Two gaps the first post-migration decoder iteration exposed, both in the same
place — the loop ends before it can observe its own result.
1. Step 1 read notifications but never looked at the agent's own open PRs. It
opened #23, labelled the issue `state/needs-human`, and stopped; nothing in
the brief would ever bring it back to a red check. Next iteration it planned
to start a new question on top.
2. It called the merge resolved on `cargo build -p sylpheed-cli` — the crate the
conflict was in. The break was in `sylpheed-viewer`, which neither side of
the merge edited: one branch added a `ComposeOptions` field, the other had a
literal construction of it. The crate you just fixed is the one place the
breakage cannot be.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
`decoder-loop.md` has had `## Each iteration` since 2026-09-04, ending in
7. Commit, `push-work`, open the PR, label the issue `state/needs-human`,
and stop.
`port-loop.md` has never had that section at all. It states the rules -- "you
cannot close your own work", "you move an item to state/needs-human", work
reaches main "through pull requests" -- but nowhere gives the STEPS that carry
them out. The Port agent was told the rule and never the procedure.
Two steps were missing entirely, and both are Phase 7 check items:
* READ YOUR NOTIFICATIONS first. They are polled and nothing pushes them, so
an agent that skips this never receives anything addressed to it -- and the
symptom is silence, not an error. That is GITEA-SETUP.md Phase 7 check 1.
* OPEN THE PR AND SET THE LABEL. That is Phase 7 check 2.
The second carries a trap worth naming in the brief rather than in a runbook:
`docker/port/bin/push-work` ONLY PUSHES. Verified by reading it -- no /pulls
call, no `Closes #`, no label; its refusals (no main, no force-push) are
intact. `propose-work`, which would do all three, is still unbuilt --
GITEA-SETUP.md's own words are "the other two thirds being manual is how they
get skipped". The capability is there (`pull_request_write` and `issue_write`
are both in the pinned MCP tool set), so this is an instruction gap, not a
permissions one.
Written to mirror the Decoder's shape while using the Port's own vocabulary --
a gate is an artifact, verify against captures rather than our renderer, state
the expected number first, `authored/` entries carry a `why`, never an invented
number where a measured one is pending.
Found by the Pi agent while answering a Phase 7 readiness question. My own
check had been decoder-only, which is why the asymmetry survived it.
Refs #11
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
§§1-9 were written on fabi-Hyrican-PC. On the other desktop the ~/.sylph-*
credentials do not exist, the Pi does not resolve, and stable is 1.90.0 rather
than the 1.98.1 §8 records -- so §8.2 (fetch the WASM bundle) and §8.4 (Phase 7)
cannot be run from here at all. Says which of the four steps can.
Measured rather than carried over: protection holds (10/10), fmt is 774 hunks
across 154 files, check-citations is 19, and the tests are 207/0/14 across 30
suites. Two of those need reading carefully:
* clippy DIVERGES. The runner is rustc 1.98.1 -- read out of job 794's log,
not assumed -- and is clean; here 1.90.0 exits 101 on only_used_in_recursion
at vfs.rs:85. That is #15 ceasing to be theoretical. It is NOT evidence that
CI's green is fake, which is the §7 lesson-5 inference in the other
direction.
* the test tally matches to the unit while measuring something else. 15
*_disc.rs files resolve disc_root() through a hardcoded absolute path, so
unsetting SYLPHEED_DISC does not skip them: the disc suites RAN here (1936 s,
mesh_consistency_disc alone 1220 s) and skipped on CI (2.4 s total) -- and
both report 207/0/14, because the skip path returns from a test that still
passes. Good news for #14, since this run is the stronger evidence; and worth
an issue, since SYLPHEED_DISC looks like a control and is not one.
I got that last one wrong first -- inferred "the counts cannot match" from "the
fallback resolves", which is §7's shape a sixth time, recorded as such.
Also: a plain `git clone` of this repo fails three ways on the pack that still
carries the 545 MB; --filter=blob:none works. And §6's tokio claim was
challenged and survived -- every use is inside a #[cfg(test)] module.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Written to be read cold by a session with no context: what is established
versus what is someone's claim, which of the four actors the gate actually
binds, and what is in flight but unpushed.
The transferable part is §7 -- one failure shape recurred five times in two
days across two agents and the assistant: a property inferred from something
ADJACENT to it rather than tested directly. Protection from a settings page,
reachability from a DNS record, a label count from a grep that matched the
function definition, Gitea's host from a DNS lookup, and a clippy version from
rustfmt output parity. The last is the sharpest: rustfmt 1.8.0 and 1.9.0, nine
months apart, both emit 774 hunks here, so formatting parity carries no
information about which clippy ran -- and the conclusion drawn from it (that
CI's green was cached or ungated) was wrong in a way that would have sent
someone chasing a phantom.
The two instances already in "Checks that were kind once" were correct
when written and decayed. The third was wrong on its first commit, and it
arrived by a different route: the check and the tree's failure to pass it
land in the same change, so the softening writes itself.
Concretely — the Clippy step had never run (no component in the
toolchain), and the tree is not clippy-clean, so fixing the step and
turning it red are the same commit. The first draft paired the fix with
`continue-on-error: true` and a comment promising removal once the debt
was paid: an expiry date nobody set, in the shape #12's closing line had
already ruled out for rustfmt. Reverted on reading it.
Adds the distinction, a table separating decay from dirty authorship, and
an earlier tell than the mechanical test:
If you are writing the softening in the same commit as the check,
the thing you want is an issue, not a flag.
The mechanical test is unchanged and still correct; this only catches the
same failure sooner, at the keyboard rather than at review.
Refs #12, #13
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01McNbzUeq1KRBWs4G6X2YVj
Two rules that look unrelated and are one failure, plus the change that makes
the second enforceable.
1. A FINDING REACHES `main` BEFORE THE CODE THAT CITES IT. A citation resolving
only on a peer branch is dead the moment it merges. Not hypothetical: 495
decoder and 366 port commits sit off `main`, and `port/scripts/boot.gd`
already cites two docs/re pages present on neither its own branch nor main.
2. A CHECK MAY ONLY SOFTEN AGAINST A CONDITION IT CAN TEST -- the Pi agent's
wording, and better than mine, because it is applicable while writing rather
than a call to be vigilant. The mechanical form:
Can this branch tell the difference between "not yet" and "no longer"?
`gitea-protect --verify` printed ⚪ "not a collaborator (yet)" and continued,
so the only instrument checking Write-not-Admin could not report that gate
being REMOVED. `check-citations` reported peer citations instead of failing
them, because under the old topology that was unfixable from the container.
Both were correct AND kind when written; neither recorded that the kindness
had a scope. Nobody edits these into being wrong -- the world moves and the
allowance stays, which is why they survive review. The smell is leniency with
an expiry date nobody set; the fix is the testable-condition rule.
check-citations gains `--for-merge`, which turns the peer class into a failure.
A flag rather than a new default because BOTH readings are still live: mid-work
on a topic branch the peer class really is unfixable noise. What the old code
could not express is where the code is GOING, and that is a condition the caller
can state. Measured on this tree: 19 citations resolve only on a peer branch --
which is the size of the #7-depends-on-#8 edge, not the 2 I had counted in
boot.gd.
The selftest gains that third class, because a flag whose classification is
unexercised is the shape this rule exists to catch. Controlled: emptying
PEER_REFS makes the peer case collapse into "nowhere" and the selftest reports
🔴 BROKEN, rc=2.
⚠️ Pre-existing and NOT from this change: the default run already exits 1 on 4
citations of `export/...` paths. Those are the generated tree, gitignored by
design, and main's copy of the tool fails identically. The CITE regex treats
`export/` as a repo prefix. Reported, not fixed -- it is the port's file and its
call whether the regex or the citations are wrong.
