The star-map family was the last unported part of the tree, and it never had
a ticket. Two of its scripts become `reach atlas map` verbs, nested under
atlas like planet (D-243: the Reach map is the ladder's top rung):
- `reach atlas map data [--check]` regenerates client/data/star_map_data.json.
The regenerated file differs by one line: `_meta.note`, which named the
old script's path.
- `reach atlas map svg` renders the concentric SVG (+ PNG), byte-identical
to the old script's output on the same data.
make check-star-map and star-map-data stay as one-line delegates, because
pre-pr-client and pre-pr-validate depend on check-star-map.
generate-star-map.py, its seed, sculpt-star-map.py and tune-star-map-topology.py
are archived, not ported. The generator rewrites docs/design/star-map.json
unconditionally from an S-NNN-keyed seed, so re-running it would erase the GJ
migration and every hand edit since; sculpt and tune only understand S-NNN
edges. .claude/rules/diagrams.md was telling agents to "edit the generator
and re-run it". It now distinguishes the live concentric render, the seven
frozen S-keyed sector .d2 files (T-1294), and the two SVGs that never had a
generator in the repo.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
tooling/db/ (a misnamed directory: connectors, not database work),
trellis-batch.sh and synth_ui_sounds.py become `reach assets`:
audio {health,generate,batch,post {convert,normalize,trim,pipeline}},
image {health,generate}, trellis {health,generate,batch}, and synth-ui.
The four audio bash wrappers are retired, and tooling/db/ is gone.
Parity, from baselines taken before anything moved:
- the four UI-sound WAVs and the harmonic-synth WAVs (exponential and linear
decay) are byte-identical
- the ffmpeg pipeline's decoded PCM is identical. Its .ogg bytes are not,
even between two runs of the OLD code: Ogg picks a random stream serial,
so the encoded file was never the right thing to compare
- the network success paths can't be run in a gate (Stable Audio and Trellis
are kept off, Gemini costs money), so tooling/test_assets.py stands up a
fake Gradio and pins every payload: the audio submit, Trellis's six-call
session sequence with its 9-input image_to_3d, and the Gemini body. It
failed when one Trellis value was mutated (7.5 → 7.0)
Failure classification, in endpoints.py, is the point of the port. The
services are OFF by design (VRAM on tower-of-joy, D-17), and the topology doc
warns against "fixing" one by restarting it. So a refused connection says OFF
and asks for the service to be turned on rather than restarted; a 4xx/5xx says
the request was rejected; 401/403 says credentials; 429 says quota; and an
unreachable Gemini blames the network, not VRAM.
Behaviour changes, each a failure that used to read as success or crash:
- audio batch and trellis batch exited 0 with failures in their summaries;
they now print the summary and exit 1
- trellis generate on a missing image crashed with a TypeError
(print(..., indent=2)); it now names the file, and checks it before the
service so a typo is not reported as an outage
- the ffmpeg pipeline left its intermediates behind when a step failed
Structure: the connectors called each other as subprocesses (batch spawned
the connector, which spawned audio_post) and parsed each other's stdout. They
are now function calls, and ffmpeg is the only exec, through core/process.
ensure_venv() is removed: it os.execv'd into .venv, which D-263's exec rule
forbids, and reach declares the dependencies itself. config.json moved into
the domain deliberately, and the local-services rule follows it.
Output contract: results are still JSON on stdout with the same keys, so skill
readers keep working. Failures are an exit status with a Fix line, never
{"ok": false}. The audio-gen, glb-gen and image-gen skills, Araminta's agent
file and the allow-list are updated to match. glb-gen's "trellis-batch.sh is
hardcoded to one category" caveat is gone: batch takes --input-dir or --names.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
`reach wiki stats` and `reach wiki gttr-hook` replace tooling/db/wiki_sync.py
and populate_gttr_hook.py. Both are output-identical to the originals:
`stats` byte-for-byte, and all 301 extracted GTTR hooks line-for-line.
wiki_sync.py moved whole, but generate_wiki() and import_from_wiki() are NOT
verbs. Before porting, the old `--generate` was run against a clean tree to get
a parity baseline. It changed all 301 system pages, +940 / -10,761, and was
reverted at once. It deletes the Celestial Bodies / Stations blocks (owned by
the Rust atlas sync, which it does not know about), deletes the
Industries / Exports / Imports rows (nothing writes those any more), and
rewrites star types where systems.db and the pages disagree. D-262, CLAUDE.md
and the wiki skill all described it as the routine, prose-preserving render.
CLAUDE.md and the skill now say not to run it; D-262 needs amending — T-1292.
Provenance moves to tooling/archive/, with a README naming what each script
did and why it is not run:
- pql-migrate/ (the T-1271 ruling)
- wiki-bootstrap/: assign-astro-ids + its catalog, migrate-s-to-gj,
patch-core-sector (hardcodes a dead path), fill-missing-globes,
generate-stubs and find-stubs (finds 0 stubs — Phase 1 is done),
backfill_cultural_corridor (a raw systems.db patch script, outside D-262),
and process-wiki-system-changes, whose last step is the destructive render
Also:
- stats() printed "run import first" and exited 0 when a table was missing;
it now fails with a remedy. generate_wiki() counted created pages after
writing them, so `created` was always 0.
- tooling/godot-cold-parse and godot-parse-sweep were never retired after
T-1283, and the pr-process skill still told agents to run them. Removed;
the skill and parse_sweep.gd now name the reach verbs.
- systems.db re-stamped: schema comments changed, and the stamp records the
schema file's SHA for tamper detection.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The sole generator of systems.db moves to tooling/domains/ledger/ and is
now `reach ledger import`. economy_import/ keeps its name (Rust comments in
server/src cite it); the entrypoint becomes service.py; schema_version.py
moves with the importer, which is where the version is defined.
