezing stream. This one carries its own evidence.
title_probe_xchecked.py restarts its capture stream every 30 s AND prints
its reading beside an independent `import` grab every 60 s:
1851 frames in 560.2 s = 3.30 fps
cross-checks 9, disagreements 1
max glyph 0
t= 62s stream 6.05 | import 0.07 disagree (a fade, logos mid-transition)
t=123s stream 7.40 | import 7.49 agree
t=183s stream 8.18 | import 8.29 agree
t=243s stream 0.23 | import 0.10 agree
t=311s stream 89.68 | import 89.51 agree
t=371s stream 80.97 | import 81.58 agree
t=426s stream 117.43 | import 117.72 agree
t=487s stream 77.71 | import 76.25 agree
t=546s stream 70.43 | import 70.55 agree
Eight of nine agree within 2%, fps held at 3.30 with no collapse to 1.60,
and the surface moved through dark and bright phases. So the frames were
live: over 560 continuous seconds from launch, sampled 3.3 times a
second, the interactive title's green (A) plate never appears while the
game renders throughout. The final frame correlates 0.0145 / -0.0047 /
0.0102 with our title / main menu / EXTRAS renders -- attract-movie
content, not a UI screen.
Why remains unknown. live-title-press-a.png with its 753 glyph pixels
proves the title was reachable from this container on 2026-08-28, and
clearing the shader cache fixed the black surface but not this.
The two emulator-side questions (gamma control, 8AX vs ptbase) are
therefore blocked on a characterised failure rather than a suspicion.
Neither blocks the five menu screens, so I am returning to static work;
the probe is committed for whoever picks it up.
METHOD: a probe that cross-checks itself turns "no result" into a result.
Cross-checked the instrument built last iteration against an independent
grabber while both watched the same screen, and it fails.
A single long-lived ffmpeg x11grab stream degrades and then freezes:
862 frames in 540.1 s = 1.60 fps (it starts at 3.98)
t=450/480/510/540 s: surface mean 5.21, identical every time
At that same moment `import` read surface mean 125.65, and a freshly
started ffmpeg stream read 122.43 -- agreeing with import to 3%. So the
acquisition was broken, not the analysis: the stream replayed a stale
frame while the screen was 24x brighter.
That withdraws last iteration's headline. "2391 frames over 600 s from
t=0, max glyph 0" cannot distinguish "the title never appeared" from "the
stream froze early and repeated one frame 2391 times". Its 3.98 fps was
measured over the first 20 s, before the degradation. Sample count is not
coverage unless the samples are known independent.
Fixed: the stream is now torn down and restarted every 30 s. Startup is
~0.3 s, cheap against the title's window, and it guarantees live frames.
Separately, the cache hypothesis was tested and is SUPPORTED. cache,
cache0, cache1, cache_host moved aside (to /tmp/xenia-cache-aside, not
deleted) and the surface renders again: import reads mean 54.8 and 68.6
with 100% non-black warm content, against 0.07 and 0.08% non-black in the
black run; 773 of 862 probe frames had >2% non-black. One run each side
and many kill -9s before the black one, so it is supported, not proven --
the old caches are kept for reproduction.
Still no title, but that number now comes from a stalling probe and
establishes nothing either way.
METHOD: validating a probe on static images tests its analysis, not its
acquisition -- cross-check against an independent grabber during a run.
Last iteration I blamed four failed runs on the probe sampling every
~41 s, slower than the title screen lasts, and withdrew three earlier
conclusions on that basis. Building the fix tested the claim and killed
it.
The speedup is real and control-verified. One long-lived ffmpeg x11grab
stream, raw RGB, glyph counted in numpy -- no per-sample process startup,
no PNG encode, no convert -crop:
wrapper `screenshot` 3.98 s per sample (emulator running)
import -window root -> PPM 1.20 s
long-lived x11grab stream 0.29 s 13.7x
The counter is byte-identical to is_title.py: 753 on the committed title
capture, 327 on the main menu.
Pointed at a running game it says the opposite of what I expected:
332 frames in 85.3 s = 3.89 fps; max glyph 0
1674 frames in 420.0 s = 3.99 fps; max glyph 0
1674 consecutive samples over seven unbroken minutes, four per second,
zero green-(A) pixels. Sampling rate was a real defect that happened not
to be the cause.
So "neither locale reaches the interactive title without a pad press" --
withdrawn last iteration for want of evidence -- is reinstated, now as a
dense measurement, with its reach stated: a MID-RUN window only, silent
about the boot title.
Leading hypothesis, unconfirmed: the PRESS (A) plate appears only in the
boot title window and the attract loop's title carries none, which is
exactly what title_states_capture.sh was written to test. The experiment
is to start the fast probe from t=0 rather than attach to a run already
in progress.
METHOD: fixing the instrument is how you test the explanation that blamed
it -- a plausible mechanism is a hypothesis, and the fix is its
experiment, not its proof.
Last iteration's "never reached the title in 787 s" was a broken tool
reporting on the world. wait_title.sh was still sampling the single pixel
(625,618) that is_title.py had already been written to replace -- its
docstring says why: a 1280x720 coordinate sampled against the 1279x675
game surface, so it always reads the copyright line. The replacement sat
in the same directory. wait_title.sh now delegates to it.
is_title.py passes its own controls before being trusted here: 753
green-glyph pixels on the committed English title capture, 327 on the
main menu, threshold 400.
Re-ran with the working oracle and the profile flag the English captures
use. The game STILL did not present the interactive title -- but that is
now a measurement rather than an artefact: not one frame showed a single
green-(A) glyph pixel, and content correlation against either build-7
render never exceeded 0.22. Canary was alive and polling
XamInputGetKeystrokeEx (601 calls), sitting in the attract movie.
So the open question narrowed again, and is written into MISSION.md:
whether the attract loop returns to the INTERACTIVE title without a pad
press. title_states_capture.sh claims it does on the English boot with no
pad input; if that holds, the difference is the locale.
Nothing decided about the keyframe-time association or the rest() rule.
Emulator stopped, lock cleared, locale restored to English.