Phases 1-4 and 6 are done on the instance. This file still opened with "Nothing
exists on the instance: no agent users, no API tokens, no labels, no milestones,
no branch protection" -- every clause of which was false by the time the merge
that carried it landed.
Replaced with a table of measured state, and each row says what was MEASURED
rather than what was run:
* protection is verified behaviourally -- a real push to main refused with
`pre-receive hook declined`, as the repository owner -- not read off a
settings page. That distinction is the whole subject of this file.
* the tokens are probed: right identity, 403 on branch_protections for both
agents, so the Write-not-Admin carve-out is demonstrated and not asserted.
* the labels are 11 because the instance holds 11.
And a standing note that this block is the part most likely to be wrong, with
what to believe instead: `gitea-protect --verify` and the issue list MEASURE,
this block REMEMBERS. A remembered status is a cache with no invalidation, which
is the same failure as a 1,227-line BLOCKED.md and as the two documents this
runbook was split across an hour ago.
Caught by the Pi agent against the live instance after Phase 4 ran. The tool
creates 5 state/*, 2 agent/*, 4 kind/* = 11.
Where the 12 came from is worth a line, because it is a shape that recurs:
$ grep -c '^mklabel' tools/gitea-setup
12
$ grep -n '^mklabel' tools/gitea-setup | grep -v ':mklabel "'
74:mklabel() { # name colour description
I counted the function DEFINITION as a call. A measurement taken one token away
from the thing being measured -- the same shape as reading protection off a
settings page and reachability off a DNS record, which is now three today. The
version that cannot make this mistake is counting what the instance holds, and
that is what found it.
Two things that read as protection while being none.
Phase 2's rule binds everyone who reaches Gitea through the API or the web, and
does not bind anyone with `gitea admin` in the container -- which includes the
supervising agent that created the agent accounts and minted their tokens. From
that shell the rule is editable and an admin token is one command away. That is
the boundary of what the phase buys, not a hole to plug there, and the document
read as though the gate were universal. Phases 1 and 2 gate the two CONTAINERISED
agents, whose design assumption is that policy lives where they cannot reach it;
a supervisor with a host shell is not in that set.
And `gitea-setup` finished by telling the reader to go and build a Gitea project
board by hand, four sections after the doc explains that a board is a second copy
of the state to hand-sync and is precisely the failure that produced a 1,227-line
BLOCKED.md. A tool instructing you to do the thing its own documentation argues
against is the drift this whole surface exists to end.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01McNbzUeq1KRBWs4G6X2YVj
The runbook existed as two documents -- a published page and this file -- with
no mechanism keeping them equal, only an intention to remember. Two versions was
the predicted outcome of that, not an accident on top of it. This is the fold,
and the rule that follows it: THIS FILE IS THE SOURCE, the page is derived from
it. When something is urgent enough to push to the page first, it lands here in
the same turn, not "shortly after".
Four things the file did not carry:
* YOUR OWN PUSHES TO main STOP. `enable_push: false` compiles to CanUserPush,
which returns false with no bypass for admins or the owner -- quoted from
the source. Three commits went in by direct push the day this was written,
so the first notice would have been mid-task. Now a check step.
* the token files' MACHINES, which the table had lost.
* do NOT add `write:repository` to the `fabi` token. That scope IS a push
credential. Written down because that advice was given, in chat, by the
author of this file.
* Gitea 1.25.5 confirmed from the desktop too, not just the Pi.
And one thing deliberately NOT folded in: the page said the desktop's outbound
HTTP was blocked, and that is false. `python3 -c 'urllib...'` returns
200 {"version":"1.25.5"} from this box. What is refused here is `curl`, by a
local permission prompt -- which I read as a network constraint and then
published as one. The Phase 3 locations stand; the reason given for them did not.
The "not verified" section now separates WRONG from UNCHECKED. Four entries are
wrong -- requiring an approval does not close the gate, the check could not have
caught that, the token scope, the reachability -- and the pattern in all four is
identical: a property inferred from something ADJACENT to it (protection from a
settings page, reachability from a DNS record) instead of tested directly. That
is the frozen-splash failure, committed in the document about avoiding it. The
first two were caught by the other agent, which is the argument for the review
gate this file exists to build.
Phase 2 as a file. Six settings where two are load-bearing and both were missing
from the first draft is the shape of thing that gets mis-clicked at 1am, so it
goes through the API: what was applied is readable in a diff, and `--verify`
can re-check it later instead of it being checked once.
--verify states its expectations INDEPENDENTLY of what the apply path sends.
A check derived from "whatever we posted" cannot fail -- it re-derives the
expectation from the thing under test, which is the same instrument-shaped
failure as a check that passes on an instance with no rule at all.
It also asserts both agents are still Write and not Admin, because an agent
promoted to Admin can edit the rule and then merge, so a green rule proves
nothing on its own. That is the `gitea-verify` card from "Still to build";
what is left of it is only putting it on a timer.
`block_admin_merge_override` stays false on purpose, and the reasoning is in
the file: approvals are whitelisted to `fabi`, and Gitea will not let `fabi`
approve a `fabi` PR -- so with the override blocked, a human-authored PR could
never reach one approval and could never merge at all. The override is not a
hole in the agent gate because the agents are Write, not Admin. Phase 1.2 pays
for that; this is where it is spent.
Reads the repository-scoped credential that already exists on the agent box
(~/.sylph-git-credentials) rather than the issue-only ~/.sylph-gitea-api-token,
which every branch-protection endpoint refuses. That keeps the setup needing no
new credential, and keeps push rights on one machine.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01McNbzUeq1KRBWs4G6X2YVj
Phase 5 of docs/agents/GITEA-SETUP.md, plus a correction to Phase 2 that the
runbook could not have known it needed.
gitea-mcp v1.7.0 goes into both images, pinned by the sha256 the release
publishes and smoke-tested with `--version` at build time, so a bad pin fails
the build instead of the agent. Each entrypoint registers it at user scope for
that container's own identity, remove-then-add so a restart is idempotent.
The token is passed BY PATH. `-e GITEA_ACCESS_TOKEN=$(cat …)` would write it in
cleartext into ~/.claude.json, which every session in the container reads;
GITEA_ACCESS_TOKEN_FILE is new in the pinned version and leaves the secret in
its read-only mount. Verified against the binary's own --help, not assumed.
The tool filter stops being an experiment. The names are in the release README:
each agent gets issues, notifications, labels, milestones and pull requests, and
NOT `pull_request_review_write`. That one matters because separate identities
open a hole the runbook did not name: Gitea refuses to let an author approve
their own pull request, and does nothing about sylph-decoder approving
sylph-port's. Two agents could satisfy `required_approvals = 1` between
themselves and then merge, since branch protection blocks pushes to main and
never blocked merges.
Withholding the tool is defence in depth; the controls are in branch protection,
and both docs now say so: approvals whitelisted to the human so an agent's
approval does not count, merges whitelisted to the human so an approved PR is
still merged by a person. Phase 2's check gains the step that actually tests it
-- approve the throwaway PR yourself, then confirm the agent STILL has no merge
button. Without that step, the check passes on an instance where the agents can
merge each other's work.
Also settles two entries on the runbook's own "not verified" list: the tool
filter names, and the Gitea version (1.25.5, whose API schema carries
enable_merge_whitelist and enable_approvals_whitelist under those names).
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01McNbzUeq1KRBWs4G6X2YVj
WORKFLOW-gitea.md said what the working surface is and why. It did not say how,
in what order, or how to know a step worked -- so it was a destination with no
route. This is the route.
Seven phases, each with a check, each marked 👤 human or 🤖 me:
1 identities two agent users, Write NOT Admin
2 protection main behind a PR + 1 approval -- BEFORE tokens exist
3 tokens three principals, three tokens, three files
4 structure labels and bundles, and deliberately NO Kanban board
5 MCP gitea-mcp v1.7.0, per-agent identity, user scope not .mcp.json
6 items migrate the live findings only -- not 1,227 historical lines
7 restart and verify the three things that must be true
Phase 1 leads because it is not hygiene: Gitea does not let a PR's author
approve it, so while an agent IS `fabi` either the human cannot approve its work
or it can approve its own. The review gate does not exist until the agents are
distinct people. (It also fixes 495 commits of agent work attributed to the
human's email.)
