run-visual verified only that the captured PNG was non-empty AS A FILE. A
blank screen is a perfectly valid ~19 KB PNG, so it passed — and once a blank
capture had been recorded as a golden, every later blank capture matched it at
0.0% and the scenario PASSED. atlas_GJ338Bd_Block and atlas_GJ445c-m1_Chunk
sat green against blank goldens while the suite's other 30 scenarios failed.
That is the worst kind of test result: indistinguishable from success, and
load-bearing for exactly the work it fails to cover. e024cfb3f recorded this
same failure once already ("the Atlas Global goldens have been measuring
nothing"); it recurred because nothing checked the property, only the file.
tooling/visual-blank-check measures the share of the frame taken by its single
most common colour. On this project's real captures the classes are far apart:
Global (real world map) 38.7% modal
Region (flat colour wash) 7.2% modal <- dither; least uniform of all
District 45.4% modal
Block / Chunk / Quarter 92.9-94.6% modal <- nothing drawn
Nothing falls between 45% and 93%, so the 0.85 default sits in open space
rather than being tuned against a boundary case. Deliberately NOT an aesthetic
judgement: the Region wash is a real product gap (T-1213) and scores 7.2%,
comfortably "content". The question is only whether a world reached the
screen.
Wired into both paths, and the update path is the one that matters — refusing
to RECORD a blank golden is what stops the trap being re-armed. Ad-hoc
--screenshot only warns, since capturing a rung that renders nothing is a
legitimate thing to want to do; that is how the empty deep rungs were found.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Two faults, one fix.
Captures were stealing the desktop. Only the golden path even tried to go
offscreen, via xvfb-run -- which is not installed here, so it took the
"using visible window" fallback; --screenshot and --movie never wrapped at
all. Every capture opened a Godot window on the machine Jeroen is working
and gaming on. Now a single wrapper covers all three paths.
gamescope, not the alternatives, for two independent reasons. It renders on
the real AMD GPU, and the goldens are pinned to this box's Mesa/AMD output
(T-1121 -- they do not port across rendering stacks), so xvfb-run's llvmpipe
would shift every pixel: offscreen must not silently mean a different
renderer. And it is the only installed option that lets the output size be
set. cage is also present and also GPU-backed, but it is a kiosk compositor
and forces its client to the headless output's default -- measured, a
960x540 request produced a 1280x720 PNG. A wrapper that quietly changes
resolution is worse than none here.
Worse, because resolution is not cosmetic on this map. D-255's extent
inversion makes the shorter viewport axis span exactly one cell of the rung,
so the viewport decides how much world a rung shows and at what cell count
-- a small capture is a DIFFERENT map, not a scaled one. 960x540 was also
16:9, so it never exercised the ultrawide aspect added in 21e263d0a, which
is the aspect actually in daily use. Raised to the panel's native 3440x1440.
It paid for itself immediately: at native, Ferrath Global reports
courses=375 drawn=0 -- every river culled, where the same build drew them at
960x540. Filed as T-1239. That is exactly the class of bug a too-small
capture hides.
Goldens are NOT regenerated here. They are stale across 15 commits already,
and blessing the current look before it has been reviewed is the trap this
suite just spent a day proving.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The visual gate had stopped measuring anything: 30 of 32 scenarios
failed, and the two that passed were the worst result of the lot.
Deleted the 15 top-down scenarios (fog, HUD, dialogue, NPC, minimap,
cursor) and their goldens. They all failed at a near-uniform ~12%, and
that uniformity across unrelated scenes is one global cause -- the
ultrawide UI stretch moved every element. They cover the renderer the
cascade freezes until Phase 5, which will need its own tests anyway, so
re-baselining would only have blessed a deprecated layer nobody is
reviewing. Jeroen's call.
The remaining problem was the goldens that PASSED. atlas_GJ338Bd_Block
and atlas_GJ445c-m1_Chunk matched at 0.0% because capture and golden
were both blank -- the same "goldens have been measuring nothing" trap
e024cfb3f caught at Global, still live at the bottom of the ladder. The
cause is that every below-Global golden descends at jump_to(ZERO), and
world-metre zero is merely the origin of the region grid, not anywhere
chosen. So _setup_atlas_golden_shot now takes an optional world_center
(default ZERO -- existing goldens are untouched), and a new
atlas_GJ820Bc_land_* set walks Region through Chunk at ONE land point,
so the rungs can be read as a descent instead of five unrelated frames.
aliveness_probe prints the placement's world metres alongside its pixel
and survey cell, since that is the coordinate the Atlas actually
navigates in.