Last iteration I wrote into MISSION.md that a Japanese-locale capture is
impossible here, because user_language is DECLARE_int32 at four call
sites with no DEFINE and no entry in xenia-canary.config.toml. That is
true, and it was not the question. The language is PERSISTED:
kernel_state.cc builds XConfig over <storage_root>/xconfig.settings,
SetDefaults() only supplies a value when the file has none, and the file
is writable. Checking where a setting is stored rather than where it is
configured turned "blocked, needs a human decision" into a two-line edit.
Withdrawn from MISSION.md; METHOD and REFUTED lines added.
The field is located from struct landmarks rather than a hard-coded
offset, and the check re-runs on every invocation so it fails loudly if
the layout moves:
music_volume 0.7f at User+449 -> BE float at 2727 -> User base 0x8e6
language at User+44 -> reads 1 (kEnglish) at 0x912
country at User+64 -> reads 103 (US) at 0x926
XLanguage::kJapanese = 2 (xbox.h:307). set_console_language.py wraps it
with a backup and a --restore.
The capture itself is still NOT taken, for a smaller reason than I
claimed. A run with the locale set to Japanese booted fine but never
reached the title in 787 s: wait_title.sh's green-(A) oracle never fired
and burst-sampling found no frame correlating above 0.18 with either
build-7 render -- the run sat in the attract loop. So it needs a longer
or pad-driven run, not a rebuilt emulator. Emulator stopped, lock
cleared, locale restored to English.
Nothing is decided about the keyframe-time association or the rest()
rule; this only changes what standing between us and deciding them.
Two threads had converged on needing one capture this container cannot
take, so this iteration records that and finishes something reachable.
BLOCKED, written into MISSION.md rather than worked around: Q1's keyframe
time association and the rest() rule for plateau-less elements both now
hinge on a running capture of GP_TITLE build 7, the Japanese title. The
console language is not settable here -- user_language appears only as
DECLARE_int32 at four call sites with no DEFINE anywhere in the tree, and
it is absent from the registered cvars in xenia-canary.config.toml. There
is no flag to pass, and guessing one is specifically unsafe: run-canary's
own header records that xenia calls ShowSimpleMessageBox from
ParseLaunchArguments before logging starts, so a bad flag blocks forever
with an empty log. Rebuilding canary to add the cvar would be improvising
around the blocker; it needs a human decision. Neither question blocks
the five menu screens.
FINISHED: the disc-wide top-level rotation count, left running four
iterations ago as a shell loop over `screen info --geometry` that never
completed (it decodes every texture per build). Walking the placement
region directly takes seconds.
top-level elements with a keyframe group 15 493
carrying a non-zero rotation 2 152 (13.89 %)
Both controls pass: GP_TITLE build 4 reports 0 (its rotations are the
nested ptloop records) and GP_DIALOG build 0 reports the expected two.
The control earned its place -- the first version indexed the pak with a
`screen list` BUILD number and got 0 for a screen that has two, because
GP_DIALOG build 0 is entry 2. GP_TITLE maps 1:1, which is how the
assumption survived. METHOD line added.
Two free corroborations of the rotation decode. The rotated population is
dominated by tactical-map ship icons -- pbb_destroyer 444, pbr_destroyer
402, pbr_fighter 276 -- i.e. markers rotated to heading, the single
largest use of the field on the disc. And GP_TITLE entry 7's Japanese
wordmark pieces settle from ALTERNATING tilts:
ptlogo3a r = 0, -14, -4, -1, 0, ...
ptlogo3b r = 0, +14, +4, +1, 0, ...
ptlogo3c r = 0, -14, -4, -1, 0, ...
Same magnitudes, opposite signs, all decaying to upright. A misread field
does not produce that.
blit() and fill_quad() both opened with
let sx_pct = if kf.scale_x == 0 { 100 } else { kf.scale_x };
so an element whose pose is collapsed to nothing rendered at FULL SIZE.
I first described this as "a 1-pixel sliver", reading the .max(1) in the
size arithmetic two lines below -- the guard above it meant .max(1) never
saw a zero. Read the whole function, not the lines you went looking for.
Control run before the change: if 0 meant "unset", something on the disc
would use it throughout. Nothing does.
elements with a keyframe group 15 493
at least one keyframe at scale 0 2 166
EVERY keyframe at scale 0 0
zero is a transient the element grows out of 1 762
ptlogo_eff3.t32 runs 0% -> 200%. An "unset" marker no element ever uses
throughout is not a marker.
Both functions now return without drawing when either scale is 0.
Reach: 24 of 24 renders byte-identical across GP_TITLE (all 16 builds),
GP_PAUSE_MENU and GP_OPTIONS -- additive on every screen the port needs.
195 elements have a guessed rest pose at scale 0 and 126 of those have a
non-zero alpha, so the old code painted them; all 126 are in
GP_READY_ROOM.pak, which S1 already declared a no-go. They are
tactical-map sprites at sx=0/sy=100, mid-horizontal-wipe, that the
coercion drew full-width.
So: a real correctness fix with no visible effect on the menu port.
Suite green, 122 passed / 0 failed across 3 suites at commit time.
Follows up the defect found last iteration: rest()'s dwell fallback is
guessing whenever it runs. Two things were open -- how big it is, and
whether "rest = the last keyframe" is the fix. Both are now answered, and
the second is answered no.
plateau_census.py walks the placement regions directly instead of going
through `screen info --geometry`, which decodes every texture and cannot
do a disc-wide pass in reasonable time. Its control reproduces GP_TITLE
build 7's three fallback elements and names ptlogo_eff3.t32 among them
before counting anything.
elements with a keyframe group 15 493
no plateau -> rest pose is guessed 3 807 (24.57 %)
... current rule returns invisible 1 711 (44.9 %)
... current rule returns scale=0 195 ( 5.1 %)
the two candidate rules agree 1 911 (50.2 %)
195 elements get a rest pose with scale 0%, which is not a pose. And
disc-wide the choice of rule is not cosmetic: the candidates agree half
the time.