Phase 2's check is a real push and a real PR, not a reading of the settings
page. The reason protection lives in the server rather than in a brief is that
it should not depend on good behaviour -- so verifying it should not either.
Phase 5's install facts are checked, not remembered: gitea-mcp v1.7.0,
`gitea-mcp_Linux_x86_64.tar.gz`, `-t stdio -H <host>`, `GITEA_ACCESS_TOKEN`.
The `--tools` filter is flagged as an EXPERIMENT that might exclude the merge
tool as defence in depth -- explicitly not a substitute for phase 2.
Ends with what is still to build (propose-work, an attachment uploader,
gitea-verify, the wiki landing page) and a "what I have not verified" section:
the approve-your-own-PR behaviour, the --tools names, the Projects API, and the
Gitea version -- the API was unreachable from my sandbox three times running.
The two loop files ARE the prompts -- `sylph-port` and `sylph-decoder` read them
off the host at launch -- so the workflow change had to land here or it would
not reach the agents at all.
PROTOCOL.md gains four sections:
* Work items -- issues, milestones as bundles, the state labels, and that
`state/blocked` uses DEPENDENCY EDGES, never prose. A prose blocker is what
let a 1,227-line BLOCKED.md go stale.
* Messages -- an ask is a `kind/ask` issue, not a SendMessage. With the part
that matters: 🔴 NOTHING PUSHES. Notifications are polled, at the top of
every iteration, and therefore an agent must NEVER wait on an ask -- set the
edge, take the next item. The channel this replaces dropped 21 consecutive
messages to a stale session id and reported success each time.
* Pull requests -- one item per branch, `Closes #N`, and you may not merge
your own. Branch protection enforces it; the rule is written down so the
agent knows it, not so it depends on the agent.
* Each iteration, in order -- notifications, sync, one unit, hand over, stop.
Also: evidence a human must look at now attaches to its issue, and a blunt
"never commit game content, under any directory name" with the 545 MB that
prompted it.
The two briefs shrink 697 -> 298 lines. They had accreted five dated focus
blocks between them -- sole-focus orders, F1-F6 queues, one-off "merge this
branch on your first iteration" instructions -- which is a queue, and a queue
belongs in the tracker. What is KEPT is what outlives its bug:
* ask of any check, what would this still report if the feature were absent?
Three instruments passed a splash that never animated.
* the instrument must sit at or above the thing that can break -- the
InputEventAction / input-map miss.
* R1, and grep REFUTED.md before proposing.
* the .pe is primary and the database is somebody's analysis of it.
* the oracle is the real game in Canary, not any renderer of ours.
⚠️ NOT YET TRUE when this lands: the agents have no Gitea users, no API tokens
and no MCP server, so the issue tooling these briefs assume does not exist yet.
The agents are stopped. Setting that up is the prerequisite for restarting them.
Takes the port branch up to 77320d5e -- the state the human play-tested on
2026-09-02 -- for SOURCE paths only. Not a branch merge: `auto/port-p6-audio`
is 366 commits and 938 files, and most of that must not land.
WHAT COMES IN (76 files, all human-confirmed working):
* the logo splash animation. 08ed3dd1 found it: `pose_at` ASSIGNED the settle
instant instead of clamping to it, so the splash never animated at all --
and the same bug manufactured a passing harness result, because the harness
photographed t past the settle. Confirmed by play-test: "cannot notice any
obvious difference from the actual game."
* gamepad input -- (A)/(B) bound additively (`ui_accept` ships with NO joypad
binding), stick latched with hysteresis at the game's own 61% digitise
threshold. This is what made (A), video-skip and Extras work at all.
* menu navigation and flow, menu audio, the exporter, the authored
declarations, and 23 verification tools under tools/port/.
WHAT IS DELIBERATELY LEFT ON THE BRANCH:
* everything after c0ae460a -- the F5/F6 title-timing investigation, whose own
tip commit calls itself a "hand-off for one-minute human checks". Unchecked
by definition; it goes through the new review gate like anything else.
* the OPTIONS menu work of 2026-09-03. Real, probably good, NOT play-tested.
* the F1 repeat mechanism, which its own commit calls "deliberately inert".
WHAT MUST NOT LAND, AND WHY THE .gitignore CHANGED:
545 MB of extracted game content was committed on that branch -- 850 sprite,
audio and transcoded video files under `export-probe/` and `export-probe2/`,
plus 246 MB of loose .wav and .tsv at the repo root. This repository's own
rule, in this file, is "never game content".
The rule was not missing. It was written, and it was tightened on that very
branch, with a careful comment explaining why BOTH `export/` and `data/base/`
had to be listed -- while the exporter was writing to a third name that
nobody had thought to list. Enumerating names is the thing that failed. So
the ignore rules now describe the SHAPE: any top-level `export*/`, game media
by extension, and loose capture output at the root. Verified both ways -- it
catches all four offenders and ignores nothing currently tracked.
Verified: `cargo check --workspace` clean; all nine GDScript files parse in
project context, with a positive control (an injected syntax error is detected,
3 lines) so the clean result means something. `tools/port/check-all` was NOT
run -- it needs the container, the export tree and a display.
The human wants to direct this project from a web UI rather than chat or Remote
Control, so Gitea becomes the working surface. No new store: adding a second
copy of the truth is this project's defining failure mode, and Gitea already
holds the code. Its first-party MCP server (gitea/gitea-mcp v1.7.0, checked) has
issues, labels, milestones, PRs, attachments and notifications.
ISSUES replace BLOCKED.md. Milestones are bundles the human defines; issues are
items agents propose and the human approves. The state labels end in
`needs-human`, which is the state the whole model turns on and the one no
off-the-shelf tool models -- the market has converged on removing the human.
`blocked` uses Gitea's DEPENDENCY EDGES rather than prose, so "the Port is
blocked on the Decoder answering X" becomes queryable and closes itself.
PULL REQUESTS, the human's proposal, adopted -- and a bigger improvement than it
looks. Today's long-lived auto/* branches have drifted 280 and 373 commits apart,
which is unreviewable by construction. One PR per item makes the human gate
NATIVE rather than a label convention, binds the change to its item, and enforces
the sizing rule: an item too big to review in one sitting was too big to be an
item.
🔴 Agents must not merge their own PRs, and pull_request_write includes merge --
so this goes in BRANCH PROTECTION on main, not in a document asking them not to.
Same principle that fixed the build-jobs cap: policy where the agent cannot reach
it.
WIKI -- the human suggested it for RE findings, and that half is declined with
reasons. A finding's value is that it sits beside its evidence, versioned with
the code that consumes it; the wiki is a separate git repo, so a decode
correction and the exporter change depending on it could never be one reviewable
PR. And wiki edits bypass review: the REFUTED.md reclassification changed the
file both agents read to decide what not to try, and as a wiki edit it would have
been an unreviewed mutation of shared ground truth. The wiki takes human-facing
orientation instead -- runbook, navigation, container notes, and a landing page,
which closes the real gap that there is no view of what is happening except
container logs.
FILES: three needs, three homes. Agent-to-agent transient stays in /exchange.
Evidence a HUMAN must look at attaches to the issue it belongs to -- it travels
with the item and cannot be orphaned from the claim. Evidence a finding cites
stays in git. Note the MCP exposes attachment_read only; upload needs a direct
REST call.
tools/gitea-setup creates the labels and bundles, idempotently, with --dry-run.
Blocked on a token with write:issue -- the push credential is write:repository
and every issue endpoint refuses it, checked rather than assumed.