The stamp survived the move, which is the thing that had to hold:
- generated_brands.toml is byte-identical (sha256 e748531…) before and after
- `reach check systems-db-stamp` reported STALE after the move (the registry
saw it) and OK after the regen
- the dry-run carries every count and warning of the baseline transcript,
and exit 2 — imported and stamped, coverage gate unmet — still reaches the
caller through @command
`make regen-db` survives as a one-line delegate, per D-263's muscle-memory
clause: about fifty files name it, including the headers of generated wiki
TOMLs and the remedies the push gate prints. `make economy-db` is retired;
it ran `reach generate brands` before the import, which the import already
does as its first step.
economy_import.errors is reconciled as DOMAINS.md asked. ImportAborted stays
as internal rollback control flow and never reaches a caller; the service
converts it to a ReachError carrying the remedy.
regenerate_brands caught cargo_binary's ReachError, printed it and raised
ImportAborted, dropping the remedy. It runs before the import transaction
opens, so there is nothing to roll back — it now propagates.
Both sys.path bootstraps are gone; they existed only because the directory
was hyphenated. The step labels ran [1/10]…[10/13]…[17/19]; one 24-step
counter now drives the event phase and progress.
Stale pointers fixed on the way: MIGRATION_SQL has lived in
economy_import/migration.py since T-1067, but the asset-pipeline rule,
DEVOPS and the schema comments still sent readers to import_economics.py;
the rule and DEVOPS also still named the check scripts T-1281 retired.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The 30-file tree moves under atlas as its third rung (D-243), ten verbs
fronting it. Each verb restates its module's options so `--help` describes
something; tooling/test_planet_router.py hands every declared option to the
module's own argparse and fails on drift, and now runs in make test-tooling.
The 2026-09-02 half of this move had converted the top-level imports and the
repo roots. Finishing it found what the half-move left:
- Lazy in-function imports, and all of sol_data/, still named siblings bare.
They resolved only through sys.path.insert hacks, so under reach the first
globe render in generate, batch or sol-import would have raised
ModuleNotFoundError. Qualified; the hacks are gone.
- 247 print() calls and a stdout progress writer that fired once per 8 KB
block. Report verbs (audit, quality) write through console.out, progress
through console.event, and download progress is throttled to 10% steps
so a job log is not tens of thousands of lines.
- Every error exit raises ReachError with a fix.
Two checks that could not fail:
- batch --verify-determinism printed a warning and exited 0 on a mismatch.
- import-provinces exited 0 with errors > 0.
Both now raise. The 271-body bake is only safe to re-run because the first
one holds.
sol-import --body is action="append" in the module but the router took one
value, so --body GJ0d --body GJ0e kept one. Now repeatable, and _flags repeats
list options.
test_conformance walked one level, so a nested group was reported as a verb
missing @command and its ten verbs were never checked. It recurses now;
proven by stripping @command from `planet quality` and watching it fail.
Stray PNGs from the 2026-09-03 runaway router-test run are parked in
.cache/t1288-stray-pngs/, not committed. Their reliefmaps differ from HEAD
while the heightmap regenerated byte-identical — filed as T-1291.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
35 payloads move to tooling/scripts/blender/ and stay outside package scope.
They run under Blender's bundled Python, which cannot see the repo venv, so
they physically cannot import tooling.core — holding them to the D-263 contract
would either fail the gate forever or force the contract to be weakened for
everyone, and the second is how a gate stops meaning anything.
Count verified by import rather than filename: 33 import bpy/bmesh directly,
and the two that do not are still payloads per their own usage lines.
garment-fit/make_logo.py is the one genuine non-payload and stays for T-1290.
The bash wrapper is retired rather than kept. Keeping it would have put the
install-resolution logic in two places, which is the duplication T-1286 had
just finished collapsing three copies of. domains/blender/service.py owns the
decisions — resolve_blender (native beats flatpak, ordering preserved),
resolve_payload, absolutise — and only run_payload performs. test_blender.py
pins all of them without launching Blender, which matters here more than
usual: the thing being launched is a 200 MB GUI application that writes GLBs.
`reach blender run` takes a registered payload name OR a path to any script,
because the wrapper served both — the spikes and the glb-gen skill hand it
one-off scripts of their own. An unknown name enumerates all 35 and exits 2.
The exclusion now defends itself. check_carve_out_stays_carved fails if
`scripts` is added to PACKAGE_ROOTS, if the payload directory empties (an empty
exclusion proves nothing), or if an __init__.py appears there (which would make
the payloads importable — the coupling the carve-out exists to prevent). All
three arms mutation-proved.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Twelve scripts retired, three domains registered. `reach` now covers nine.
generate: `generate-brands` and `generate-corporations` were the second and
third copies of the same 24-line build-if-missing-then-exec bash `tooling/atlas`
carried, so they collapsed into `core.process.cargo_binary` rather than being
ported. `import_economics` shelled out to the first of those, so it now calls
that helper — `generated_brands.toml` comes back byte-identical, and the stamp
registry swaps the retired wrapper for `core/process.py`.
pr: `watchlist-diff` derives its watched set from `generator_sources.py` instead
of restating it, so it cannot drift from the stamp check.
dev: the environment scripts split decision from performing, per D-263's
guarded-exec rule. `godot_plan()` and `worktree_plan()` decide what would
happen; `install_godot()`, `install_rust()` and `setup_worktree()` do it.
`tooling/test_environment.py` pins the version pin, both override precedences,
the already-current skip, the platform refusal and both worktree refusals —
none of them performed. `make setup` now installs reach first, since the
targets that install rust and godot are reach verbs.