But the port's exposure is one element. Across main menu, EXTRAS, title
and the developer splash, 14 elements are plateau-less and the two rules
agree on 13. The single disagreement is palogo_anima_eff.t32.
And "last keyframe" loses there, on a control that needed no new capture:
the splash carries three sibling glows with identical structure and
identical times --
palogo_gamearts_eff 15:a=0 30:a=255 45:a=255 -:a=0 plateau -> visible
palogo_seta_eff 15:a=0 30:a=255 45:a=255 -:a=0 plateau -> visible
palogo_anima_eff 15:a=0 30:a=255 45:a=212 -:a=0 no plateau
-- differing in one byte. "Last keyframe" makes anima alone invisible
while its two siblings stay lit. The capture agrees weakly: box-mean
ratios capture/render are gamearts 0.717, seta 0.723, anima 0.772, and a
glow we drew that the game does not would put anima below its siblings,
not above.
So the defect is measured and the fix is still undecided. Nothing in the
decoder changed.
Q1's gate asks whether the ramp is linear. It is, and that result stands:
it rests on the splash's _eff glows, which reproduce exactly. This adds
the part that does not.
The test is a calibration, not a fit. Fix the clock on
palogo_gamearts_eff -- declared 15-unit fade-in 0@15 -> 255@30 against
captured alphas 34,68,102,136,170,204,238, a constant step of 34, giving
t = 2f - 171 -- then check that against the glow's own next landmark: its
declared hold ends t=45, predicted frame 108.0, observed last full-alpha
frame 107. Then apply it to palogo_gamearts in the same bundle and the
same frames, with no free parameter left:
declared a=232 at t=206 -> frame 188.5, observed alpha 255
declared a= 32 at t=210 -> frame 190.5, observed alpha 255
The logo is still at full alpha nine frames after it should read 32; its
fade-out runs ~17 frames late; its declared 80-frame fade-in is never
drawn. Not culling -- the same element is submitted down to a=7 on the
way out. Calibration-free version: the declared fade-out spends 12 of 16
units dropping 23/255 of the alpha, and the capture has no such plateau.
Candidate, offered and NOT adopted: if +36 held the NEXT keyframe's time,
the fade-out shape fits (RMS 4.05 vs 12.13, two elements) and the
decoder's "last block's time is unreadable" special case disappears --
the last block would simply have no successor. Rejected for now because
it explains neither the missing fade-in nor the lateness, and because the
_eff elements cannot discriminate between the readings at all (with four
blocks the shift only relabels the phases). Decoder unchanged.
Also withdrawn, mine, within the iteration: "the _eff glows hold a
constant alpha 33". They ramp 34 -> 255 in steps of 34. I printed the
series minimum and read it as its range, with a "14 distinct colours"
column sitting next to it saying otherwise.
The rotated quads on the title screen come from the keyframe block after
all. The earlier negative -- "every GP_TITLE build 4 element has all three
angle words at zero" -- read the right bytes over too small a region: it
walked the top-level declaration table, and the rotated elements are the
nested leaf records ptloop01.rat / ptloop02.rat.
Confirmed against the framebuffer rather than against our own renderer.
The two records declare +12 = 30 and -45; the GPU capture submits their
quads at +30.26 and -45.28 degrees -- magnitude and sign, two different
values. Corroborated by shape in GP_BUNK 117ca14f, where +12 ramps
0 -> 360 with position, scale and alpha constant: a spin in place.
Identifying which draw it was needed edge lengths, not bounding boxes:
400x1076 and 400x1444 against pteff03/pteff03a 399x180 at the elements'
two different declared scales, 600% (1080) and 800% (1440). The same test
names three known-positives in the capture (ptlogo1, ptcopyright,
ptbtn00), so it passes its own control.
Keyframe gains rotation_deg plus unknown_4/unknown_8, carried rather than
dropped. NOT rendered -- ui_layout::blit is axis-aligned only, so the
reference renderer and the port will both draw these upright until a
rotating blit exists.
The census tool ships with the trap that broke its first version: nested
RATC blobs are not 4-byte aligned, so an aligned scan found 0/3 of its
own control blocks and missed 16 341 blocks. Disc-wide +12 is non-zero in
14.50 % of 83 862 blocks.
sylpheed-formats tests, SYLPHEED_DISC set: 131 passed, 0 failed across the
6 suites finished at commit time; the run had not yet completed.
Finishing the step I left open last iteration rather than starting
something new. Offsets and packet counts were in hand; what was missing
was proof they are actually waves.
slb_extract_wave.py wraps a (bank, offset, packets, channels, rate) slice
in a synthesized XMA1 RIFF, following the layout the Rust decoder already
uses. The three located cues decode to 0.533 s, 0.344 s and 1.016 s of
mono 48 kHz audio, audible from sample 0, each with the percussive
attack-and-decay envelope of a UI blip. Bitrates come out at 12-15 kB/s,
about half the stereo BGM rate, which is what mono should be.
The control matters more than the results. The SAME wrapper applied to
BGM_001's first wave decodes to 173.808875 s -- identical to the duration
that bank's own on-disc RIFF header produced back when Q10 was answered.
So the header I synthesized is not approximately right, it reproduces a
known-good decode exactly, and the cue durations are trustworthy for the
same reason.
I did not commit the decoded audio. Three commands regenerate it from the
disc, and the corpus's job is measurements and tooling rather than
extracted game assets. The offsets, the packet counts and the tool are
the deliverable.
new-game intro has a name
Q9, answered twice over -- once from the disc and once from frames I had
already captured and not used.
The movie manifest's first eight slots are the whole boot-side flow, and
the slot key is the role: LOGO1-4 -> logo1-4.wmv (not on the disc, which
is why the splash is a screen), ADVERTISE_MOVIE -> ADV.wmv, STAFF_ROLL ->
the credits reel, MS00A -> S00A.wmv, MS01A -> S01A.wmv. So the new-game
intro is S00A.wmv, decoded -- which is the half I could not test at
runtime, because A on NEW GAME hangs the emulator.