Two new findings from the human, both about WHEN a title animation starts, and
both handed over rather than guessed:
F5 Does (A) SNAP the title to finished, or ACCELERATE it? The human says they
cannot tell and is right that they cannot -- a three-frame acceleration and a
one-frame cut look identical to an eye. Two routes that should agree: a
per-frame capture (acceleration shows intermediate alphas, a cut shows none)
and the code (assigning a target time and raising a rate multiplier are
different instructions). Their "looks more like a snap on multiple attempts"
is recorded as a PRIOR, not a result.
F6 The title's sweeping white glow -- ptloop01/ptloop02, the blue PCB-like lines
-- starts only when the plate appears in the real game, and starts earlier in
the port. A lead from the exported declaration, mine and unverified: those
elements are keyed at t = 0, 70, 100, 238, 250 while the plate reaches full
alpha at 236, with pteff02 keyed at exactly 236 and ptlogo_back2eff and
ptcopyright at 238. 236-238 is a synchronisation point in the declared data
and a human just reported a behaviour change there. Flagged AGAINST itself
too: 238...250 looks equally like an exit ramp -- ptcopyright uses that shape
and starts nothing -- and the sweep lives in a nested .rat leaf with its own
timeline.
F6 bears on clock: "shared" and on F4: if a title element does not move until
the plate arrives, either the declared data says so and our keyframe reading is
wrong, or something at the plate's arrival STARTS it, which is a mechanism
nobody has proposed.
And the process change, which is the human's and outlives this item:
"attacking the 'whole' mission was too big for them to handle. Split the given
missions and tasks into even smaller tasks which they can tackle and give to
a human for feedback."
PROTOCOL.md gains "Work in units a human can check in a minute". A milestone is
not a unit of work, it is a bag of them. A unit is right-sized when it ends in
something a person can judge in under a minute WITHOUT READING ANYTHING, and
each one states its question, what the human looks at, and what it does NOT
cover. Do one, hand it over, stop -- an unverified fix under a second change
makes a regression two-variable.
The evidence for the rule is this week: the splash sat through a whole milestone
and took one day once scoped to "does it animate?". The bar is a HUMAN check,
not a green tool -- three instruments passed a frozen screen.
I wrote "whether the game snaps both clocks forward" into yesterday's F4 and the
human asked which clocks. There are none: authored/flow.json sets
`clock: "shared"`, so the title's two composited builds -- build 4 the artwork
(finishes t~=118) and build 2/3 the plate (full alpha t=236) -- run on ONE clock
started together. Left standing, that phrasing sends an agent hunting for a
second clock this corpus says does not exist.
Corrected in both briefs and in the playtest page, marked as a correction rather
than silently edited.
And the question is better than I first framed it. `clock: "shared"` is
AUTHORED, and the port's own plate-arrival-halves.md calls it "not falsified...
not confirmed to better than ~20 % either", with an unresolved anchor
disagreement inside one binary: the reconciliation picked t=118 while
settle_time() returns 160 and the boot prints "settles at t=160".
So F4 is a TEST OF THAT PREMISE, and the discriminator is observable -- press (A)
early, while the wordmark is still building in, and watch the ARTWORK rather
than the plate:
advances the shared clock -> the artwork SNAPS to finished
only forces the plate -> the artwork KEEPS ANIMATING its build-in
Both briefs now say to answer F4 before building on `shared`, and tell the port
not to choose what "jump" means.
"Menu walk and navigation is fine. Video skips too. Extras open. New Game
shows new game intro video." -- 2026-09-02
P5 is done. Its gate was "a human clicks through it", the retro said it had been
waiting on that and not on code for the whole milestone, and it has happened.
PORT-MISSION.md updated. The NEW GAME gap is accepted as-is.
Four findings, three of them the Decoder's:
F1 THE MENU REPEATS ON A HELD DIRECTION AND OURS DOES NOT. One step per
deflection was authored as the conservative choice because nobody knew; a
human has now watched the real game and it repeats, "at a medium pace... slow
enough to see which item is selected". That settles the existence half of H1
against us. The RATE is still unmeasured and must not be guessed -- the
description bounds it and supplies no number. Decoder measures initial delay
and repeat interval as frame counts; the port implements the mechanism and
waits for the numbers.
F2 THE SFX ARE TOO LOUD BECAUSE THERE IS NO MIX AT ALL. Measured: confirm
-17.7 dB mean / -0.0 dB peak, 3 dB hotter in mean than the music and 6.4 dB
above move. No volume or gain value exists anywhere in export/ or authored/,
so every clip plays at unity on one bus. Decoder: is per-cue or per-bus gain
on the disc -- the cue table is the obvious place and cue 1103 is already
decoded. Port: gains at PLAYBACK as data, and explicitly NOT normalisation in
the exporter, which destroys the relationship between clips and cannot be
undone by a modder.
F3 SOMETHING IS MISSING ON THE TITLE SCREEN. The export carries one music file
and the port plays nothing on the title. Which cue does the title play, and
is there a sting on the plate or on accept? A negative needs a positive
control: find the menu's cue by the same method first.
F4 (A) SKIPS FORWARD THROUGH THE BOOT AND WE IMPLEMENT TWO OF THREE PRESSES.
In the game: skip video, reveal plate immediately, accept plate. The middle
one is missing here. Whether the game snaps both clocks forward or only
reveals the plate is a question, not a detail -- and it is a cheap second
route to the plate-arrival question, since a press that skips to the plate
says where the game thinks the plate belongs.
H3, the plate delay, is ACCEPTED -- "feels the same... sufficient". Left
unattributed rather than closed green.
"Looks good! Cannot notice any obvious difference from the actual game.
Mark logos as done." -- 2026-09-02
Not "the check passes": a person compared the port against the real game and
could not tell them apart. That is the oracle, and it is the strongest result
this port has produced. The sole-focus order is lifted; both agents return to
their milestones.
The fix was one word -- pose_at ASSIGNED the settle instant instead of clamping
to it, so every query returned the settled pose whatever the clock said. The
same line manufactured the false green: the capture harness shoots after two
frames, so it was photographing t~=2 units, which looked settled only because
everything looked settled. The 0.01 % agreement that closed H2 was measured
through the accident. One bug produced the defect AND the evidence of its
absence.
Verified here before it went to the human, by film rather than by claim:
motion 16.4 % -> 27.7 %, distinct luma states 26 -> 43, the publisher ramp 6
steps -> 13 in one continuous run, and the developer splash's interrupting
0.50 s freeze gone. The publisher trajectory rises to a peak and settles back --
the crossfade signature.
The port then closed a gap motion-census names in its own header ("a wrong ramp
that moves every frame passes here") with a shape check pre-registered from the
disc, measured off a film, on a non-overlapped strip, in ratios so the texture
divides out: rise:last declared 1.20, measured 1.20 exact.
Kept as the standing lesson, because it is the fourth instance: an instrument
that sits below the thing under test cannot see it fail. Ask of any new check
what it would still report if the feature were entirely absent.
Explicitly NOT claimed: P5's gate is "a human clicks through it" and nobody has
said the milestone is met. The briefs say so, and say not to record it on the
human's behalf.
The decoder's end-to-end pipeline work returns to normal priority rather than
being dropped -- it is what decides whether the port's 60 units/s matches the
game. The ramp is now right in SHAPE and unverified in DURATION.
A human on a GPU at ~140 fps: "the logos just switch, there is no animation at
all." Measured from a real boot with --film at 0.05 s, then per-frame change:
splash moves 1.30 s of 7.95 s = 16.4 %
publisher splash 0.30 s of motion, then 3.20 s FROZEN
developer splash 0.35 s + 0.25 s, then 2.40 s FROZEN
distinct luma states in 7.95 s 26
A 45-unit build-in cannot be drawn in 26 states, and a fade does not hold one
picture for 3.20 s. The frame counter says 24.8 fps achieved; both are true --
the port is DRAWING 25 times a second and CHANGING almost never.