Two live bugs found while porting:
- The clerk read its decision index from `decisions/README.md`, a path that
stopped existing when the DQR tree moved to `governance/`. Every clerk agent
has been grepping blind; its prompt pointed at the same dead directory.
- The conformance exec-check matched any `x.system()` regardless of receiver,
so `platform.system()` read as `os.system()`. Narrowed and re-proved against
a real mutant.
`process.run` gains `input=`, `timeout=` and a `ProcessTimeout` subclass so a
killed run stays distinguishable from a verdict. The pre-push hook no longer
merges the clerk's stderr into its stdout — under streaming the last merged
line is a JSONL event, which would read as an unrecognised verdict and block.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
reach atlas db / names / systems-done / check / verify / commit-and-sync /
update-field / flatness. Eight scripts retired, five of them bash. verify
reproduces the original exactly: 2 errors across 301 proposals, exit 1.
The binary has more verbs than its wrapper documented. The bash usage text
listed four; atlas-commit-and-sync calls four more it never mentioned. All
eight are declared so reach atlas --help is a complete index, and unknown verbs
are still forwarded — a hand-maintained list falls behind the binary it
describes, so rejecting on it would break the day someone adds a subcommand.
commit-and-sync now stages by default and commits only with --commit. Nothing
else in reach writes to git history, and committing as a side effect of "sync"
is a different risk class from writing a file; the default prints the message
it would use, leaving the decision where it was.
Two real bugs found in that script while porting it. It ran atlas-verify and
never checked the exit code, so a proposal that FAILED verification was still
wiped, committed and synced — bad data in systems.db is far harder to undo than
a failed command, and it now refuses. And it hardcoded a pinned
"Co-Authored-By: Claude Opus 4.6" into every atlas commit, which the git-commit
skill names as the root cause of attribution drift.
update-field gains two guards the original lacked. Its field→table map lived
inside a bash heredoc string where nothing could check it, and an unknown field
produced an UPDATE against a table of None; it now names the nine accepted
fields. And it checks rowcount, so a system_id that does not exist is a failure
rather than a silent no-op reported as success.
The three Python scripts moved with the usual treatment — prints to console
events, argparse replaced by typed functions, __file__ roots to
config.repo_root(). No root bug this time: checked before moving rather than
after, three domains running.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
reach godot parse-sweep / cold-parse, reach visual diff / blank-check /
thumbnail. Five scripts retired, and the callers rewired — tests/run-visual
invoked three of them by path at four sites, which is a wider blast radius than
the make targets were.
The godot pair were grep pipelines encoding five hard-won lessons as comments
nobody could test. They are Python filters now, with the reasons attached, and
the engine invocation is a guarded exec. Verified on the real client: 229
scripts, clean.
Their three not-ok states stay distinct, because only one is a verdict about
the code. An engine that crashed or is missing is not a parse failure —
reporting it as one blames the tree for a broken toolchain. A sweep that
emitted no completion marker checked nothing, and zero errors from a check that
never ran reads as clean, which is the false-green the sweep exists to close.
The deliberate asymmetry between the two checks is preserved and documented:
cold-parse filters "Cannot infer the type", the sweep does not, because that
suppression is why cold-parse stayed silent about a helper that genuinely does
not parse.
All three visual scripts carried the same root bug as validate-checklist:
Path(__file__).parent.parent, correct at tooling/ and two levels too deep at
tooling/domains/visual. Fixed during the move rather than after, having learned
that it fails silently — paths resolve to nothing, the work appears to have
nothing to do, and the tool reports success. Three domains now where that would
have shipped a false pass.
Two bugs my own transformation introduced, both found by running rather than
reading. Multi-line print(..., file=sys.stderr) became console.event(...,
file=sys.stderr), and console puts unknown kwargs into the payload — a file
object would have reached json.dumps at the exact moment something was already
being reported as an error. And the replacement script wrote escaped quotes
into three files. Mechanical transformations need mechanical verification.
sys.exit removed from four sites: a service must not end the process.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
reach validate content / checklist / ron / name-collisions. The three old
scripts are retired, their make targets with them.
Print statements go through the logging sink rather than a collector. The
validators emit their findings as console events as they run, so a long content
validation streams instead of going quiet and dumping at the end — the message
strings and their order are unchanged, only the destination. That also
satisfies the conformance rule forbidding print() in the package, which is what
forced the question.
validate-ron was three languages deep: bash dispatching on a flag, a Python
heredoc doing collision detection, cargo run for schema validation. Logic
embedded in a shell string cannot be imported, tested, or found by anything
that indexes Python, so it became Python; the cargo call became a guarded exec.
It also split into two verbs, because --check-name-collisions answered a
different question from the default path: whether the SET of cultures is
coherent, versus whether ONE file is well-formed.
The move broke something, quietly, which is the point of doing these one at a
time. validate-checklist computed ROOT as Path(__file__).parent.parent — the
repo root while it lived at tooling/validate-checklist, and tooling/domains
once moved. Both its schema and gauntlet paths silently repointed at nothing,
the gauntlet directory "did not exist", and it reported success having checked
zero files. Caught by running it beside the original: old exit 1, new exit 0.
Now config.repo_root(), and load_schema raises ReachError instead of calling
sys.exit, which a service must not do.
Parity on the live tree: content reproduces the original byte for byte
including its counts, name-collisions likewise. Tests pin what those runs
cannot reach — the detection path, since the repo currently has no collisions,
and the argument errors.
Two things found and left alone: validate-content FAILS on the live tree with
13 missing schemas, pre-existing and unrelated to this port; and the ticket's
claim that validate-content sits in the pre-commit hook is wrong — that hook
runs only check-fact-ids and pql decisions validate, so there was no shared
edit to coordinate.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Bash(reach) and Bash(reach *) join .claude/settings.json beside the pql pair.