And there is no separate boot-intro slot: ADV.wmv IS the advertise movie,
the boot just plays it first. Confirmed independently by matching 19
attract frames against five candidates -- 15 hit ADV with a playhead that
advances monotonically at the sampling rate and ends at 137 s, its full
length, with the title back on the next sample. So the attract movie
plays to its end; nothing cuts it short.
That corrects me. Two iterations ago I recorded the attract movie as
"~85 s, so probably not ADV.wmv" -- arithmetic on a start I never
observed, since sampling began 39 s in. REFUTED and METHOD both take it,
along with the matcher's real failure mode: a near-black frame has no
signature, and its runner-up is not evidence.
Skippability I did NOT settle, and the corpus contradicts itself: one
page says A skips a movie every time, while the boot harness deliberately
never taps during one because it breaks the title. Named the one-boot
test rather than picking a side.
not something to concatenate
Q10, and it starts by withdrawing the question's own premise. BGM_001 is
not three sub-waves of 10 KB / 4.47 MB / 4.67 MB: the 10 KB is the bank
header. A bank is exactly TWO waves, and across all 32 BGM banks on the
disc the two always have the SAME duration -- equal to 0.01 s over
lengths from 37 s to 277 s. That alone kills intro+loop and kills two
halves of one piece, both of which require unequal lengths.
Four banks appear to break the rule and do not: BGM_106-109 are the known
leading-region straddle, and the giveaway is that the entry named
BGM_107.slb contains BANK id=1108. The seek packet counts pin each join
exactly, so they realign to the same two-equal-waves shape rather than
being dropped as noise. That trap goes in METHOD.
Then the roles, by decoding both waves to PCM. They are
sample-synchronous: transient-envelope correlation searched over +/-5 s
peaks at lag +0.00 s, and both waves stop at the same millisecond,
167.663 s. Two stems of one performance, meant to sound at once. Wave 1
is quieter, has almost no bass and is far more L/R-decorrelated, which
reads as a surround-rear pair or a second intensity layer -- I cannot
separate those two from the file, and say so: ChannelMask is 0x0002 on
both, and this game's channel metadata is already documented as
meaningless.
Two things the port needs that are NOT on the disc, both marked as
authored: the track is not a seamless loop (BGM_001 fades out and is
followed by 6.15 s of silence, no loop-point field found), and nothing
names which bank the menu plays -- all 32 BGM cues are numeric.
its timing is on the disc
Q7. Every title-side screen carries a full-screen black .prm quad that
paints last, and its keyframe group IS the transition: black at T0, clear
by T1, clear until T2, then back to black on exit. Read with the corpus's
start-of-a-ramp rule and Q1's time unit that gives 0.87s for EXTRAS,
0.97s for the main menu, 4.08s for the title -- from the file, not from a
stopwatch.
The disc-wide check is per-pak all-or-nothing rather than the 41% the
headline count suggests, and GP_TITLE's 6 of 12 is the useful row: the
six builds carrying a fade quad are exactly the six SCREENS, and the six
without are exactly the six overlays. GP_DIALOG is 0 of 133. That is
independent corroboration of the overlay finding from two iterations ago.
One piece is NOT on the disc and says so: the fade-OUT length. The fourth
keyframe has no time slot, because a group's last block stops four bytes
short. Measured instead, at 30fps, ~0.4s and the same both directions.
And a warning I earned: the luminance rise after a transition is NOT the
quad's ramp. The incoming screen's own elements animate in after the quad
has cleared -- 1.47s observed against a declared 0.97s. Time the fade
from where the frame is pure black.
Rig: screenshot samples at 0.5 Hz and cannot see a 0.4s fade at all,
which is why an earlier burst called this an instant cut. ffmpeg x11grab
at 30fps instead; both go in METHOD.
method lines withdrawn
Q5, measured off two boots: up/down move one item and WRAP at both ends
on the 5-item main menu and the 3-item EXTRAS alike; left/right do
nothing; B goes up one level and restores focus to the item you came
from (4/4); B on the main menu returns to the title; B on the title does
nothing. The menu opens on TUTORIAL -- the middle item -- 2/2, though a
third recorded run implies NEW GAME, so that one is reproducible rather
than invariant and says so.
Q4 by driving: LOAD GAME opens the save-slot list, TUTORIAL the lesson
list, OPTIONS the settings menu, EXTRAS build 6, MISSION SELECT the stage
list. NEW GAME is not tested -- A on it hangs the emulator and this
iteration needed the session. The GamePart ID behind each is NOT
measured: it is the entry of the decoded id table whose name matches the
screen I saw, and the page says so rather than wearing the badge.
And the withdrawal. Last iteration I wrote that these menus drop d-pad
presses shorter than ~0.3s. They do not. Once wrap is measured, every
press count I had is exactly right -- four presses moved four steps
THROUGH the bottom, which lands one above where a non-wrapping menu
would. I invented hardware flakiness instead of testing the ends of the
list. METHOD keeps the withdrawal rather than deleting the line.
Also: label brightness is not a cursor oracle here -- the background art
outshines the highlight on some rows. menu_focus.py reads the focus ring
instead, 254 vs <82, no tuning.
The trickiest reading on the disc lived in the Bevy viewer: resolving a
cutscene's voice to a continuous byte REGION of the sound stream, because the
movie voices are one XMA stream chunked into VOICE_*.slb entries whose
boundaries do not match the cues -- a cue routinely spans two chunks, so a .slb
need not hold the track its name claims.
That put the logic most likely to be re-derived incorrectly in the crate least
likely to be reused. The Godot port's exporter needs the same answers, and there
must be one implementation of them.
New `sylpheed_formats::media` owns every case where the bytes of one playable
thing are not one archive entry: segment-spanning reads, multi-sub-wave banks,
and the voice-region resolution. Callers supply bytes through a `DiscSource`
trait, so the viewer keeps its ISO/directory abstraction and a headless consumer
gets `DirectorySource` for free.
The seam is deliberate: this module returns XMA RIFFs, not PCM. Decoding means
shelling out to FFmpeg, which is native-only and a policy decision for the
consumer -- everything up to "here are the bytes that belong together" is disc
knowledge, everything after it is a codec choice.