🔴 Why every check passed, which matters more than the bug:
frozen sweep drives the clock BY HAND -- proves the renderer can draw
pose N, never that the poses are drawn in sequence
settled compare 0.01 % against the capture -- a screen frozen 84 % of the
time matches a settled reference PERFECTLY, that is what
frozen means
achieved fps counts frames DRAWN -- the same pixels 25x/s scores
identically to animating
Every one measured throughput or a pose. None measured CHANGE. Same shape as
InputEventAction bypassing the input map: the instrument sat below the thing
that was broken, so the break could not appear in it.
tools/motion-census closes the class. It measures change and nothing else, and
its --selftest asserts it separates a fade (97.4 % moving) from a switch (2.6 %)
from a frozen film (0.0 %) -- a detector that cannot tell those apart would
report the same green line on all three.
Both briefs: this is the SOLE focus. The port reproduces before changing
anything and gates every fix on a film rather than a still. The decoder maps the
whole pipeline end to end -- disc bytes, the game's per-frame update (does it
interpolate between keyframes or hold?), what is submitted per frame, and what
Canary does to it before a capture records it -- delivered as a SERIES, not a
settled value.
The port should also record the refutation against itself: H2 reads ANSWERED on
the strength of the frozen sweep. The mechanism half stands, the blur is a baked
companion texture. The behaviour half does not.
Both are pushed. The decoder's carries the R1 register reclassification and
tools/stale-instrument; the port's carries the two input fixes, verify-input
and BLOCKED H1-H3. Each branches from that agent's own tip, so it is a
fast-forward on the line they are already on -- and the port must merge before
touching input or it will re-derive a fix that is already asserted.
A human played the port on real hardware for the first time (2026-09-01) and
found four things. Two were port defects, fixed. Two are open and are now both
agents' focus: the PRESS (A) plate arrives late, and the splash fade/blur is
weaker than the game's.
Their verdict on method is the reason this is a brief change and not a ticket:
"the agents were essentially guessing and trying to copy what one would see,
but while they did get close it still is not quite right"
Close-but-not-right is the signature of reproducing APPEARANCE instead of
deriving MECHANISM. So the Decoder's focus block asks, in order: is there a
post-process pass at all, what is it, where do its parameters come from -- and
only then what curve. Both routes, dynamic (GPU state, shader constants, render
targets; add logging to Canary, it is theirs read-write) and static (.pe, the
DB, the paks), with each fact labelled by which produced it.
TEMPORAL-VERIFICATION.md is the other half, and it generalises past the
splashes. We have been photographing the game at time t, and t is never the
same twice: emulator speed varies with host load, Canary presents at ~28.1 fps,
the capture path costs a variable 0.1-10.8 s, and a long-lived x11grab stream
degrades and then freezes. The register already carries FOUR refutations of
exactly this shape. The replacement rule: record a film, not a photograph;
align by CONTENT, not by clock, and report the lag as a measurement rather than
minimising it away; prefer ordering, counts, durations and shape over any value
at a wall-clock instant; anchor on an event; report achieved fps against
requested fps.
Also into both briefs: the input set. The port had no joypad binding for (A) or
(B) and nobody noticed for a whole milestone, because --script sends
InputEventAction, which BYPASSES the input map -- so every check asserted the
code below the map and nothing about the map. The Decoder is asked to DECODE
the full set the game reads rather than discover it by pressing buttons; the
Port is told input is verified at the device level or not at all.
And both briefs now point at the R1 register reclassification, because two of
the ten re-opened entries land on this focus: "the declared keyframe timeline
reproduces the captured splash" is 🟡 our-reader, and the rest() pair is open
in BOTH directions -- while the two splashes are the only screens that reach
that fallback.
The mounts landed but the agent could not learn of them: I documented them in
CONTAINER-NOTES.md, which the decoder's prompt does not list, and then restarted
the container -- so a fresh session with no memory of the exchange had a 586 MB
database and a decompressed image sitting unmentioned in its filesystem.
Now in the PROMPT itself, not only in a document, because the prompt is the one
thing a new session is guaranteed to read. CONTAINER-NOTES.md is also added to
its reading list.
And the caveat that matters more than the asset. The .pe is PRIMARY -- the bytes
the console executed. The database is somebody's ANALYSIS of them, produced by a
disassembler that had to guess, and it is wrong in the ways disassemblers are
wrong: misdecoded mnemonics where data was read as code, function boundaries
short or long or merged or split, coverage missing entirely for code reached only
by indirect dispatch, and names that are derived rather than symbols.
So a finding resting on a database row is not established until the bytes agree:
read the same address out of the .pe and check. Where they disagree the image
wins, and the disagreement is itself worth recording, because it tells the next
reader which parts of the database to distrust.
A fast index into 9.2 MB of machine code, not a source of truth.
The decoder had neither, and reported the gap precisely: four scripts in this
repo READ /work/xenia-rs/sylpheed.db and nothing produces it, so the whole static
PPC route was consumers with the producer missing.
Both exist on the host and are now mounted read-only: the 586 MB database (25 481
functions, 851 classes with RTTI, EH tables, imports, 1.8M indirect-dispatch
candidates) and the decompressed image.
The image is the more useful of the two. It is a FLAT VA DUMP -- file offset =
VA - 0x82000000 -- so reading a known address needs no XEX decrypt, no LZX, and
no booted emulator. The decoder had independently recovered the same bytes by
dumping /dev/shm/xenia_memory_* and validating against the GamePart table, which
is good work and a sound method, but it noted itself that needing a running
emulator is a bad dependency for something the entire static corpus rests on. It
does not need one.
Also recorded that an earlier claim the .pe was STALE was tested and refuted, so
nobody re-litigates it, and that instructions.raw is an INT rather than hex.
Written down as reference material, explicitly NOT a deliverable: they are
read-only, they come from outside the repository, and a fresh checkout elsewhere
has neither. Reimplementing the producer belongs in sylpheed-formats, and until
it exists every static finding rests on an artefact this project cannot rebuild.
Both agents read the protocol, their mission and the shared tooling from their
OWN checkout, and both work on topic branches -- so without an explicit sync
they follow whichever version of the rules existed when the branch started.
Found concretely: tools/audio-capture and two protocol revisions were on main
while the decoder worked for hours from a branch that had neither. The port had
merged on its own initiative and did have them, which is exactly the kind of
divergence nobody notices until the two disagree about what the rules say.
The audio work was three parts and I shipped two. The transcode-fidelity method
and the pinned 5.1 downmix landed; the null sink -- the only one that answers
"what does the GAME play" -- I deferred to "the next natural rebuild window" and
then rebuilt both images four times without doing it.
pulseaudio-utils is now in both, with tools/audio-capture wrapping it: a null
sink is a real device as far as an application is concerned, so Canary and Godot
open it normally and parec records what they emit.
This unblocks the decoder's Q8. The cue-to-event bindings are currently a name
match against the authors' own identifiers -- a plausible guess, not a
measurement -- and capturing what the game plays on a menu move converts them.
`audio-capture run` reports the peak level and warns when the capture is silent,
because silence is the failure that looks like success: a WAV of exactly the
right duration, full of zeroes, because the application opened a different sink.
A duration check alone passes it, which is how a confident wrong number gets
made.
Three defects, all mine, found by checking instead of assuming.
**They never exchanged a word.** SendMessage=0, ListAgents=0 across both new
sessions. PROTOCOL.md specified in detail what a message MAY and MAY NOT do and
never said how to send one or that the other agent was addressable -- they knew
that last time only because the human told them directly, and rebuilding with
fresh volumes wiped it. Policy without mechanism is prose. Now documented with
the two addresses, a worked example, and an instruction to introduce themselves
on the first iteration rather than waiting to have a question.
**The decoder lost the disc and the ISO.** They used to arrive inside the project
mount and silently stopped when /work became a clone. Silently is the word: the
disc-gated tests SELF-SKIP without SYLPHEED_DISC and report green, so a whole
test suite would have passed while measuring nothing. Both are now mounted
explicitly, the ISO at a stable path so run-canary does not depend on host
directory names.
