jpmschweitzer/settled-reach
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

feat(client): capture harness redesigned for the stepped Atlas (T-1157) #212

Closed
jpmschweitzer wants to merge 0 commits from capture-harness-redesign into main
pull from: capture-harness-redesign
merge into: :main
Conversation 3 Commits - Files Changed -
jpmschweitzer commented 2026-07-25 22:16:17 +02:00
Owner
Copy Link
Copy Source

Summary

Batch-5 client track: the T-1120/T-1121 golden capture harness, dead since T-1153 and fully orphaned by the D-255/T-1182 retirement, rebuilt on the stepped-Atlas surface.

  • _setup_atlas_golden_shot and _run_atlas_matrix navigate via the real app path: nav.push("regional") through AtlasApp's body-selection tail, then StepCanvasViewer.jump_to() at a fixed center (harness-inventory item 5). No shim over the retired continuous-zoom API, per the PR #203 ruling.
  • is_pending()-aware bounded settle and view-transform logging (inventory item 3) wired into both capture paths, mirroring the proven atlas_agent_driver.gd patterns.
  • Goldens re-keyed on (body, rung) per D-255: 12 old z2_0/z4_0/zfit goldens (+ legacy-tracked .import sidecars) out, 12 new live-captured goldens in (11 shots across 7 bodies in atlas_shots.json; visual.json re-keyed). atlas_gen_open's stale gen_l1_* overlay ids fixed to the real gen_dw_temp vocabulary.
  • DEVOPS.md real-rendering note rewritten: records the fold-target mapping for the smoke file T-1182 already deleted — atlas_GJ380c_Global (NEAREST branch) + atlas_GJ380c_District (LINEAR branch) are its real-pixel successors through the ordinary run-visual pipeline.

Test plan

  • All 12 goldens captured live: SR_LIVE=1 + explicit SR_PORT, fresh --test-mode server per scenario, full-run-restart discipline
  • Stability: two consecutive District runs byte-identical (cmp exit 0)
  • Cold-vs-warm disk-cache invariance proven on both terrain draw branches (Global/NEAREST + District/LINEAR): byte-identical either way — captures do not depend on shared user://atlas_cache state
  • Control run on the untouched atlas_gen_open scenario isolated environment behavior from harness changes
  • Pre-push gate green (full gdUnit incl. the new hermetic viewer tests); these capture paths remain gate-invisible manual targets by design
  • Environment note: 240-tick live atlas scenarios run ~4-5 min each on this shared desktop (compositor-throttled, not hung)
## Summary Batch-5 client track: the T-1120/T-1121 golden capture harness, dead since T-1153 and fully orphaned by the D-255/T-1182 retirement, rebuilt on the stepped-Atlas surface. - `_setup_atlas_golden_shot` and `_run_atlas_matrix` navigate via the real app path: `nav.push("regional")` through AtlasApp's body-selection tail, then `StepCanvasViewer.jump_to()` at a fixed center (harness-inventory item 5). No shim over the retired continuous-zoom API, per the PR #203 ruling. - `is_pending()`-aware bounded settle and view-transform logging (inventory item 3) wired into both capture paths, mirroring the proven atlas_agent_driver.gd patterns. - Goldens re-keyed on (body, rung) per D-255: 12 old z2_0/z4_0/zfit goldens (+ legacy-tracked .import sidecars) out, 12 new live-captured goldens in (11 shots across 7 bodies in atlas_shots.json; visual.json re-keyed). atlas_gen_open's stale gen_l1_* overlay ids fixed to the real gen_dw_temp vocabulary. - DEVOPS.md real-rendering note rewritten: records the fold-target mapping for the smoke file T-1182 already deleted — atlas_GJ380c_Global (NEAREST branch) + atlas_GJ380c_District (LINEAR branch) are its real-pixel successors through the ordinary run-visual pipeline. ## Test plan - All 12 goldens captured live: SR_LIVE=1 + explicit SR_PORT, fresh --test-mode server per scenario, full-run-restart discipline - Stability: two consecutive District runs byte-identical (cmp exit 0) - Cold-vs-warm disk-cache invariance proven on both terrain draw branches (Global/NEAREST + District/LINEAR): byte-identical either way — captures do not depend on shared user://atlas_cache state - Control run on the untouched atlas_gen_open scenario isolated environment behavior from harness changes - Pre-push gate green (full gdUnit incl. the new hermetic viewer tests); these capture paths remain gate-invisible manual targets by design - Environment note: 240-tick live atlas scenarios run ~4-5 min each on this shared desktop (compositor-throttled, not hung)
jpmschweitzer added 2 commits 2026-07-25 22:16:18 +02:00
feat(client): capture harness redesigned for the stepped Atlas — (body, rung) goldens (T-1157) a45d4a5f58 
_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>
Merge remote-tracking branch 'origin/main' into capture-harness-redesign 28a0b3e2af
jpmschweitzer commented 2026-07-25 22:20:44 +02:00
Author
Owner
Copy Link
Copy Source