The four moved functions were previously untested; `tests/media_disc.rs` now
pins them, including the negative the corpus paid for -- an unbound movie must
stay unvoiced rather than borrow a neighbour's clip, which was tried and played
the WRONG recording.
The algorithm is unchanged, moved verbatim (same window sizes, same fallbacks).
The new disc tests pass; the broader audio suite was not re-run in this pass.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
BGM_001.slb is three sub-waves (10 KB, 4.47 MB, 4.67 MB) and the decoder
concatenates them into one 347 s track. That is a default nobody chose, not a
decision: two near-equal halves could be intro + loop, two variations, or two
halves of one piece, and a menu that loops its music needs to know which.
Found while wiring the Audio Library up to the shared banks. Recorded in
HANDOFF.md as a trap too, so the port does not build looping on top of the
concatenated track before the question is answered.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Censusing ptc_pack's own naming vocabulary turned up a third variant of the
prefix trap, and this one had been corrupting a number the corpus carried.
268 of ptc_pack's names do not start with eff_ at all. They start with EF_IDX_,
as in EF_IDX_eff_d001_f. A regex anchored at eff_ chops that prefix off and
merges distinct names, which is exactly where the earlier figure of 532 came
from. Enumerating maximal [A-Za-z0-9_] runs gives 727. The two earlier traps were
a STORED name being longer (rot_n001_break) and a BOUND name being a prefix
(eff_f0002 inside eff_f0002_barnhaze); this is the third - a prefix the pattern
cannot see at all, because its anchor sits in the middle of the real name.
Looking each bound name up bare AND under EF_IDX_ resolves 25 of the 34 that were
unlocated. The map is now 128 of 137, and the residual is 9, small enough to
print: eff_e0044, eff_f0002, eff_f0002_barn, eff_h308, eff_j002_e01,
eff_j002_e02, eff_m010_wep_85, eff_m011_wep_85, eff_n0071. All 17 eff_l### are
among the recovered.
This withdraws my own previous correction. I had recorded Base.xpr (53) as
holding more bound effects than ptc_pack (46), and struck out "ptc_pack is the
effect library". With the prefixed keys counted ptc_pack holds 71 - it IS the
larger library, and the 46 was an undercount from the same truncating pattern.
Two shared libraries remains right; which one is bigger does not.
The suffix vocabulary: 106 distinct tokens over the 727 names - IDX 223 (the
prefix above), _f 137, _e 119, _root 87, _col 54, _mdl 45, _break 43, _ring 38,
_ALL 17, _haze 14, _thunder 10. That census counts ALL tokens rather than
trailing ones, which is precisely how the EF_IDX_ PREFIX surfaced inside what I
had first labelled a suffix list - the mislabel found the bug.
Testing the structural candidates the way _hangar was tested, does the suffixed
name have a bare parent:
_ALL 17 names 17 of 17
_root 87 64 of 87
_break 30 15 of 30
_e 74 0 of 74
_f 61 0 of 61
_root is strictly terminal - 87 of 87, and it never appears mid-name. The
compound shapes put it outermost: _e_root 19, _f_root 18, _break_root 13, bare
_root 30. So the order is <stem>_[<faction>|<break>]_root and _root reads as a
hierarchy marker rather than a variant - though 64 of 87 having a bare parent
means it is not simply the parent of an existing node, and _break at 15 of 30 is
likewise not a plain destroyed-twin-of-everything.
_e/_f never have a bare parent, 0 of 135. That is independent asset-side
confirmation of the faction law: an effect is authored per faction and there is
no faction-neutral original for either side to derive from.
effect-homes.txt changes 5/30 and every line pairs: five values changed (103->128,
34->9, ptc_pack 46->71 and its sort position, the residual header, 3-digit
80->105 of 110) plus 25 pure deletions, exactly the 25 recovered names. All are
3-digit, so the 4-digit line is unchanged at 23 of 27. The other sixteen
artefacts are byte-identical.
Chasing the 17 unlocated eff_l### turned up their shape first: they come in
_e/_f PAIRS - eff_l101_e + eff_l101_f, and the same for l102, l104, l105, l106,
l201, plus _e-only l010/l011/l107/l108 and _f-only l002.
Partitioning every eff_<letter><digits>_<e|f> binding by the ID letter of the
OWNING unit (one GP_MAIN_GAME_* pak = one user):
effect _e effect _f
UN_e### 33 0
UN_f### 0 61
94 of 94 agree and both off-diagonal cells are empty. The control reads the
factions straight off the IDs: UN_e### -> ADAN (42 objects), UN_f### -> TCAF
(26), UN_n### -> TTRL (2, tutorial, binding neither). So an effect ending _e
belongs to an ADAN ship and one ending _f to a TCAF ship - the same visual is
authored twice, once per faction, which is exactly why eff_l### arrives in pairs.
What the 34 unlocated ARE is now also clear, even though where they live is not.
They are one job, not a scatter: Generic binds 32 of the 34, Explosion 19,
Shell 9, Level_0 and Weapon 2 each. The binder fields rank LowerHPFxModel 252,
HitFxModel 144, then JetFxModel_00N and AfterBurnerFxModel_00N. They sit in the
six GP_MAIN_GAME_* paks at 130 bindings each plus 32 in DefTables.pak. Since
LowerHPFxModel is the damaged-ship effect, the residual is largely the
per-faction battle-damage and hit visuals. None of the 34 is a record name and
only one is a field name, so they are asset references.
Stated plainly: they remain unlocated AS ASSETS. Knowing the family and its
naming law does not say where the geometry lives - the .xpr route is exhausted
for them and the parsed pak payloads hold references, not meshes.
Also fixes a defect in the artefact shipped last commit. effect-homes.txt came
back with two equal-count lines swapped: Counter.most_common() breaks ties by
insertion order, so the package listing was not deterministic. Now sorted by
(-count, name) and verified to regenerate byte-identical twice running. This is
the corpus's own rule - any map built by iterating a set or Counter needs
sorted() - and the new tool had violated it.