**Both agents shared one Claude state directory.** They share the host's
~/.claude, and once both working directories became /work they resolved to the
same projects/-work/ -- two supposedly independent agents writing to one place,
which undoes the point of separate checkouts. Each now has its own volume, seeded
once from the host with credentials only, so a token refresh writes locally and
neither can corrupt the host's auth.
Also widened the pacing rule. It banned ScheduleWakeup by name; the decoder then
scheduled itself an hourly cron job -- not harmful, but the same instinct that
ended a run yesterday, through a door I had left open. Now: no self-scheduling by
any route.
Mount audit after the changes: shared and intentional are the exchange volume and
the read-only credential seed. Everything else -- repo, Claude state, cargo,
target, canary, disc, ISO -- is per agent or one-sided.
The last structural fix for the collision class that has bitten three times. Both
containers now clone the repository into their OWN named volume instead of
bind-mounting a human's working tree, so an agent's local git config cannot
capture a human's commits, a credential helper cannot leak a container-only path
onto the host, and a `git add -A` cannot sweep another party's in-flight files.
Cloned once at startup and never auto-pulled: pulling under a running agent
moves files out from under whatever it is mid-edit, which is the same bug again.
Accepted knowingly: Claude Code keys per-project memory off the working
directory, so moving off the host path starts that memory empty. The corpus in
docs/ is the memory that matters and it travels with the clone.
Other changes:
* docker/agent -> docker/decoder; the launcher is sylph-decoder. Roles, not
"the agent", now that there is more than one.
* /reborn is gone -- one repository now, so the port reads HANDOFF from its own
checkout rather than through a live read-only mount of someone else's tree.
* Canary mounts separately at /canary; it stays a fork tracking upstream.
* A shared `sylpheed-exchange` volume at /exchange, with tools/ on PATH so
`share` is available in both.
* The decoder's credential file gets the .host-copy treatment the port already
had -- `credential.helper=store` rewrites by rename-over-target, which is
EBUSY on a bind mount and reports a fatal that is not one.
* Budget split deliberately: decoder 5 cpu / 6 GB, port 3 / 4, leaving room for
the planned Referee. "Half the host" was right when there was one agent.
Prompts move to docs/agents/ and are rewritten around the protocol: the oracle
is the running game, dynamic RE stays with the decoder, each iteration must
attempt to refute one claim of the other, and neither may verify its way out of
its own role.
**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.
Two fixes to push-work and the policy that goes with them.
**Tags.** The port's exporter depends on sylpheed-formats BY REVISION, so a
commit of ours is part of its build -- and the commit it pinned lived on one
auto/* branch and nowhere else. Deleting that branch orphans it; squash-merging
it is worse, because squash creates NEW commits, so main appears to contain the
work while the pinned sha becomes unreachable and the port stops building for a
fresh checkout. Silently, at their build, long after the breakage.
push-work now pushes with --follow-tags, which publishes annotated tags
reachable from the pushed commits, and MISSION.md says to tag whatever the port
needs. formats-pin-2026-08-29 at 76653ca is the first, created after the fact.
**Credentials.** This script set credential.helper as --local config, which
PERSISTS in the repository. The repo is a bind mount the host also uses, so the
host's git inherited a path that exists only inside the container and every host
push failed with "unable to get credential storage lock". Now applied with -c to
the single push. Same fix already landed on the port side; this copy still had
the bug and would have re-leaked on the next push.
Committed with a pathspec so the agent's in-flight work is untouched.
Probed all 97 movies. 28 are wmapro 48 kHz 6-channel 5.1 -- ADV.wmv and every
S*.wmv story cutscene; the other 69 are wmav2 48 kHz stereo, every RT*.wmv and
hokyu_*.wmv. The split is cinematics vs in-mission radio chatter.
Both movies the menu milestone needs, ADV.wmv (the boot/attract intro) and
S00A.wmv (the new-game intro), are in the SURROUND group.
Why it matters: one ffmpeg command over dat/movie/ produces two different kinds
of result and records neither. The 69 stereo files pass through unchanged; the 28
surround files get downmixed 5.1 -> stereo by ffmpeg's DEFAULT matrix, folding
centre-channel dialogue into L/R at a weighting nobody chose and which is not
stable across ffmpeg versions. That is a content decision inherited by accident,
so it should be stated explicitly and recorded beside the command.
Credit where due: found by the human while checking a transcode, verified
independently here and widened from one file to the whole disc.
Two METHOD entries from the same episode, both about measurement rather than
format: don't probe a file another process is still writing (a half-written
transcode reported 33 s against a 137 s source, no error, nearly a filed bug),
and a difference-signal RMS is meaningless before cross-correlation alignment
(-34.2 dB against a -25.3 dB source looks like failure and is inconclusive).
Three things, all from parsing ptbtn0Nf.rat as a build.
1. THE RING SPINS. Its two keyframes differ in exactly one field: rotation_deg
ramps 0 -> 360 with position, scale, alpha and tint all constant. A spin in
place, the same shape as the GP_BUNK example already recorded.
2. THE ORACLE CONFIRMS THE GAME RENDERS IT. In the OPTIONS-focused capture the
ring's bright head sits in a completely different angular position from the
sprite's own -- caught mid-spin. This is a SECOND independent confirmation that
rotation_deg is drawn, now on a different screen and a different element from the
ptloop sweeps, and it raises rotation's priority: it is not a title-only concern
that sits off-screen at rest, it is the main menu's focus marker.
NO ANGLE IS QUOTED. A brightest-region centroid says ~250 deg, but the control
refuses that precision -- rotating the sprite by a known 30/90/180/270 and
re-measuring gives errors up to 19.8 deg. What survives the error bar is that a
<=20 deg error cannot manufacture a ~250 deg displacement.
3. WHICH PLACEMENT WINS -- correcting this page's own earlier caveat, which said
to use the leaf only for elements the parent does not declare. Right for a BASE
record, wrong for an f record: the parent declares NO element for ptbtn0Nf.rat at
all (zero of build 5's 16), so the f record's placement comes from its leaf for
BOTH elements, label included. The label's (-7,-7) is load-bearing -- the f
sprite is 13px larger per axis and -7 keeps them concentric (535+96/2 = 583 vs
542+83/2 = 583.5). Corroborated against the oracle: the focused-minus-unfocused
region is x 505..703, and the leaf predicts a right edge near 707 where the
parent reading predicts 714.
Also exposes UiBuild::records (name -> (offset, size) of a nested .rat leaf).
Nested records were parsed into a PRIVATE map, so a consumer holding a UiBuild
could not locate a leaf's bytes at all -- which is exactly what blocked the port
from reaching the ring.
The port is forbidden from reimplementing media assembly and Static.slb is
exactly that case: no RIFF, no seek chunk, no XACT container, just a packed run
of whole 2048-byte XMA1 packets, so a wave is defined only by (offset, packet
count) and the header has to be synthesized. That step now happens once, in the
crate that owns the format, instead of in each consumer.
`slb::xma1_wave_riff` wraps raw packets; `media::se_wave_riff` looks the bank up
and reads just the packets asked for. Both reuse the existing synth_xma1_fmt /
build_riff, which are already byte-identical to what tools/re-capture/
slb_extract_wave.py writes -- so this is exposure, not a second implementation.
It reads a TARGETED range rather than the whole bank, and that is load-bearing:
Static.slb is the ONE entry of sound.pak's 9 519 whose declared extent runs past
the end of the extracted segments -- by exactly 616 768 B -- so reading it whole
fails outright on this extraction. Every cue we need is in the first few hundred
KB. Recorded rather than worked around silently.