Recorded because it will be asked again: the ladder is anchored via a
CityPlacement, but that is a match record -- a pixel, an archetype, an
orientation -- not built geography. No settlement exists anywhere yet
and none is due before T-1207, so the empty deep rungs are the expected
state. What the ladder judges is the nature layer.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Two parsing bugs in the summary, found while measuring suite times.
DOUBLE COUNT. gdUnit4 prints one "Statistics:" line per suite and then a
single "Overall Summary:" line whose numbers are the sum of all of them. The
pattern matched both shapes and summed all 87 lines, so every total was
exactly twice the truth: a full run reported 3,660 tests against an actual
1,830, and a 26-test suite reported 52. It was invisible because it doubled
UNIFORMLY — nothing ever looked inconsistent, only large. Every count quoted
from this harness, in this session and before it, was 2x.
Now prefers the Overall Summary, which is gdUnit4's own arithmetic over the
whole run and so cannot disagree with itself; per-suite summing survives only
as a fallback for a run that dies before printing it.
ANSI. gdUnit4 colourises output and the escape sequences sit BETWEEN the
fields of the summary line, so patterns matching the raw log silently fell
through to the weaker "Executed test cases" fallback — which cannot see skips
and reported a fully skipped suite as 26 FAILED. All parsing now runs against
a de-ANSI'd copy, including the load-error guards.
SKIPS are now parsed and surfaced as their own JSON field, and excluded from
passed. Counting a skipped test as passing is the same false-green shape the
harness guards exist to prevent, and it stops being hypothetical the moment a
suite is deliberately skipped.
Verified against a fully-skipped suite (26 total / 0 passed / 0 failed / 26
skipped, was 26 FAILED) and a full run (1,830 total / 1,804 passed / 0 failed
/ 26 skipped, was 3,660/3,660).
Pair session with Jeroen, 2026-07-27.
Co-Authored-By: Claude <noreply@anthropic.com>
Found by walking into it. test_step_canvas_annotation_layer.gd had a parse
error from an earlier edit in this session, so gdUnit4 could not load it and
ran the other suites instead. The harness printed 3610 passed / 0 failed and
exit 0. Fifty tests had not run for hours and nothing said so — the full
suite reports 3660 with the file repaired, and that difference was invisible.
Two states are now hard harness failures rather than test results:
load_error — a suite failed to LOAD. Any pass count excludes it, so a green
number is a lie. The hint names the offending file.
no_tests — zero tests executed. A run that executes nothing can never be
a pass; previously a mistyped --filter printed "Tests passed".
Both add a "harness_error" field to the summary JSON and exit 2. The exit
code cannot inherit gdUnit4's, which returns 0 in both states — that is
precisely why they were invisible.
Verified by injecting each failure rather than by reasoning about it. The
load_error guard was checked in the case that actually matters: one broken
file among many, where total stays large and failed stays zero. That run now
reports 3610/0 WITH harness_error and exits 2, where before it was
indistinguishable from success.
Also repairs the file itself: a missed set_frame() argument (the parse error),
and a cell-placement test still asserting pre-inversion spacing. Rewritten to
assert the invariant that survives the extent inversion, the viewport aspect
ratio and panning — half the SHORT axis is half a rung cell — instead of a
literal. Two things it deliberately does not assert, both of which the
previous version got wrong: "the corner is half a district away" holds only
on a square canvas, and the canvas is one district WIDE without sitting ON a
district. It is a free-floating window centred wherever the player panned;
zoom is stepped, pan is continuous. A rung names a scale, not a cell you are
inside. A second test pins that with a deliberately unaligned world centre,
so a future change that snaps the canvas to the rung lattice — making pan
step instead of slide — fails here.
Pair session with Jeroen, 2026-07-27.
Co-Authored-By: Claude <noreply@anthropic.com>
Tyre finding 2: _log_atlas_view_transform's footprint_px now logs the
real on-screen pixel footprint (StepCanvasTransport.canvas_footprint_px
* canvas_scale — the same public pure function the terrain layer uses),
with the old cell-count field kept as canvas_cells. Live-verified the
divergence the fix exposes: Global logs footprint_px=(1528, 760) vs
canvas_cells=191x95 — the T-1192 fit-multiplier class the log exists to
root-cause, previously invisible under the mislabel.