The other sixteen artefacts are byte-identical; effect-homes.txt changes only in
the tie-break ordering of the five 1-count rows, with every line pairing.
Enumerating eff_* names per .xpr across all 166 packages and matching the bound
names EXACTLY gives a real home for 103 of the 137, up from the 71 ptc_pack
alone accounted for. Only 36 packages carry an effect name at all, and two
dominate:
Base.xpr 53
ptc_pack.xpr 46
Stage_S28.xpr 2
five rou_f001_wep_NN.xpr 1 each
So there are TWO shared effect libraries, not one - and ptc_pack.xpr is the only
*_pack bundle on the disc, so no third shared library is hiding. By digit-width:
3-digit 80 resolved of 110, 4-digit 23 of 27. The previous iteration's split
survives and sharpens - the four-digit series really does live outside ptc_pack
(that zero stands), and now we can say where: Base.xpr.
Correction to the previous commit. It reported eff_f0002 and eff_f0002_barn as
present in Base.xpr. Both were SUBSTRING artefacts: what the file actually holds
is eff_f0002_barnhaze, one longer resource name that grep -l eff_f0002 and
grep -l eff_f0002_barn each match inside. Neither bound name is there. This is
the corpus's own paid-for prefix lesson arriving from the other direction - last
time it was rot_n001 vs rot_n001_break with the stored name longer; here the
BOUND name was the prefix. The new map is exact-keyed and does not have this
failure mode, so the earlier positive is withdrawn.
34 names remain unlocated, dominated by a family the last pass did not single
out: eff_l### with 17 of the 34, then h 4, s 4, j 2, m 2, t 1, and four
four-digit names - eff_e0044, eff_f0002, eff_f0002_barn, eff_n0071.
Scope note worth keeping: the j 22 / t 14 clustering reported last time was the
residual against ptc_pack ALONE; against all packages those families are largely
accounted for and l is what is left. Both numbers are right for their own
population, which is exactly why a residual has to say what it was measured
against.
New artefact with its regenerator: tools/re-capture/effect_homes.py ->
docs/re/data/effect-homes.txt, which lists all 34 by name. All sixteen existing
artefacts byte-identical.
This corrects the previous commit. 文字列 is not a developer's leftover: it is
one member of a six-word Shift-JIS TYPE vocabulary, and the records carrying it
are a machine-readable schema for the unit datasheet.
The whole non-ASCII population on the disc is 6 distinct values out of 99328 -
0 of 3496 record names and 0 of 12173 field names - and all six are type words:
文字列 string 366 uses
NS_"文字列" NS_ string 12
整数 / 整数値 integer 66 / 6
浮動小数値 floating-point value 504
浮動小数値[0〜1] float in [0,1] 36
990 type-valued fields. So the reader defect noted last time is real but bounded
to these six strings, and name_hash re-encodes Latin-1 byte-for-byte, so hashing
was never affected.
They sit in 15 records x 6 GP_MAIN_GAME_* paks = 90 instances, i.e. 15 records
with ONE user. The names are exactly the unit substructure family, and six carry
a literal wildcard: Turret_???, Hatch_???, Bridge_???, Thruster_???,
ShieldGenerator_???, Versatile_???, and NS_*. ??? is the numeric-suffix wildcard
at record AND field level - Turret_??? is the schema for Turret_000..00N, and
inside it CannonFrame_??? / MuzzleFrame_??? stand for the numbered slots.
Where a field's type is an enumeration the schema holds an EXAMPLE value instead
of a type name: Yes for the five booleans, Vessel for Generic.Type (the 43
Craft + 71 Vessel split), Ship_ for the ID prefix convention. Maneuver is the
one fully-typed record, 34 of 34. Every ResistanceTo* and every Color_* channel
is declared FLOAT[0..1] - normalised by declaration, matching the sampled values
in unit-datasheet-static. Generic.NozzleSpec_??? has its own type NS_"文字列"
and NS_* is a record, so the nozzle spec is a nested sub-schema.
Control separates schema from data cleanly: the _??? records and NS_* exist ONLY
as schema, 6 of 6 instances typed, while the eight real substructure names are
typed in 6 instances and untyped in the rest - Generic 6 of 3651, the others
6 of 684 each.
Turret_??? carries the game's own typo NomalModel beside DamagedModel.
This gives the port an authoritative field-type table: types the disc declares,
rather than types inferred from sampled values.
New artefact with its regenerator: tools/re-capture/datasheet_schema.py ->
docs/re/data/datasheet-schema.txt. All fifteen existing artefacts byte-identical.
The remaining named false-positive mode - "107 rows solve to a 64K-boundary
base, a bare addis with no addi of its own, so any scatter of displacements
votes for it" - is refuted by its own measurement.
New positive test in the tool: simulate lis/addis rD,r0,HI + addi rD,rA,N +
or rD,rA,rA forward through each row's function and ask whether the solved
base lands in the solved register.
64K-boundary bases ("low confidence") : 107 / 107 confirmed
non-zero low half ("trustworthy") : 8 / 154 confirmed
A round base is the case where the compiler needed no second instruction, so
`addis r11, r0, 0x820B` stands in the code in full. A miss on the other class
is silence (base built in the caller or loaded from memory), not refutation.
Control: every row the corpus independently validated against the disc has a
64K-boundary base - debriefing, career, save, leaderboard, the 205-name PG*
HUD roster, material slots, the S16 boss collision/frames/motions and its
loader. 13 rows over 10 functions. The dense-short-string false positives the
corpus did name (r31 = 0x8202xxxx) all sit in the "trustworthy" class.
The 0x820B0000 cluster is DUPLICATION, not error: 60 of its 82 rows are one
function emitted 60 times, exactly 491 instructions each, two instructions
differing (both global data pointers), identical 41-address string sequences.
40 resource names written into a per-copy global via sub_8217FA08 at 24-byte
strides. 38 of the 40 are disc GameResourceID values (480 distinct); rot_n001
and rou_e202 are not, and no disc GameResourceID uses the rot_ prefix.