Verified as an artifact, not a compile: all three cues decode through ffmpeg to
mono 48 kHz PCM at 0.533 / 0.344 / 1.016 s, non-silent (rms 2085 / 2985 / 4327,
peaks 29813 / 16973 / 32767). The refusal path is exercised in the same run --
an impossible packet count is rejected rather than returning a short stream,
because a truncated XMA decodes to plausible-sounding garbage.
Also adds docs/re/captures/ORACLE-CAPTURES.md: an index of the nine canary
framebuffer captures already in this repo, and a plain statement that THEY are
the reference and `screen render` is not.
The port needed ptbtneff01.t32's placement and was about to author it from an
eyeballed PNG measurement. It does not have to: a `.rat` leaf needs no new
reader. Its first 32 bytes have a bundle header's shape -- "RATC", 0x3c
declaration-entry size at +4, element count at +20, design 1280x720 at +24/+28 --
so ui_layout::parse_build reads it unchanged.
The control is the base record, whose position is known independently: the parent
screen reports ptbtn01.rat resting at (542,162), and parsing the leaf alone
returns ptbtn01.t32 at (542,162). It reproduces all five buttons.
Positions are absolute design-space top-left. The ring rests at (500, 156/236/
316/396/476) for buttons 1-5 -- a uniform (-42,-6) from each button's own rest,
identical in the Japanese bundle. The bright label is a uniform (-7,-7).
Two things recorded rather than smoothed over: a leaf's placement DUPLICATES the
parent's rather than being relative to it, and the two copies are not always
byte-equal (ptbtn04's parent says y=401, its leaf says 402) -- the parent is what
compose honours, so the leaf is the source only for elements the parent does not
declare, which is exactly the ring. And `screen render --focus` is blind to the
ring for the same reason the port's exporter was: el.focused is name-based on
top-level elements and neither walks into the leaf.
1 SPLASH ADDRESSING (was blocking P3). No content predicate exists: design size
fails (every extra composable bundle sampled is 1280x720, like every screen) and
element count fails (fragments run 2..15, the splash halves are 3 and 7). But
GP_TITLE needs none -- `--all` adds exactly four bundles there and all four are
real screens, with the --all index equal to the pak entry index 1:1. And there
are TWO splash screens: 11/14 are the developer logos, 10/13 are the SQUARE ENIX
publisher wordmark, which the port did not have and which the boot shows first.
2 FADE-OUT (was blocking P3). It is (a), and it is bigger than the fade quad.
Every element ends on exactly ONE untimed keyframe, which rules out (b); that
block is where the screen plays out -- quad to a=255, buttons/labels/glows to
a=0, frames hold. (c) is refuted by a null test that discriminates: a black quad
alone holds the button/background brightness ratio constant, and through the
fade it falls 6.50 -> 1.94, 3.4x monotonic.
3 FOCUS (saves P5 rework). Over-vs-instead is unobservable -- the focused sprite
covers the base at 100.0% of base-visible pixels on three pairs once aligned
(true offset (7,7); the centre alignment reads a misleading 78-84%), and
compositing both ways differs by RMSE 1.1 inside the button rect. The real defect
is the focus record's SECOND element: ptbtn0Nf.rat declares ptbtneff01.t32 (a
42x46 glowing ring, focus only) plus the bright label, where the base declares
one sprite. That ring is the marker the port draws nowhere.
5 GAMMA. The capture is not neutral: capture ~ 255*(render/255)^g, g ~ 1.34-1.49,
and the chain says it is a ramp the GAME installed, not a capture artefact. So
RMSE against captures has a floor. Reach stated: the flat patches are all dark
(render ~0-60), so midtones and highlights are unconstrained.
4 ROTATION is a human's call and is recorded in MISSION, not acted on -- the port
rotating while the reference renderer does not would make verify-screen report a
large diff meaning "the port is right". The RE half is answered: rotation is
about the declared pivot, measured against a GPU capture.
The focus record's +20 element-count word is marked 🟡 not ✅ -- read on GP_TITLE's
ten button records only; the disc-wide check is written and still running.
Two runs were killed this session for looking stuck. Measured: 1 318 s serial,
no output while it runs. Also records that `build-reborn test` forces
--workspace and silently ignores a `-p`, so scoping needs the cargo
passthrough (`build-reborn t -p <crate>`).
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).
`ratc::parse` named each child by scanning backwards for the last printable run
of bytes before its magic. The format states the name explicitly instead, in an
`opt ` block: `"opt " | BE32 len | name | NUL | 3 bytes | magic` -- the same
block `ui_layout::opt_link` already read for a button's focus link.
The scan agrees with it 17 918 times out of 17 942 and is wrong 24 times, every
one the same failure: the 3 trailing payload bytes are themselves printable and
beat the real name. For `pteff05.t32` those bytes are `38 41 58` = `8AX`, so the
full-resolution background of all five menu screens registered under a name no
element declares, resolved to no sprite, and `compose` dropped it through an
early `continue` that -- unlike the two arms above it -- records nothing. The
screen lost its background and `screen render` still reported "all resolved".
`8AX` was never a name. Docs that treated it as one are corrected here.
Disc-wide, and the control is the 17 918 the scan already got right: the `opt `
reading reproduces every one of them. Effect on the five screens is the
signature of the same art at twice the resolution -- mean brightness unmoved,
high-frequency detail x1.15..x1.30 -- which is what the separately-measured
`ui-8ax-fullres-background` result said the game draws.
Also closes a long-standing dangling reference: `pmbase.t32`, recorded as "on
the disc nowhere", is the `GP_STAGE_CLEAR` child the scan called `8AX`. RATC
sibling references now resolve 10 148 of 10 148.
Verified: 114/114 sylpheed-formats unit tests (including two new ones pinning
the `8AX` case byte for byte and the no-block fallback), and every disc-gated
integration suite in sylpheed-formats/sylpheed-cli.
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.
ui-prm-primitives recorded that where a .prm paints on a screen without a
measured order is unsolved. For the developer splash it is now measured.
Every frame opens with two full-screen draws before any sprite. The
second is untextured in all 212 frames with a constant vertex colour of
FF000000 -- opaque black -- matching palogo_eff0.prm's declaration
exactly: kind 0x10, pivot (640,360) -> 1280x720, one keyframe, a = 255.
So the splash backdrop is an opaque black full-screen quad from the
bundle itself, painted behind every sprite, which is why a splash render
needs --black rather than the default backdrop.
Not a general rule, and said so: the measured main-menu order puts
pteff02.prm at position 4 and pteff00.prm LAST, the latter being the
screen-transition fade.
And a correction to an existing row. measured_paint_order returns
[0, 2, 4, 6, 1, 3, 5] for the splash, described as "the .prm, then all
three glows, then the three logos". But the glows and the logos never
appear in the same frame -- 0 overlapping frames in 235 -- and two
elements that never co-occur have no observable relative depth. Between
those halves the vector records the order they were SEEN IN, not a
front-to-back relationship.
That does not make the render wrong, and element 0 is a real depth
observation since the .prm co-occurs with everything. But the type of the
claim matters: reading the vector as depth invites compositing all seven
elements at once, which is exactly what does not reproduce the screen.
METHOD: two things that never co-occur have no observable relative order;
when recording an order, note which pairs actually appeared together.
The activation decision is code, and MISSION already says the port
authors the sequence -- so the useful move is to hand over the sequence
measured rather than chase the code. From the 235-frame draw capture, at
1 frame = 1/30 s (2 units/frame, 1 unit = 1/60 s, both settled in Q1):
publisher: SQUARE ENIX logo f1-90 90 frames 3.00 s+ at 0.00
(gap, nothing drawn) f91-93 3 frames 0.10 s at 3.00
developer: both glows f94-115 22 frames 0.73 s at 3.10
developer: both logos f116-211 96 frames 3.20 s at 3.83
Three limits, stated with the numbers rather than after them. The capture
opens with palogo_sqex already at a=255, so the publisher phase began
before the window and 3.00 s is a FLOOR -- every "starts at" is relative
to the capture, not to boot. palogo_anima and palogo_anima_eff get 0
draws in all 214 frames, so a third pair's phase is not in this
measurement. And it is one capture, one run: the glow->logo switch being
a single frame boundary with no overlap is a strong shape, but each
duration is one sample.