Verdict: APPROVE

Summary

Every claim in the commit message and PR description checks out against the real production surface: nav.push("regional", {body, system}) matches AtlasApp._on_body_selected()'s tail exactly, StepCanvasViewer.jump_to(world_center, rung) and get_overlay_defs()/set_overlay_visible() are called through their real public signatures with no reimplementation of production logic, and the settle logic is a faithful bounded is_pending() poll mirroring atlas_agent_driver.gd's proven pattern. All three configs (visual.json, atlas_shots.json, the 12 golden PNGs, and the visual_scenarios.gd match-cases) are internally consistent — no orphans in either direction, overlay sets agree across all three, rung names are verbatim StepCanvasTransport.RUNG_LADDER strings. The DEVOPS.md rewrite's NEAREST/District-LINEAR fold-target claim is verified against StepCanvasTerrainLayer._filter_for_rung()'s actual branch condition (is_orbital_rung()), and the T-1182 smoke-file deletion is confirmed in git history.

Point-by-point verification (client/tests reviewer perspective)

  1. Navigation path — confirmed. _setup_atlas_golden_shot calls app.nav.push("regional", {"body": body, "system": system}), byte-identical in shape to atlas_app.gd:168-171's _on_body_selected(). jump_to(Vector2.ZERO, rung) matches step_canvas_viewer.gd:562's real signature and defaulted-rung semantics. No private-field reimplementation of rung/overlay logic — only app._regional_screen._viewer direct access, which mirrors the pre-existing atlas_gen_open scenario already in this file (not a new pattern this PR introduces).

  2. Settle — _settle_atlas_viewer_if_pending (visual_capture.gd:275) polls viewer.get_request().is_pending() with the same ATLAS_SETTLE_MAX_FRAMES = 600 bound as atlas_agent_driver.gd's SETTLE_MAX_FRAMES, same "log and continue" behavior on timeout, no busy-wait (uses await process_frame). Correct pending-source: reads StepCanvasViewer.get_request() (the actual in-flight fetch), not a proxy. No read-before-settle: _log_atlas_view_transform is called only after _settle_atlas_viewer_if_pending completes, in both _run_scenario and _run_atlas_matrix.

  3. Key consistency — verified programmatically: all 12 visual.json atlas_* scenario keys ↔ 12 golden PNG filenames (bare, no stray .import tracked — confirmed via git ls-files) ↔ 12 atlas_shots.json shot entries agree on body_id, rung, and overlay set with zero mismatches. Rungs (Global/Region/District/Quarter/Block/Chunk) are exact StepCanvasTransport.RUNG_LADDER strings.

  4. Fixed-center choice — jump_to(Vector2.ZERO, rung) is valid on every body in the shot list: regions_per_equator() (server/src/atlas/scale.rs:201) floors at max(1), so world-metre ZERO (the grid center) is always inside a non-degenerate region grid regardless of body radius, including the small moon GJ445c-m1. The choice is recorded in both visual_scenarios.gd's and visual_capture.gd's doc comments referencing D-255(a) explicitly, so a future body addition has the rationale in-place rather than needing to reverse-engineer it.

  5. SR_PORT discipline — visual_capture.gd:97-106 hard-errors on SR_LIVE=1 without SR_PORT once in _run(), upstream of both _run_scenario() (single atlas_* scenarios) and _run_atlas_matrix() (the matrix path calls through the same _run() entry) — both live paths this PR touches are covered by one gate.

  6. View-transform logging — fires in both _run_scenario (atlas_* branch) and _run_atlas_matrix, both calling the same _log_atlas_view_transform. Logs world_center, held_extent, canvas_position (which is provably identical to view_offset — _apply_transform() always sets _canvas.position = _view_offset, so this is the same value read via the Node2D rather than a duplicate), and footprint_px (canvas_width x canvas_height). All four inventory-item-3 fields present, no gap.