Tyre finding 3: atlas_shots.json's one_shot_per_body note replaced by
rung_coverage stating the true distribution — 13 goldens across 7
bodies (per-rung coverage map, GJ380c's 3 shots as the invariance-proof
body, curated-subset rationale per T-1121). Corrects the prior commit
message's '12 new goldens' miscount: the true count everywhere is 13,
verified against shots array, visual.json entries, and PNGs on disk.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
_setup_atlas_golden_shot and _run_atlas_matrix rebuilt on the real D-255
surface: nav.push('regional') through AtlasApp's own body-selection tail,
then StepCanvasViewer.jump_to at a fixed center (inventory item 5) — no
shim over the retired continuous-zoom API. is_pending()-aware bounded
settle (mirroring atlas_agent_driver.gd) and view-transform logging
(inventory item 3) wired into both capture paths. The 12 z2_0/z4_0/zfit
goldens are replaced by 12 (body, rung) goldens captured live against
fresh --test-mode servers; atlas_shots.json/visual.json re-keyed;
atlas_gen_open's stale gen_l1_* overlay ids fixed to gen_dw_temp.
Verification: two consecutive District runs byte-identical, and a
cold-vs-warm disk-cache invariance proof on both terrain draw branches
(Global/NEAREST, District/LINEAR) — byte-identical either way, so
capture output does not depend on the shared user://atlas_cache state.
DEVOPS.md's real-rendering exception note now records the fold-target
mapping for the smoke file T-1182 already deleted (Global + District
goldens exercise its two real-pixel draw branches). Old legacy-tracked
.import sidecars go with their PNGs; new goldens ship bare per
.gitignore's client/**/*.import rule.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The _comment_atlas_goldens key inside scenarios{} hit _print_list's .get() on a bare String (reproduced: Invalid call on String at visual_capture.gd:699). Moved to a top-level key alongside resolution/golden_dir; _print_list now also skips _-prefixed and non-Dictionary entries so future comment keys cannot crash it. Verified: --list runs clean, JSON valid.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
AtlasViewer gains public set_view(zoom, offset)/get_view_offset (clamped, review-flagged API gap) + gdUnit suite. visual_capture.gd gains an atlas_matrix scenario (one boot, tests/atlas_shots.json-driven: 7 verified bodies x zoom x overlay sets, signal-gated waits on atlas_layers_received/city_names_received with timeout fallback). 12 individually-addressable golden scenarios registered in tests/visual.json. Root-level settings.db* gitignored (capture-spawned servers write their settings store to cwd). Fixes found live: GDScript lambdas capture value-type locals BY VALUE (wait-flag moved to Array carrier); RegionalScreen.enter() no-ops when the screen id is unchanged (direct enter() per body after the first).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
apply_move no longer writes Facing (was the only movement-facing coupling,
player-only); the player's view changes solely via explicit SetFacing.
NPC path-follow now sets Facing to the step direction — a strict improvement
recorded as a D-252 correction: NPCs previously never received Facing from
movement, their cones sat at spawn direction while walking. Bump-to-turn
retired (blocked moves change nothing) with a regression test for each
semantic. Wire schema unchanged; player_facing docs now say view/aim.
Gauntlet fixtures regenerated (facing octants now reflect view-only
semantics); client replay + live-roundtrip suites green against the new
server.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
New SR_LIVE sandbox scene: CharacterVisual composited in a 3D greybox world
derived from server snapshots. Per-leg constant-velocity interpolation keyed
to the stance throttle, 'server feet / client eyes' facing (wire octant while
moving, client aim octant idle), cadence-synced gait state machine on
AnimationPlayer custom blends, D-148 orthographic follow camera (-30deg
default, T-cycle presets), sim-space grid shader, camera-side wall cutaway,
accumulating never-evict tile store with four-state visibility tint.
Additive seams only: InputMapper.facing_angle_provider (2D path unchanged),
CharacterVisual.play_animation blend_time param + get_animation_player().
Visual harness gains per-scenario scene field + SR_AUTOPILOT input scripting.
210 new gdUnit assertions across five suites; verified live (230/230 total,
clean smoke, screenshot at .cache/screenshots/locomotion_idle_live.png).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
atlas_gen_open opens the real atlas app via HudGroups, navigates to GJ71c's
regional screen, and enables the Layer-1 overlays — exercising the production
opener + the live request→cascade→render pipeline against a --test-mode server
(SR_LIVE=1). Capture with: tests/run-visual --screenshot atlas_gen_open.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Ad-hoc capture scenario that overlays the ImplantPending "generating" indicator
on the booted scene for visual sign-off — `tests/run-visual --screenshot
implant_pending`. No golden committed; this is an inspection scenario, not a
regression gate.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The procedural server cascade (Phase 4) and the frozen names-only pool
supersede the Python atlas geometry generator and the LLM namer. Retire:
- generate_atlas.py (geometry production — cities/roads/rivers placement)
- gemma_naming.py, naming_core.py + tests (test_batch_naming,
test_register_selection, qa_naming) and run-atlas-naming.sh (the LLM
place-namer; its output is now the frozen pool)
- apply_name_fixes.py (name-field patches), fix_fewshot_bleed.py /
prune_atlas_features.py (geometry tools)
- import_city_names.py (redundant with import_economics name-pool path)
Pipeline updates: drop the generate_atlas step + atlas-generate /
test-atlas-determinism targets from the Makefile; remove generate_atlas
from the stamp registry (import_economics is the sole regen-db generator);
drop run-atlas-determinism from tests/run-all; refresh stale references in
schema_version, backfill_cultural_corridor, earth_blocklist (kept as
reference data), populate_terrain_reference, and heightmap.rs.