Artefact diff 13/4, confined to the replaced section; the 261-row table and
the 64K histogram untouched; byte-identical on a second run. Fourteen other
artefacts byte-identical.
sub_825F2CF0 / sub_825F2F88 (r0, 30 names each, 97% disc names) are not
name-block readers at all: they are the compiler's VMX save/restore helper
pair. 145 instructions each, 72 differ, every difference stvx<->lvx over
v14..v127.
Root cause: in `addi rD, rA, N` the rA slot reads as literal zero when it
names r0, so `addi r11, r0, -N` is `li r11, -N` - vector spill offsets, not
displacements. The solver excluded r1 only; the bogus base landed inside the
stage-settings name block, dense enough to "resolve" 30 real field names.
name_block_bases.py now skips r0 as well. Rows 277 -> 261, functions 190 ->
176, non-64K bases 170 -> 154, data-table rows 53 -> 50; 154+107=261 and
277-16=261. Artefact diff -100/+3, every removed row line carries r0, control
sub_82341A20 r30 = 0x82088F94 217/226 untouched.
Nothing lost: all 30 names are a strict subset of sub_8230D1F8 r29 (129
names), the stage-settings loader already in the corpus. String-xref join
agrees - FinalPassBG, FogMinDistance, ScreenColorR, ExposureKey_BeginValue
each exist once with exactly one xref, from sub_8230D1F8.
Fourteen other artefacts byte-identical; name-block-bases.txt byte-identical
on a second run.
name_block_bases.py extended with a per-row data-table test; artefact +57/-0,
byte-identical across two runs (now ~2 min 12 s -- it adds a disc-wide pak scan).
The test is objective, not by eye: a row is a data-table schema if its names are
IDXD record/field names on the disc (13450 such names disc-wide). 53 of 277 rows
are >=50 % disc names with >=8 names; the other 224 are engine/XDK vocabulary,
compiled key lists, or noise.
The two axes are independent: against base confidence, solved bases split 34
table / 136 not, round bases 16 / 91. "Round base" and "not a table" are
different questions.
The 53 contain every loader already known -- that is the control. Five rows in
the 53 are unowned, each noun grepped and appearing in no docs/re/ file:
sub_823BDAA8 r11 (33) = the S16 boss's muzzle/attach frames (GN_MainGun_*_Muz*);
sub_823BDAA8 r10 (25) = motion names (Motion_stand, Motion_attackA_start), the
EnumMotions family DefTables declares; sub_82315AE8 r11 (20) = the Guardian
record's own fields, i.e. the S16 boss loader; sub_8219E560 r11 (18) = the
leaderboard screen keys; sub_825F2CF0 + sub_825F2F88 r0 (30 each, same base) =
post-processing (FinalPassBG, FogMin/MaxDistance).
Four rows that look new are not, and their disc-overlap says so -- 53-70 % rather
than ~100 %, because they mix arsenal fields the corpus owns (ConditionToDevelop,
WeaponDesc, SilhouetteModel) with literal screen coordinates as strings.
Not settled: none of the five was opened -- this iteration produced the shortlist,
not the findings.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PMRJjbxLqZtsb5Vb7KunPE
StageResource.EnumerateAIParams names the table, and every name resolves.
29 StageResource records; 28 carry EnumerateAIParams (the one without it is the
_Test template); 23 distinct table names declared; 23/23 hash to an AIParams
object key under the prefix "stage\"; 0 objects left unnamed; 0 declared names
with no object.
AIParams_S01..S16 (16) + AIParams_S24..S29 (6) + AIParams_Tutorial.tbl shared by
six (UnitGroup_S18..S23, the tutorials) = 23 tables over 16 + 6 + 6 = 28 records.
The arithmetic closes both ways and matches the 23-object count found last
iteration.
Same sharing shape as the settings family (stage-settings-table.md: 24 objects,
StageParameter_Tutorial shared by six tutorials). Two independent families agree
on how the tutorials are handled -- n=2, a pattern rather than a rule.
aiparams_census.py extended with the stage mapping; artefact +7/-0, byte-identical
across two runs; the other fourteen verify unchanged.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PMRJjbxLqZtsb5Vb7KunPE
CORRECTION FIRST. Last iteration I recorded sub_8233C368 as unblocking a
NEEDS-HUMAN item -- "the corpus carries the AI tail of Maneuver as
NEEDS-HUMAN/runtime; it is statically reachable after all." That was wrong.
stage-mission-tables.md already documents AIParams_S02.tbl as exact original
values obtained by static RE, directly portable, listing all 20 field names and
both shapes. I grepped FiringLength and saw the file but did not read the
section. Finding the owning doc is not reading it. The only genuinely new part
was the loader's name.
What is new: the census generalises Stage 02 to the disc. 23 AIParams objects,
identical in all six GP_MAIN_GAME_* paks, sharing ONE declared-name set of 34
profiles; 782 profile records = 23 x 34; 0 declared names without a record in
their own object. So "34 AI profiles" is not a Stage-02 fact -- every stage
carries the same 34 and only the values move. The roster is declared by an
Enumerate_AIs record whose field names are the profile names, the same
declaration-table mechanism that closed DefTables.
Type predicts the field count with exactly two exceptions: Fleet -> 6 fields is
253/253 zero partials; Squad -> 20 fields is 483/529. The 46-record residual in
full: AI_Test and AI_CraftSquadron_Test, both Type = Squad with only the six base
fields, in all 23 objects. No profile's shape varies between objects.
New regenerator aiparams_census.py, 45-line artefact, byte-identical across two
runs; the other fourteen verify unchanged.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PMRJjbxLqZtsb5Vb7KunPE
277 rows over 190 distinct functions (a function can read several blocks).
name_block_bases.py extended with a confidence split.