What is solid is the part that matters: the 0.73 s and 3.20 s phases are
each within 2% of their element's declared span, so the durations are the
bundle's own and only the ordering is authored. That is the difference
between a port transcribing timing and inventing it.
Last iteration I withdrew "a bundle is a palette" because the evidence
did not choose between selective activation within one bundle and two
compositions shown in sequence. The alternative can be killed from the
disc, which I had not tried.
Hypothesis 2 needs a bundle declaring the GLOWS WITHOUT THE LOGOS. There
is none. Every GP_TITLE entry carrying palogo elements:
10, 13 (publisher twins) palogo_eff0.prm, palogo_sqex, palogo_sqex_eff
11, 14 (developer twins) palogo_eff0.prm + all three logos + all three glows
Four entries, and each developer entry declares the complete set of six.
So whichever bundle was active across frames 94-211 -- entry 11, entry
14, or both in turn -- it declared the logos and the glows, while the
game drew two sprites at a time in disjoint phases.
Therefore only a subset of a bundle's elements is drawn at any moment,
whatever the bundle-loading story is. The conclusion no longer depends on
resolving how many bundles are involved, which is why the texture-base
test's failure stopped mattering.
So the claim is reinstated -- this time by eliminating the alternative
rather than by assuming it away. What worked was not a better capture but
asking what the competing hypothesis would REQUIRE on the disc and
finding it absent.
METHOD: a hypothesis that predicts an artefact can be killed by looking
for the artefact, which is often far cheaper than measuring the behaviour.
Last iteration I wrote that a bundle is a palette whose elements are
selectively activated. The disjoint glow/logo phases have two
explanations and I asserted one:
1. one bundle, some elements run then others;
2. two bundle-loads shown in sequence (entries 11 and 14 are twins
declaring identical sprites).
The draw log's tex[base=...] looked like it would separate them. It does
not, and the control is in the same table:
publisher splash f1-90 0x11C30000, 0x10000000
glows f94-115 0x11C30000, 0x10000000
logos f116-211 0x11C30000, 0x10000000
The publisher splash is certainly a DIFFERENT bundle from the developer
splash, and it uses the same base. So 0x11C30000 is a reused upload slot,
not a bundle identity, and the test cannot choose between the two
hypotheses.
Survives: a bundle's declared elements are not what gets drawn.
palogo_anima and palogo_gamearts carry byte-identical keyframe times and
in the same run one is drawn 95 frames and the other none -- and
whichever twin was active declares both. The phases are strictly disjoint
(0 overlapping frames in 235).
Withdrawn: the mechanism. The practical consequence is unchanged --
compositing every element of a bundle does not reproduce what the game
shows over time -- but the why is not established and I stated it as
though it were.
What would separate them: a per-draw capture recording the bundle each
draw came from, or a file-IO log showing whether a second RATC entry is
read between frames 115 and 116.
METHOD: a shared resource address does not identify the resource's owner;
and state the mechanism as a separate claim from the observation, or the
weaker one inherits the stronger one's evidence.
Two checks on last iteration's "sequential, not simultaneous" reading.
First, the phases really are disjoint. If glows and logos ever shared a
frame the claim would be wrong. Across all 235 captured frames the count
of frames containing both is ZERO, and the switch is a single clean
boundary -- f110-115 draw 1280x720 + 262x108 + 525x90, f116 onward
1280x720 + 243x86 + 499x72. Two sprites either side, no transition frame.
Second, and larger: a third of the bundle is never drawn. Entry 11
declares three logo/glow pairs and only two appear.
palogo_gamearts / _eff 95 / 22 frames
palogo_seta / _eff 95 / 22 frames
palogo_anima / _eff never
palogo_anima declares the SAME keyframe times as palogo_gamearts. Two
elements with byte-identical data, 95 frames and 0 frames in one run.
Reach: the capture covers frames 1-214, so this is "never in the window".
So a bundle is a palette, not a script. Its elements say what to draw and
for how long; which of them run, and when each starts, is decided outside
the placement data. That is the same conclusion the boot-order work
reached from the other end -- the driver is code, not data -- now with a
per-element measurement behind it.
For the port, concretely: compositing every element of a bundle does not
reproduce what the game shows over time. It is the right thing for a
static screen that settles, and it is not a timeline.
METHOD: two elements with identical data and different outcomes is the
strongest possible evidence that the decision is elsewhere.
Tested whether the splash timeline, played, reproduces the capture --
the last gap in the animation model. Half of it does.
Durations match. At 2 units/frame under the shifted reading, from the
235-frame draw capture of the developer splash:
glows drawn f94-115 (22 frames = 44 units) declared ~0..45 = 45 97.8%
logos drawn f116-211 (96 frames = 192 units) declared 15..210 = 195 98.5%
Each element is on screen for its declared span to within 2%.
Starts do not. Every glow declares the same times 15,30,45 and every
logo the same 15,30,190,194,206,210, so on one clock they would overlap
almost entirely -- and they do not overlap at all. The glows run 94-115
and the logos 116-211, strictly sequential, the logos starting the frame
after the glows end. Fitting one origin needs f0 ~ 93.5 for gamearts_eff
and ~103 for gamearts, about 19 units apart, and aligning one throws the
other off by ~9 frames at both ends.
The obvious candidate is refuted. parse_placements reads each group
header as (element index, keyframe count) plus one undecoded LEAD-IN
WORD -- exactly where a per-group start offset would live. It is
0x00000000 for all seven elements, glows and logos alike.
Reach: not the keyframe times (identical within each family), not that
word (zero), not declaration order (which interleaves logos and glows
where the observed sequencing groups them), not the RATC child order.
What remains is that the sequencing is code-driven, which agrees with
what the boot-order work concluded independently.
For the port: a group says how long an element animates and what it does,
not when it starts relative to its neighbours. The observed order on the
developer splash -- both glows, then both logos -- is measured for one
screen, not a decoded rule, so the sequencing has to be authored.
METHOD: when a model reproduces durations but not positions, the missing
piece is an origin, not a rate.
The model's sharpest prediction, tested with its control. The draw log
says that on the developer splash the _eff glows are drawn on frames
94-115 and the logos on 116-211, so at the moment the reference capture
was taken EVERY glow is already finished -- including the two that have
plateaus and which rest_plateau therefore renders visible. Suppressing
them should help on the splashes and hurt where a screen genuinely
settles.
publisher splash +0.9604 -> +0.9982 +0.0377
developer splash +0.9659 -> +0.9980 +0.0321
title (control) +0.9500 -> +0.9480 -0.0020
main menu(control) +0.9460 -> +0.8544 -0.0916
EXTRAS (control) +0.9440 -> +0.8370 -0.1070
Both splashes jump to about 0.998; all three persistent screens get
worse. The control is what makes this a finding rather than a
coincidence: the same edit helps exactly where the model says it should
and hurts exactly where it says it should not.
So rest_plateau is not over-drawing in general -- it over-draws on
TRANSIENT screens. A plateau mid-animation means the element is held at
that point in the timeline, not that it is on screen once the screen has
settled. Where a screen settles, the held pose IS the settled pose and
the rule is measurably right.
And that answers the question left open several iterations ago -- what
"rest" means for a transient element. It does not mean anything: the
splashes never rest. A static composite of them can match a chosen frame,
and about 0.998 is what these captures' frame is worth, but the format
does not answer a question the screen never poses.
For the port: play the timeline for the two splashes, which the settled
keyframe timing now supports, and composite statically for title, main
menu and EXTRAS.
METHOD: an edit that improves one set of cases is only interesting once
you have shown it damages the cases where it should.
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.