Two entries, not the one the ticket asked for: a rule ending in " *" does not
match the bare word, and bare `reach` is a real invocation now that it prints
the domain list. pql, make, cargo test and ruff check each carry a bare-form
entry alongside the wildcard for exactly this reason, and adding only the
wildcard would have left `reach` prompting while `reach check ...` did not.
This is the line Q-124 was actually filed about. Ten hand-written
Bash(tooling/...) entries each cover a single script and every unlisted tool
prompts; one command with subcommands is one rule covering everything. The ten
stay for now — the old scripts are still the working tools until T-1253.
On verification, since the ticket warned specifically against declaring this
done on the wrong evidence: real calls run clean, but that is NOT proof the
rule matched. The same calls succeeded before the rule existed — there was no
Bash(reach ...) entry in either settings file and no blanket grant — so the
session was already permitting them and the observation cannot distinguish "the
rule matched" from "the rule was never consulted". settings.json is read at
session start, so this cannot be self-verified from the session that wrote it.
Proof is a later session, in a prompting mode, where reach runs without asking.
One accepted limitation, documented rather than worked around: rules
prefix-match the whole command string, so an env-prefixed call like
SR_REPO_ROOT=... reach ... will still prompt. An environment override is a real
departure from normal invocation; the ordinary form is what needs to be
frictionless.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The relationship between wiki/, the generators, systems.db and the runtime is
a directed graph with two edges running opposite to the obvious direction and
one running backwards into its own producer. Prose renders that badly: every
document that has described it states a single ownership direction and is
therefore wrong about part of the tree. D-262 makes the diagram the source of
truth and points CLAUDE.md, Skill(wiki), project-structure.md and
wiki/GOVERNANCE.md at it.
The correction that matters most: body pages were described everywhere as
machine-owned and reverted on sync. They are not. scaffold_bodies.py writes
one once and never overwrites it, and import_economics then reads that
frontmatter directly as input — so a hand-edit is not reverted, it is obeyed,
and silently changes world generation. Worse than being overwritten, and the
actual reason GOVERNANCE.md forbids the edit.
New: tooling/check-dataflow-graph.py, wired into the Makefile and the pre-push
hook. It asserts every repo path named in a hand-authored diagram still
resolves — and its docstring states plainly what it cannot do: verify that an
edge still MEANS what it says. If wiki_sync.py stopped writing body pages
tomorrow, every path would still exist and the check would still pass. Edge
semantics stay a human check against the tool's source, so nobody reads a green
gate as a verified map.
Verified by breaking it: pointing one label at a moved path fails with exit 1
naming that path; restoring it passes. Building the checker also caught two
real vaguenesses in the diagram — "GJ-*/index.md" and "bodies/{id}/index.md"
were written without their wiki/star-systems/ prefix, which is precisely the
ambiguity this map exists to remove. Generated star-map .d2 files are excluded
by name; their correctness belongs to their generator under D-223.
Also files Q-124 + T-1246 (tooling): whether the 123 Python files under
tooling/ should become one Rust CLI of pql's calibre. The friction is real and
mostly not about the language — the permission gate prefix-matches whole
command strings and a blanket Bash(python3 *) grant is forbidden, so each tool
prompts near-individually, while a single binary is one allowlist entry. The
record requires pricing the cheap alternative (a Python dispatcher entrypoint)
before recommending Rust, and flags the hard constraint: import_economics is
stamped by source SHA, so any port must keep that contract intact through the
transition rather than disabled during it.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
d2 emits SVG natively; its PNG path wants a ~150 MB headless-Chromium
download and prompts interactively, so every PNG here was produced by an
out-of-band magick step. There is no Chromium on this system. Dropping PNG
removes the dependency rather than trading one format for another, and cuts
docs/diagrams/ from 17 MB to 3.3 MB. SVG renders in Gitea and in clide
(`clide draw --file <path>`, which takes .d2 source directly), and diffs as
text.
One PNG is kept on purpose: design/star-map-concentric.png has no .d2 source.
Also renders the 7 star-map .d2 files for the first time. star-map-plan.md
listed their renders as a deliverable in March and the step never ran; the
new `make check-diagrams` is what surfaced it.
New: docs/diagrams/data-flow/wiki-generator-flow.d2 — which way the arrows
point for any file under wiki/. Every edge was read in the tool's own source
rather than inferred. It records the trap that keeps costing us: scaffold_bodies.py
writes a body page once and never overwrites it, and the generator then reads
that frontmatter directly — so a hand-edit there is not reverted, it is obeyed,
and silently changes world generation.
Two rendering traps found the expensive way and now written down:
- A d2 `|md` block becomes an SVG <foreignObject>. ImageMagick and flutter_svg
both silently drop it, so the legend was in the file and invisible in every
viewer except a browser. Plain labels render as real <text> everywhere.
- Container boxes fight the layout engine. Grouping nodes whose flow-depths
differ forces long edge routes; this diagram went from an unreadable 2.4:1
sprawl to a legible 0.75:1 by deleting five containers and changing nothing
else. Colour classes carry the grouping instead.
make diagrams / make check-diagrams render and gate. Repo-specific rules in
.claude/rules/diagrams.md; d2 syntax and the traps live in the user-scope
d2-diagram skill, whose PNG default was flipped to SVG to match.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Jeroen asked whether I had read the python that writes the frontmatter. I had not — only grepped it. Reading body_definition_parser.py and scaffold_bodies.py properly overturned what I had written twice today.
scaffold_bodies.py NEVER OVERWRITES ('Only creates files that don't exist yet... existing body index.md files are skipped'), and the generator reads that frontmatter directly. So a hand-edited body page is not reverted, it is OBEYED, and it silently changes world generation — worse than being overwritten, and the actual reason GOVERNANCE.md forbids it. System pages behave the opposite way: wiki_sync.py re-renders their READ-ONLY blocks, so edits there ARE reverted. Three cases, not two.