The tool's false-positive mode, measured and named: 107 of 277 rows solve to a
base on a 64K boundary -- a bare "addis rX, r0, 0xHHHH" with no addi, so any
scatter of displacements votes for it. 82 are 0x820B0000: about 60
near-identical functions in 0x8281xxxx-0x8284xxxx all "naming" the same rou_e0NN
list. The 170 rows with a non-zero low half are the trustworthy set. A round
base is not automatically wrong -- sub_822215D0 sits on 0x820A0000 and resolves
205/206 -- so read the ratio, not the base.
The index re-derives every loader we already knew (unit 217, stage settings 129,
PlayerParams 90, hangar 81, squadron orders, missile guidance, shell movement,
substructures, six camera/fog readers) -- that is the control.
The find: sub_8233C368 reads the AI behaviour table. r28, base 0x8208583C, 20
names -- Enumerate_AIs, FiringLength, GuardLength, AutoGuardLength, CounterLength,
MusterLength. stage-mission-tables.md owns those field names on the data side,
but Enumerate_AIs appears in no document and no reader was known; the corpus
carries the AI tail of Maneuver as NEEDS-HUMAN/runtime. It is statically
reachable after all. The same base also serves sub_82338EE0 (97 names, Weapon
TargetType SpecialWeaponType ReticleType IsCharging ...) -- the weapon datasheet
loader, also not previously named.
Five unowned blocks surfaced and NOT opened: PGHUD_*/PGREMAIN_NUM HUD part names
(205/206), STAGE_RESULT/stage_num_shoot_down_aircrafts/EX_OVERVIEW,
g_mWorldViewProjection/NormalMap/GlossinessMap engine material slots,
Boss16Collision* (cross-links the S16 Guardian object), and roh_n001_menu1_cam_pos
menu camera tags.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PMRJjbxLqZtsb5Vb7KunPE
New reusable tool, tools/re-capture/name_block_bases.py -> docs/re/data/name-block-bases.txt
(2880 lines, ~65 s, byte-identical across two runs).
A loader that reads a table by field name keeps one base pointer and emits
"addi rX, rBASE, -N" per name, so no static xref sees the strings. Solve the base
from the DISPLACEMENT SET alone: every (string address, displacement) pair implies
a candidate base, and the true base collects a vote from every name it explains,
so it wins outright. My first cut took candidates from ONE displacement and
scored the unit loader at 52/226 against the right answer's 217/226 -- vote over
the whole set, not a probe.
Control passes with no prior knowledge: the tool recovers sub_82341A20 -> r30 =
0x82088F94 at 217/226, and independently recovers sub_8230D1F8 (129/132),
sub_822F9498 (90/91) and sub_822AE628 (81/108). 277 name-block-reading functions
image-wide, with the schema each names.
The analog block is SOLVED: sub_821A6CF0, r29 = 0x820A1630, 22/24. In code order
it names ControlTweakName, YawMagForNormal, the 12 Tweak fields, the 8
AnalogRevice_* curves and GP_MAIN_GAME -- the whole schema in the object's own
order plus its pak. r29 is built at 0x821A6D34 as addi r29, r11, 5680 =
0x820A0000 + 5680, matching the solved base exactly. It is the same function that
reads PlayerParams.
Two of my own verdicts withdrawn: "referenced by nothing" and the softened "not
found by these routes". The measurements behind them were right; the conclusions
were wrong. The base was solvable from the data the whole time.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PMRJjbxLqZtsb5Vb7KunPE
All six language copies carry the identical 337 unnamed IDXD hashes. Partitioned
by record-name shape (53 shapes), 333 map onto families the corpus already
documents: 131 weapon datasheets, 114 unit datasheets, 64 unit Faces tables, 10
unit message sets, 8 chatter rule tables, 5 enumerations/formations/placement.
114 is exactly the corpus's unit count (43 Craft + 71 Vessel) -- these are the
tables the corpus has always worked with, reached by SHAPE because they have no
names. Naming them adds nothing.
The Enumerate object in each GP_MAIN_GAME_* is EMPTY, zero fields, which is why
route 2 named 1283 entries in DefTables and 0 here.
The 4 unclassified objects are new -> docs/re/structures/player-tuning-tables.md:
the analog stick response curves (8 axes, 11 samples + a named Count = 11, tested
8/8; yaw/roll/throttle are the identity ramp, the shaping is all on pitch and the
camera axes; adv_yaw is non-monotone and unexplained), the player craft's flight
envelope (Booster, 50 fields -- the player side of the AA_/AV_ pair documented
for NPCs -- plus TacticalManeuver, SpecialAttack + three gauge bands,
SpecialWeapon, Misc), the Stage 16 boss (identified by Shell_S16Boss_* ids;
Guardian HP 65000, Core 42000), and one unidentified Generic naming eff_n0071.
New regenerator main_game_unnamed.py, 112-line artefact, byte-identical across
two runs; the other twelve artefacts verify unchanged.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PMRJjbxLqZtsb5Vb7KunPE
The string harvest could never name DefTables because those names are not
spelled out as "something.tbl" anywhere on the disc -- they are declared.
An IDXD object whose single record is named Enumerate is a DECLARATION TABLE:
its field NAMES are the names of other objects, each resolving as
name_hash("<field name>.tbl"). EnumLODSet_test.tbl declares 676,
EnumGameModel_test.tbl 360; the disc holds 144 such objects (138 in DefTables,
one in each GP_MAIN_GAME_*) declaring 1298 distinct names.
Route 2 names +1283 entries route 1 could not. 130 + 1283 + 12 = 1425, no
overlap, 99.2 % coverage. Zero partials: of 5 suffixes x 6 prefixes, ('', .tbl)
scored 1036/1036 and every other combination scored 0. Residual in full: 8
declaration tables nothing declares, 2 LOD sets (Model rou_e004 / rou_e013), 2
motion sets; 15 declared names have no pak entry at all.
REFUTED alongside it: the 40 XPR2 manifests are not the naming source -- their
82 Name= values and 82 DataFile/Source paths resolve 0 entries under any of the
5 suffixes. They share the MODEL namespace only: 40 of the 82 appear as the
Model field value inside the tables.
Artefact +17/-18, every line paired, byte-identical across two runs.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PMRJjbxLqZtsb5Vb7KunPE