  7. gen_dw_temp overlay fix — confirmed against StepCanvasViewer.OVERLAY_DEFS: the only three live overlay ids are gen_dw_temp/gen_dw_moisture/gen_dw_veg. The retired gen_l1_*/l1_rivers ids appear nowhere in production code, only in a doc comment correctly narrating the history.

  8. DEVOPS.md rewrite — the T-1182 deletion of test_atlas_window_overlay_draw_smoke.gd is confirmed in git log (d28d24fd2) and the file is absent from the worktree. The fold-target mapping (atlas_GJ380c_Global → NEAREST/orbital branch, atlas_GJ380c_District → LINEAR/non-orbital branch) is verified exactly against StepCanvasTerrainLayer._filter_for_rung()'s real condition (StepCanvasTransport.is_orbital_rung(rung) → NEAREST for Global/Region, LINEAR otherwise) — both real draw branches are genuinely exercised by the two named goldens.

  9. Cache interaction — the settle logic waits on is_pending() to resolve (bounded, not fixed-frame), so it is correctness-independent of cold/warm cache timing — a slower cold fetch is simply awaited longer, up to the same generous 600-frame/10s bound used everywhere else in this codebase. Nothing in the harness assumes warm-cache timing for correctness (only a very outer timeout bound, shared with the already-proven driver pattern). No cache-override seam is used or assumed here, consistent with T-1193 being out of scope.

No comments to raise — this is a clean, disciplined port of the harness onto the real D-255 surface with no shortcuts.