It also explains T-1244's whole measurement: body_definition_parser resolves each field override > direct read > derived > inferred > SEEDED RANDOM. Continuous axes vary because they fall to the random tier; categorical axes are concentrated because they are read from the bodies table. The variance question belongs to the atlas CLI catalog, not the wiki.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The cold test worked as an experiment: a fresh agent with Skill(wiki) cited it first, refused to hand-edit body frontmatter, knew the corp regen-db stamp trap, and knew corp_specialization is missing from its own template. It also found four things the skill had wrong or missing, all verified before folding in: body frontmatter is a MIDDLE layer (atlas CLI -> systems.db -> scaffold writes the page -> import_economics reads it back), not the origin GOVERNANCE.md implies; the four empty categories are Q-118, an open scope question rather than an invitation; some bodies are visual-regression goldens and nothing in wiki/ says so; and status is editorial, not an import gate. Also: check current state before editing, since the test's own task described a change that was already true.
T-1244 corrected in the same pass — tectonics is derived from planet_class via a lookup (body_definition_parser.py:563), so the measured 68% 'low' is a projection of the class distribution, not an authoring choice. The ticket's question changed accordingly.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
An experiment, at Jeroen's request: predict how the wiki seed data is structured
WITHOUT reading it, seal the prediction, then score it. The prediction is
a1addf7e2, committed before wiki/ was opened so it could not be retrofitted.
The score, against a rule fixed in advance:
RIGHT — markdown + YAML frontmatter, TOML for economics tables, the body path
shape, more trees than the three I had seen.
WRONG — "a source, never an output". That holds for 253 pages and is backwards
for 3,262: star-systems/ is GENERATED from systems.db by tooling/db/wiki_sync.py,
its <!-- READ-ONLY --> blocks are renders, and body frontmatter IS the body
definition rather than a description of one. Also wrong: "probably no schema
docs" — there are 17 templates, ten authoring guides, economics/schema.md and a
GOVERNANCE.md that states the ownership models plainly.
ABSENT (the expensive bucket) — the two ownership models running in OPPOSITE
directions; the GTTR prose channel; terrain.npz/globe.png; that stations and
districts have NO wiki directories; that `description` frontmatter exists so
agents can filter before loading; and the scale, 11,864 files.
ROOT CAUSE, and it is not missing documentation. The wiki documents itself well.
It was unreachable: wiki/ appears in NEITHER CLAUDE.md's Project Structure block
NOR .claude/rules/project-structure.md, the annotated tree whose entire job is
orienting an agent. The largest tree in the repo — the seed for the whole Reach —
was invisible from both files a session reads first. Every item in the absent
bucket follows from that one omission. The proof is this session: it spent three
days fixing Ferrath's terrain rendering and never once saw
wiki/star-systems/GJ-820B/bodies/GJ820Bc/index.md, the file that defines Ferrath.
Fixed here: wiki/ enters both structure documents with the ownership split stated
where it will be read, and Skill(wiki) carries the traps — never hand-edit a
READ-ONLY block or body frontmatter, stations have no directories, the id is
spelled two ways, editing corp PROSE stales systems.db, and absent variance is
often deliberate rather than a gap.
That last point cost two false findings in one measurement and is worth the
warning: chemosynthetic:false on every body is a namespace reservation for
dextro-DNA-style biochemistry once geology and nature spawn to the 1x1m pixel,
and enabled:false on ~65% is staged rollout — clean planet types first, generator
scripts for the rest after. Both read as defects without the roadmap.
Also measured, since the seed's job is to supply variance: continuous axes are
rich (unique seed per body, 460-716 distinct values across orbit/tilt/ice/land)
while the categoricals that gate morphology are concentrated (68% tectonics low,
51% planet_class frozen). Filed as T-1244 with the design question stated first —
whether the distribution is intended — rather than as a defect.
Method caveat recorded in the findings: the aggregator reads scalar frontmatter
only, and atmosphere_color's "100% null" was a parser artefact, not a finding.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
A Write(<path>) permission rule matches nothing. File permission checks
consult only Edit(<path>) rules, which already cover every file-editing
tool — Write, Edit and NotebookEdit alike. Claude Code now warns about
the dead shape at session start.
All eight removed here sat directly beside their Edit() twin, so the
allow grant over the repo tree and the ask gates guarding settings and
hook files kept working throughout. Behaviour is unchanged.
That ask block remains the pattern worth copying to the other repos in
this tree — it is the only one that stops an agent quietly widening its
own permissions, and it has to be ask rather than deny to stay fixable.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Placed in the pre-push hook rather than /pr-process, because the hook is where
enforcement actually lives — and notably the hook never ran godot-cold-parse
at all, so until now nothing enforced "does this script parse" for any file
outside the startup path.
Ordered BEFORE the test suite deliberately. That makes failures cheaper rather
than the gate slower: a script that does not parse is caught in ~4s instead of
after ~135s of tests that could never have covered it. A clean push pays 3.7s;
a broken one saves over two minutes.
Not redundant with the suite. gdUnit4 reports the suites that DID load as a
clean pass, so an unparseable file reads as success — guarded now in
tests/run-godot, but only for test files. The sweep covers all 226 scripts,
including the roughly half of the codebase no test ever loads.