The Gemma prompting methodology is preserved in
docs/gemma-naming-methodology.md (separate commit).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Addresses Tyre, Hoshe, and lead review comments on PR #137:
- **Audit doc amendment** (Tyre E1 / Hoshe H1 / Lead): add "Lead override
(2026-04-21)" section at top of docs/architecture/sprint-37-878-audit.md.
Rewrites the conclusion to "DECISION: STRIP" with the cascade-based
rationale. Preserves the original audit body as the pre-override record.
- **Regression tests** (Lead 2a-2b / Hoshe H2 / H3): add POSITIVE
assertions of the new uniform behavior so silent reintroduction fails.
- `phase2_container_verb_labels_uniform_regardless_of_player_state` —
two trials (empty KG, POI-bearing KG) assert container verb labels
equal Phase-1 defaults.
- `monologue_pool_selection_uniform_no_archetype_key` — two observers
with divergent MonologueState both draw from OBSERVE_NPC_LINES.
- **Decision record amendments** (Lead 3 / Tyre S2): D-032, D-035, and
D-057 amended with Phase 6 deferral wording. "Retired pending Phase 6,
not deferred with scaffolding." Reintroduction gate: a confirmed
Phase 6 character-model design.
- **types.rs doc fixes** (Tyre S1 / Hoshe H5): StartupMessage protocol-
flow comment updated to reflect no-version handshake (D-192).
ObserverSnapshot version-history block grows a "Sprint 37 wire-format
shifts" section documenting D-192 + #878 schema drops.
- **observer/tests.rs:944 comment** (Hoshe H6): rewritten to cite
cascade rationale instead of the stale D-032-SUPERSEDED premise.
- **tests/run-atlas-determinism exit** (Hoshe H7): exit 0 when EXIT_CODE=2
(venv/DB missing = skip, not fail). Preserves skip semantics for
tests/run-all on machines without the Python venv.
Follow-up tickets filed:
- #895 (server, low): expand check-systems-db-stamp GENERATOR_SOURCES
to cover gemma_naming.py + naming_core.py (Tyre S3).
- #896 (planning, low): add CLAUDE.md carveout for server wiki writes
closing coverage gates (Tyre S4 / Hoshe H8).
H4 investigation: v01_integration_playthrough.rs was not the only E2E
handshake→tick→snapshot test; coverage preserved by bridge_ipc.rs,
bridge_tcp.rs, and game_loop.rs (the latter is pre-existing-broken
per #885). No replacement test needed.
1142/1142 lib tests pass. cargo clippy -- -D warnings clean.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Adds tests/run-atlas-determinism — imports generate_atlas as a module
and calls process_body() twice with seed=42 and dry_run=True, comparing
the returned markers dicts as JSON. No wiki files are written.
Guardrail against determinism regressions in terrain analysis, city
placement, A* road routing, infrastructure MST, and gate terminal
placement. GJ892f (domed, population 300, 1 city) is the smallest
well-exercised case.
Makefile target: make test-atlas-determinism.
Wired into tests/run-all alongside run-ipc-integration and run-visual.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- test_character_creation_sprint28: before_each now seeds
_selected_bookmark_id and _selected_location_id so the new disabled-
guard in _on_start() (round 2) doesn't silently block 5 existing
tests that call _on_start()/KEY_ENTER without setting up a valid
bookmark selection. Restores the 2 tests Hoshe flagged as R2-H1 plus
3 siblings that would have degraded the same way under the guard.
- tests/run-godot: LOG_FILE now includes $$ (PID) so concurrent runs
across worktrees don't clobber each other's logs. Path is echoed
back via the stdout JSON "log" field and the stderr hint line, so
callers never need to predict it (R2-H2).
Makes tests/run-godot self-containing so neither humans nor LLM callers
have to remember to wrap it in a timeout or pipe it into a file. A hung
test now kills cleanly at 300s with a clear TEST_TIMEOUT marker and
bisection hint instead of silently burning an hour of wall clock (as
Sprint 36 learned).