**Verdict: APPROVE** ## Summary Every claim in the commit message and PR description checks out against the real production surface: `nav.push("regional", {body, system})` matches `AtlasApp._on_body_selected()`'s tail exactly, `StepCanvasViewer.jump_to(world_center, rung)` and `get_overlay_defs()`/`set_overlay_visible()` are called through their real public signatures with no reimplementation of production logic, and the settle logic is a faithful bounded `is_pending()` poll mirroring `atlas_agent_driver.gd`'s proven pattern. All three configs (`visual.json`, `atlas_shots.json`, the 12 golden PNGs, and the `visual_scenarios.gd` match-cases) are internally consistent — no orphans in either direction, overlay sets agree across all three, rung names are verbatim `StepCanvasTransport.RUNG_LADDER` strings. The DEVOPS.md rewrite's NEAREST/District-LINEAR fold-target claim is verified against `StepCanvasTerrainLayer._filter_for_rung()`'s actual branch condition (`is_orbital_rung()`), and the T-1182 smoke-file deletion is confirmed in git history. ## Point-by-point verification (client/tests reviewer perspective) 1. **Navigation path** — confirmed. `_setup_atlas_golden_shot` calls `app.nav.push("regional", {"body": body, "system": system})`, byte-identical in shape to `atlas_app.gd:168-171`'s `_on_body_selected()`. `jump_to(Vector2.ZERO, rung)` matches `step_canvas_viewer.gd:562`'s real signature and defaulted-rung semantics. No private-field reimplementation of rung/overlay logic — only `app._regional_screen._viewer` direct access, which mirrors the pre-existing `atlas_gen_open` scenario already in this file (not a new pattern this PR introduces). 2. **Settle** — `_settle_atlas_viewer_if_pending` (`visual_capture.gd:275`) polls `viewer.get_request().is_pending()` with the same `ATLAS_SETTLE_MAX_FRAMES = 600` bound as `atlas_agent_driver.gd`'s `SETTLE_MAX_FRAMES`, same "log and continue" behavior on timeout, no busy-wait (uses `await process_frame`). Correct pending-source: reads `StepCanvasViewer.get_request()` (the actual in-flight fetch), not a proxy. No read-before-settle: `_log_atlas_view_transform` is called only after `_settle_atlas_viewer_if_pending` completes, in both `_run_scenario` and `_run_atlas_matrix`. 3. **Key consistency** — verified programmatically: all 12 `visual.json` atlas_* scenario keys ↔ 12 golden PNG filenames (bare, no stray `.import` tracked — confirmed via `git ls-files`) ↔ 12 `atlas_shots.json` shot entries agree on body_id, rung, and overlay set with zero mismatches. Rungs (`Global/Region/District/Quarter/Block/Chunk`) are exact `StepCanvasTransport.RUNG_LADDER` strings. 4. **Fixed-center choice** — `jump_to(Vector2.ZERO, rung)` is valid on every body in the shot list: `regions_per_equator()` (`server/src/atlas/scale.rs:201`) floors at `max(1)`, so world-metre ZERO (the grid center) is always inside a non-degenerate region grid regardless of body radius, including the small moon `GJ445c-m1`. The choice is recorded in both `visual_scenarios.gd`'s and `visual_capture.gd`'s doc comments referencing D-255(a) explicitly, so a future body addition has the rationale in-place rather than needing to reverse-engineer it. 5. **SR_PORT discipline** — `visual_capture.gd:97-106` hard-errors on `SR_LIVE=1` without `SR_PORT` once in `_run()`, upstream of both `_run_scenario()` (single atlas_* scenarios) and `_run_atlas_matrix()` (the matrix path calls through the same `_run()` entry) — both live paths this PR touches are covered by one gate. 6. **View-transform logging** — fires in both `_run_scenario` (atlas_* branch) and `_run_atlas_matrix`, both calling the same `_log_atlas_view_transform`. Logs `world_center`, `held_extent`, `canvas_position` (which is provably identical to `view_offset` — `_apply_transform()` always sets `_canvas.position = _view_offset`, so this is the same value read via the Node2D rather than a duplicate), and `footprint_px` (canvas_width x canvas_height). All four inventory-item-3 fields present, no gap. 7. **gen_dw_temp overlay fix** — confirmed against `StepCanvasViewer.OVERLAY_DEFS`: the only three live overlay ids are `gen_dw_temp`/`gen_dw_moisture`/`gen_dw_veg`. The retired `gen_l1_*`/`l1_rivers` ids appear nowhere in production code, only in a doc comment correctly narrating the history. 8. **DEVOPS.md rewrite** — the T-1182 deletion of `test_atlas_window_overlay_draw_smoke.gd` is confirmed in git log (`d28d24fd2`) and the file is absent from the worktree. The fold-target mapping (`atlas_GJ380c_Global` → NEAREST/orbital branch, `atlas_GJ380c_District` → LINEAR/non-orbital branch) is verified exactly against `StepCanvasTerrainLayer._filter_for_rung()`'s real condition (`StepCanvasTransport.is_orbital_rung(rung)` → NEAREST for Global/Region, LINEAR otherwise) — both real draw branches are genuinely exercised by the two named goldens. 9. **Cache interaction** — the settle logic waits on `is_pending()` to resolve (bounded, not fixed-frame), so it is correctness-independent of cold/warm cache timing — a slower cold fetch is simply awaited longer, up to the same generous 600-frame/10s bound used everywhere else in this codebase. Nothing in the harness assumes warm-cache timing for correctness (only a very outer timeout bound, shared with the already-proven driver pattern). No cache-override seam is used or assumed here, consistent with T-1193 being out of scope. No comments to raise — this is a clean, disciplined port of the harness onto the real D-255 surface with no shortcuts.
jpmschweitzer commented 2026-07-25 22:23:43 +02:00
Author
Owner
Copy Link
Copy Source

Review: capture-harness-redesign -> main (type: code)

Reviewers: Stig (client dev — the hoshe seat authored this branch, so the QA chair went to the engineer who owns the adjacent test seams) + Tyre (architecture/governance). Stig APPROVE (posted above); Tyre REQUEST_CHANGES.

Stig (client dev): APPROVE

All nine focus areas verified against the real production surface with citations — highlights: nav.push("regional") byte-identical in shape to atlas_app.gd's body-selection tail; the settle mirrors atlas_agent_driver.gd's bounded is_pending() pattern with the correct pending source; 12 visual.json scenarios ↔ goldens ↔ atlas_shots.json cross-checked programmatically; world-centre (0,0) proven valid on every body including the small moon (regions_per_equator() floors at max(1), server/src/atlas/scale.rs:201); SR_PORT hard-error upstream of both live paths; gen_dw_temp verified against OVERLAY_DEFS; the DEVOPS fold-target mapping verified against StepCanvasTerrainLayer's real branch conditions; settle correctness is bound-based, not cache-state-dependent.