/pr-process gains a scope note instead of a second invocation: cold-parse sees
only the startup path and filters "Cannot infer the type" (which hid a
genuinely broken file for five months), so it must not be read as a general
parse check. Per team-patterns.md the skill does not duplicate the gate.
Hooks run from .config/hooks via core.hooksPath, so this is live without an
install step.
Pair session with Jeroen, 2026-07-27.
Co-Authored-By: Claude <noreply@anthropic.com>
table_baroque, chair_modernist, desk_scifi (furniture, with mask
sidecars) and lion_statue (props, non-tintable) enter
client/assets/models/ under the category-first convention, byte-
identical to the spike sources (MD5-verified; spikes/ untouched,
vw_beetle excluded per the Q-067 deferral). Postprocess verified
already-applied by Blender node-graph inspection (mat_primary,
roughness=1.0/specular=0.0 shader inputs — the convenience properties
read stale defaults, the graph is authoritative), so not re-run. All
four registered in manifest.json at 'planned' with footprint_tiles
[1,1] (a flagged judgment call — normalized spike geometry erases
real-world scale; the brief ruling goes to araminta in review).
In-engine proof: clean headless import + ResourceLoader load +
instantiate for all four. Discovery documented in conventions.md §2 +
glb-gen SKILL.md: Godot externalizes the GLB's embedded texture as a
<model>_Image_0.png sidecar referenced by binary UID — it must be
committed or loading breaks with an undiscoverable dependency error.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Instantiates the five D-244 catalog files (models/textures/artwork/icons/
effects — icons carries the 13 as-built SVGs, stance icons traced to their
consumer and marked final), creates client/assets/models/{furniture,props}
with a schema-documented manifest.json mirroring the character manifest,
fixes glb-gen SKILL.md's stale 'path does not exist' paragraph, and authors
docs/assets/visual/conventions.md: category-first model naming (araminta
ruling), mask sidecars per the character convention, D-235-token-keyed
texture naming, footprint_tiles as manifest metadata. Door-state and
TileSet conventions are recorded as explicit deferrals (Phase-5 owned).
visual/README.md carries both this ticket's count/link updates and
T-1052's palette-section rewrite (shared file, committed here).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The interactive panel is no longer hand-composed per render: support/
panel-template.html carries the widget markup (CDS tokens, pick-up/detail
sendPrompt contract) and support/make-panel fills it with live pql data —
actionable WIP + unblocked leaf work from the active phase AND the
maintenance initiative T-1037 (the unblocked head T-1174 lives there, not
under the phase epic — a phase-only query missed it), sorted
status→priority→recency with overflow noted. SKILL.md §Interactive panel
now instructs: run the script, paste stdout as widget_code verbatim.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Second surface for the skill: an in-chat widget panel (WIP + unblocked
ready work, capped ~10 rows, native pql projections) with Pick up ↗ /
Detail ↗ per row and Refresh board / Next batch footer actions. Defines
the pick-up prompt contract ('pick up the following ticket: T-NNN — title'
→ single-ticket batch activation per the /whats-next step-3 flow). Records
the verified platform boundary: widget read-back is unsupported — panels
talk to Claude via sendPrompt only; the artifact stays reference-only.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
tooling/pql-board-html: self-contained interactive board snapshot from
pql-native JSON (ticket list --full + batched --with-blockers for non-terminal
tickets + plan status) — status-grouped rail with filter/type chips, ticket
dossier with status pills, dep/children chips, deep links, keyboard nav;
clide's visual identity (amber on warm near-black, mono data type). The
/pql-board skill regenerates and redeploys to the canonical artifact URL so
the user's open tab survives refreshes across sessions.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Jeroen confirmed the .internal-to-.net proxy migration was intentional
(2026-07 weekend maintenance): git.schweitz.internal's vhost is gone,
git.schweitz.net is live with a LE cert and AdGuard LAN hairpin.
tea-cli.md + local-services.md repointed (tea's own config already
switched). tooling/db/config.json: the bare tower-of-joy hostname has
no DNS entry since the migration — Stable Audio/Trellis endpoints now
by IP. Workshop archives under docs/workshops/ keep their historical
.internal mentions (records, not operative config).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
An explicit --repo slug makes tea skip local-repo setup, but pr create unconditionally needs the local repo handle and dies with 'local repository required' — the documented 'all flags explicit' rule was the trap. Root-caused shipping PR #175. tea-cli.md + pr-process template fixed.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Cleared stale STANDBY markers, swept dead DISCUSSION.md pointers, disambiguated overlapping personas, corrected asset-gen paths. Armed 7 discussion agents (gestalt/gore/mellanie/miri/nigel/ozzie/paula) with Bash so they can run the pql preamble. Integrated inigo into the roster + briefing (audio, standby). Removed tiger — localization dropped from scope (R-013). Part of T-1099.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Live-command corrections (atlas corridor-status, real body-ID naming), failure-proofed glb-gen/sprite-gen render scripts, Trellis API reference extracted. image-gen: fixed the output-path bug and de-forked the local image_connector.py to the canonical tooling/db/ copy. ticket skill consolidated to point at ticket-cli.md (setparent-none fix applied there too). Part of T-1099.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
clide SKILL.md re-synced to the clide repo's canonical copy (picks up the dock/D-87 slot; de-forks). d2-diagram/frame0-wireframe/skill-create deletions landed in the prior commit; their repo-specific tailoring is preserved in docs/design/diagram-wireframe-conventions.md. Part of T-1099.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
tooling/worktree-setup <branch>: one command to create a usable worktree —
adds it under the gitignored .worktrees/, symlinks .venv (so make/python
tooling resolves .venv/bin/python), relies on the post-checkout hook for the
pql --vault rebuild, and prints the in-worktree reminders. Tested end-to-end
(venv linked, pql.db populated on create).