- Godot+gdUnit4 output goes to /tmp/sr-run-godot.log (overwritten each
run). Nothing streams to stdout/stderr — 20k+ lines of test log into
a terminal or an LLM context is unworkable.
- Stdout: one-line JSON summary, with a "log" field pointing at the
file. On timeout adds "timeout":true and "timeout_sec":300.
- Stderr: a short hint block. On pass: one line. On failure: three
commands to inspect the log. On timeout: a bisection recipe.
- Single well-known path instead of an env var — worktrees each want
their own value and the indirection makes the hint lines meaningless.
Concurrent runs are the caller's problem.
- timeout(1) --foreground --kill-after=10 to escalate to SIGKILL if
Godot ignores SIGTERM.
- HashMap → BTreeMap throughout econ-sim for deterministic iteration (D-010)
- Fix cost_factor: multiplicative gate×zone instead of additive (trade.rs)
- Extract derive_seed to shared prng.rs, consolidate FNV-1a implementation
- Rename run_shock_test → run_no_explosion_check (not D-179 Test 3)
- Deduplicate cross-zone FX rate collection in Test 4
- Replace ORDER BY RANDOM() with deterministic ordering + ChaCha8Rng
- Make commodity coverage failure a hard error consistent with D-175
- Fix gap-fill off-by-one (4 corps → 3 when coverage = 0)
- Correct test report: EconEvent exists, location_type is body/station
All four D-179 stability tests still pass.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Add D-180 EconEvent struct (target, effect, duration, visibility enums)
with no-op handler to satisfy #809 spec. Import MARK_PRIMARY and MIXED
currency zone assignments from wiki/economics/currency_zones.toml (D-172).
All four D-179 stability tests now pass.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Test plan covering #806, #807, #808, #809 acceptance criteria aligned
with D-179 stability tests and Phase 2 economics deliverable requirements.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Add live server lifecycle to tests/run-visual (start/stop server per
scenario, parse LISTENING:{port}). Add MessagePack snapshot replay to
visual_capture.gd via Protocol.decode_snapshot() — exercises the full
client pipeline from wire bytes to rendered fog. Three replay scenarios
(hub_spawn, fog_theater, hub_after_movement) plus one live scenario
(fog_live_hub). Add gen_gauntlet_fixtures.rs to produce .msgpack fixtures
from the Gauntlet test world. Add max_diff_pct threshold to visual-diff.
Makefile: add fixtures-gauntlet target, fix build-client double-import,
preserve .godot cache in clean.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- gen_fixtures.rs: version comments now say PROTOCOL_VERSION instead
of hardcoded 14, so they stay correct across bumps
- run-ipc-benchmark: remove --iterations flag that was parsed but
never forwarded to the Rust test (compile-time constant governs)
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Six test runner scripts at tests/: run-rust, run-godot, run-ipc-fixtures,
run-ipc-protocol, run-ipc-integration, run-all. Plus run-ipc-benchmark
for Layer 3 timing. All produce structured JSON stdout, support --filter,
and exit 0/non-zero. Makefile targets updated to delegate to scripts.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Document non-blocking receive contract in perf_bench.rs docstring,
confirming no TCP deadlock race (Hoshe #1, critical)
- Make shadowcast parser order-independent — flush on new config header
instead of requiring Recursive after Symmetric (Hoshe #2)
- Fix p95 calculation: use floor(0.95*(N-1)) nearest-rank instead of
ceil(0.95*N)-1 which was off-by-one at N=50 (Hoshe #4)
- Error on --compare when no baseline file exists (Hoshe #5)
- Add D-031 10tps assumption comment to TICK_BUDGET_US (Tyre #1)
- Strengthen snapshot assertion: require warmup + half measurement
window instead of warmup + 1 (Tyre #2)
- Regenerate baseline with corrected p95 (356µs, was 526µs)
Hoshe #3 (.PHONY) was already addressed — perf-baseline is in the
.PHONY declaration on Makefile line 10.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Adds `make perf-baseline` — boots the full server plugin stack with
real content, measures 50 ticks (5 warmup), captures per-tick timing,
entity counts, and process RSS. Includes shadowcast benchmarks. Saves
structured JSON to tests/perf/baseline.json for regression detection.
Supports --compare mode (>20% threshold).
First baseline: mean 366µs, p95 526µs (0.5% of D-026 100ms budget).
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Create client/, server/, tooling/, tests/, .config/ (tracked) and
.cache/ (gitignored). Update CLAUDE.md project structure to reflect
the new layout including workshops directory.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