whats-next §3c now calls the helper and documents the three in-worktree
gotchas (pql --vault, tea-from-main, read-only content agents); tea-cli.md
notes tea must run from the main checkout.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
ticket list now omits description by default (D-27 upstream); batch
selection uses --fields id,type,parent_id,priority,status,title — the
238KB phase-subtree dump is now 23KB, no output spill.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Ad-hoc projections over pql output are the wrong layer — the compact-list
gap is filed upstream (pql FR-1, --fields projection). Until it lands,
use narrow pql invocations and Read spilled output files.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The parent-dir pattern put worktrees outside the permission sandbox,
breaking agent file access. .gitignore already documented the in-repo
convention; the skill predated it.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The mandatory Commit Message Format block hard-coded 'Claude Opus 4.6', which stayed stale for months after the model moved on. Replace with the generic 'Co-Authored-By: Claude <noreply@anthropic.com>' and note that the harness-appended model-identifier variant is equally valid. Pinned versions are the root cause of this drift class (CLAUDE-CLEANUP.md section 4).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
settled-reach is always a pql repo, so make the rule repo-canonical (CLAUDE.md +
ticket-cli.md): routine ticket/decision CLI is covered by the rules, but the
load-bearing model lives only in the pql skill — changelog is write-through (never
git-restore it), `export` doesn't replicate ticket_idmap, post-checkout rebuilds
pql.db from the changelog, so cross-branch planning moves must re-create via
`pql ticket`, not changelog surgery. Back up pql.db before risky ops.
The clide-side equivalent (read the clide skill when running inside clide) is an
editor choice, not a repo property — it belongs in a clide-owned SessionStart hook,
not here.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
D-244 ratifies what the character architecture (D-159..D-164, runtime
CharacterVisualDescriptor compositing) and the Trellis env pipeline already implied
but no decision had recorded: the in-world view renders 3D objects directly; the only
flattened 2D content is textures + flat 2D artwork (paintings/flags/billboards/signage).
There is no per-object sprite layer.
Roots out the drift: the early-spike 3D->2D sprite pipeline (#541) and the sprite-centric
docs/assets/visual catalog were never cleaned out when the project went 3D, so the
2026-06-12 fable-ous audit read them as live and re-injected the dead sprite model into
T-1049/T-1050. Fix:
- docs/assets/visual/README.md + docs/assets/README.md re-scoped to 3D models + textures
+ flat artwork (dropped the sprites/tilesets-as-entities framing).
- /sprite-gen + renderer/README repurposed as the 2D-artwork generator (paintings/flags/
billboards/signage), not the in-world object format; legacy 4-direction object-sprite
mode flagged as retired.
- T-961/T-1049/T-1050 already held in backlog pending this.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
import_economics.py 2,620 → 309 lines — a thin orchestrator keeping the
exact CLI, single-transaction/rollback contract, and exit codes. The 16
import steps, MIGRATION_SQL, brands shell-out, validators, and stamp
write now live in tooling/economy-db/economy_import/ (db, migration,
economy, corporations, brands, bodies, atlas, specialization, traits,
validators, stamp, paths, errors). Full type hints throughout.
tooling/generator_sources.py replaces the triplicated source registry
(importer / stamp checker / pr-process watch list — the skill now derives
its list via --list). The registry stamps itself, and economy_import/
modules are globbed fail-closed, so a future module is stamped the moment
it exists — closing the silently-weakened-stamp failure mode.
Rider: connector config helpers centralized in tooling/db/common.py.
Byte-identical behavior proven: full-import table dump diff EMPTY over
107,843 lines / 37 tables (volatile timestamp fields excluded); dry-run
output parity; generated_brands.toml sha unchanged. make test-tooling
PASS; ruff clean.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
- make test-tooling: planet-gen determinism guard + import_economics
--dry-run, wired into pre-push on TOOLING_CHANGED; ruff widened to
E4/E7/E9/F/W (90 safe auto-fixes applied; E402/E702/F841 ignored with
documented counts)
- one-generator reality fixed in DEVOPS.md, asset-pipeline rule, CLAUDE.md
(import_economics sole generator since #951/D-223); dead check-protocol
target deleted; DEVOPS hook/config sections rewritten from the actual
hook sources; team-patterns gate description updated (client+tooling)
- project.yaml: 0.2.0 → 0.4.0 per the 0.{phase}.{n} scheme, description
refreshed from the v0.1 Sova narration to cascade reality
- stale comment sweep: voxel.rs stub claims (all 8 families implemented),
cascade.rs TODO recited to T-1044, main.rs D-192 handshake claim,
relationships.rs/chunk_streaming.rs version targets → phase language
- gitignore: client/settings.db* e2e-run artifacts
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
- workers/pool.rs: 5 new tests; catch_unwind keeps worker threads alive on
handler panic (in-flight request loss unchanged, pinned by test + #843
docs); stubs.rs no longer falsely claims the pool is tested
- save/load: execute_save_load pinned .after(Storyteller) so the scheduler
cannot legally save pre-Input state; exclusive-system exception recorded
in tick_phases.rs rules
- surname corpus extracted to bin/shared/surname_corpus.rs (both economy
generators import it; byte-identical output verified on 23.6MB+1.45MB
TOMLs); all three stamp/watch registries updated
- generator_spike gated behind non-default 'generator-spike' feature
- economy.rs: 11 new D-181 signal-derivation tests on the new
econ_sim Simulation::from_economy in-memory constructor
- perception exemption comments now state the consumer sort contract;
unused bytemuck removed; rayon comment corrected; the 22 allow(dead_code)
documented as serde schema enforcement
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Add step 1a: check open PRs first and recommend /pr-review (review -> merge)
before selecting a batch. Open PRs are in-flight WIP; starting new work on top
piles up the board, and the cascade-correct next batch is often dependency-gated
on an in-review PR (won't show as --unblocked, or must stack on an unmerged
branch). Renumber the existing 1a-1d substeps to 1b-1e.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Add `cargo test --quiet` to the pre-push Rust block (sharing clippy's
target/ guard) so the suite runs automatically on every push touching
server/. Nothing ran the tests before — pre-push did only fmt/clippy/deny
and there is no CI — so a Rust regression could reach main unverified.
Document the resulting agent-spawn rule in team-patterns.md: don't have
implementation agents pre-run fmt/clippy/test pre-emptively, since the push
gate now enforces all three. The gate is authoritative; the lead patches any
fallout at push.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The repo moved to /var/mnt/data/projects/settled-reach (flat layout,
no more /main worktree subdir). Drop the stale autoMemoryDirectory
override (pointed at the old /home path; revert to the native
per-slug memory store) and gitignore .worktrees/, which now holds
per-machine git worktrees under the repo root.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The pql cutover is stable, so remove the superseded SQLite planning tooling. Surgical
— only the ticket/decision/raw-SQL scripts (all settledreach.db-bound and replaced by
pql) are deleted; the asset/audio/wiki connectors and shared common.py stay.
Removed:
- tooling/db/{ticket,decision,decisions-sync,decisions_sync.py,sqlite-query,sqlite-exec,
sqlite-init,sqlite-seed,sqlite_connector.py}
- tooling/{db-backup,db-install} + docs/backups/settledreach.db.backup (the binary-DB
backup ritual; tickets now live in the git-tracked .pql/changelog/)
- tooling/check-decision-ids (dead stub, superseded by `pql decisions validate`)
- Makefile db-backup/db-install targets; SR_DB_PATH + tooling/db/{ticket,sqlite-*,
decision*} entries from .claude/settings.json (audio entries kept)
Updated docs to pql: DEVOPS.md (SQLite Access + Decisions System → pql), project
structure, ticket-cli closing note, asset-pipeline raw-SQL warning.
Kept (verified still imported by the asset connectors via common.ensure_venv): common.py,
config.json, audio/image/trellis/wiki connectors. The live settledreach.db file
(gitignored, repo-parent) is left on disk as a cold rollback only.
ruff clean; pql decisions validate ok (357 decisions / 1013 tickets).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Replaces the labels-as-milestones pattern (carried over from the retired SQLite
milestone subsystem) with pql's native initiative/epic hierarchy, which already
modelled this: initiative T-745 (Development Cascade) holds the six phases as epics.
A ticket is in a phase by being parented under that phase epic — self-maintaining,
no label to apply or forget.
The `phase:4` label had already drifted (66 tickets under the Phase-4 epic T-750 but
only 42 labelled). Fixes:
- T-750 (Phase 4) -> in_progress to mark the active phase (sequential per D-166).
- Re-parented the 2 strays (T-974 Atlas-to-tile epic, T-1008 economy task) under T-750.
- Dropped all 42 phase:4 labels (tombstoned in the changelog).
- /whats-next, /ticket, ticket-cli.md, CLAUDE.md now gate on
`pql ticket list --under <active-phase-epic> --unblocked` instead of `--label phase:4`.
- Fixed the label-action verb in the docs: pql uses `add|rm`, not `add|remove`.
- pql-requirements #9 flipped: no milestone entity / labels-as-milestones needed — the
initiative/epic tree is the answer.
Verified: `pql plan rebuild` reconstructs the mutations; `--under T-750 --unblocked`
returns the active phase's ready work.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The session now runs inside the clide IDE, which exposes its UI surface as a
`clide <subsystem> <verb>` CLI (observe + drive parity). Skill copied from the
clide project; documents `clide capabilities` as the authoritative discovery
command and the observe/drive verbs (editor/files/git/search/pane/ui/pql).
clide has the pql planning store wired in (Tickets + Decisions sidebar panels),
so the just-merged migration powers the IDE's planning UI.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Completes the Phase 4 consumer cutover: adds Bash(pql) / Bash(pql *) to
.claude/settings.json so the pql workflow runs without prompts. Added manually by
the lead — the auto-mode classifier blocks an agent from self-widening wildcard Bash
permissions even via /update-config (a deliberate security boundary).
SR_DB_PATH and the Bash(tooling/db/*) allowlist stay until Phase 6 (rollback path).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Repoints the kanban workflow skills from tooling/db/* to the pql CLI:
- /whats-next: milestone queries -> `pql ticket list --label phase:4 --unblocked`
(--unblocked does the dependency walk, collapsing the old milestone + per-ticket
deps + milestone_deps steps); WIP via `pql ticket list --status in_progress` (pql
doesn't enforce the limit — warn manually); refinement append via `pql ticket append`;
activation via `pql ticket status`.
- /ticket: full reference rewritten to `pql ticket`/`pql plan`; milestone subcommands
-> phase labels; points at .claude/rules/ticket-cli.md.
- /pr-process: step 8 extracts T-NNN (maps bare #NNN -> T-NNN) and calls
`pql ticket status T-<id> review`.
- /bug-report: bug creation via `pql ticket new bug`.
- clerk: decisions-sync mechanism line -> `pql decisions sync` (factual fix only; the
agent's review logic stays for the #965 re-enable).
- workshop-start IMPROVEMENTS: the old `decision claim` non-idempotency footgun no
longer applies — `pql decisions claim` is side-effect-free.
pr-review/git-commit needed only the Phase-1 path repoint (no command refs); the
settledreach.db.backup exclusion in pr-review stays until Phase 6.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