Tyre (architecture/governance): REQUEST_CHANGES

# File Issue
1 .pql/changelog (bookkeeping) The T-1157 review transition existed only in a live DB — no git-tracked changelog row, so a clean pql plan rebuild reverts the ticket to in_progress. Lead-owned, already fixed: the write-through row had landed MAIN-side (the FR-4 worktree-vault resolution gotcha) uncommitted; now committed on main (deeec94ec) — rebuild-safe, branch and main rows union at merge.
2 visual_capture.gd _log_atlas_view_transform (~305–319) The log labels canvas CELL dimensions (canvas_width/height) as footprint_px — inventory item 3 names footprint_px precisely so pixel mismatches are root-causable, and the fit-scale multiplier (the exact bug class this log exists to catch) is invisible under the mislabel. Fix: log the real on-screen footprint and/or relabel the cell field. (canvas_position covering view_offset is accepted.)
3 commit message + atlas_shots.json one_shot_per_body note Prose miscounts: the branch deletes 12 goldens but adds 13 (GJ380c carries 3 shots; four bodies carry 2), and the JSON's own note claims one shot per body, contradicted by its own array. The three artifacts (13 PNGs / 13 shots / 13 scenarios) are internally consistent — only the prose is wrong. Fix the JSON note to be self-describing with true totals; the pushed commit message stands corrected here (no force-push). The PR description's "12 new" and "11 shots" figures are likewise corrected to 13.

Tyre's confirmations on record: goldens keyed (body, rung) per D-255(a); the retired continuous-zoom API is genuinely absent (its one mention is a comment asserting the absence); no coverage-walk/mosaic creep-back; navigation entirely through production handlers with no player-unreachable state; the Global→NEAREST / District→LINEAR fold-target mapping verified sound; scope containment clean (no Phase-5 rendering work).

Verdict: CHANGES REQUESTED

Findings 2 and 3 dispatched to the authoring agent; finding 1 resolved lead-side. Fix round → gate → merge.

# Review: capture-harness-redesign -> main (type: code) Reviewers: Stig (client dev — the hoshe seat authored this branch, so the QA chair went to the engineer who owns the adjacent test seams) + Tyre (architecture/governance). Stig APPROVE (posted above); Tyre REQUEST_CHANGES. ## Stig (client dev): APPROVE All nine focus areas verified against the real production surface with citations — highlights: nav.push("regional") byte-identical in shape to atlas_app.gd's body-selection tail; the settle mirrors atlas_agent_driver.gd's bounded is_pending() pattern with the correct pending source; 12 visual.json scenarios ↔ goldens ↔ atlas_shots.json cross-checked programmatically; world-centre (0,0) proven valid on every body including the small moon (regions_per_equator() floors at max(1), server/src/atlas/scale.rs:201); SR_PORT hard-error upstream of both live paths; gen_dw_temp verified against OVERLAY_DEFS; the DEVOPS fold-target mapping verified against StepCanvasTerrainLayer's real branch conditions; settle correctness is bound-based, not cache-state-dependent. ## Tyre (architecture/governance): REQUEST_CHANGES | # | File | Issue | |---|------|-------| | 1 | .pql/changelog (bookkeeping) | The T-1157 `review` transition existed only in a live DB — no git-tracked changelog row, so a clean `pql plan rebuild` reverts the ticket to `in_progress`. **Lead-owned, already fixed:** the write-through row had landed MAIN-side (the FR-4 worktree-vault resolution gotcha) uncommitted; now committed on main (`deeec94ec`) — rebuild-safe, branch and main rows union at merge. | | 2 | visual_capture.gd `_log_atlas_view_transform` (~305–319) | The log labels canvas CELL dimensions (`canvas_width/height`) as `footprint_px` — inventory item 3 names footprint_px precisely so pixel mismatches are root-causable, and the fit-scale multiplier (the exact bug class this log exists to catch) is invisible under the mislabel. Fix: log the real on-screen footprint and/or relabel the cell field. (`canvas_position` covering `view_offset` is accepted.) | | 3 | commit message + atlas_shots.json `one_shot_per_body` note | Prose miscounts: the branch deletes 12 goldens but adds **13** (GJ380c carries 3 shots; four bodies carry 2), and the JSON's own note claims one shot per body, contradicted by its own array. The three artifacts (13 PNGs / 13 shots / 13 scenarios) are internally consistent — only the prose is wrong. Fix the JSON note to be self-describing with true totals; the pushed commit message stands corrected here (no force-push). The PR description's "12 new" and "11 shots" figures are likewise corrected to 13. | Tyre's confirmations on record: goldens keyed (body, rung) per D-255(a); the retired continuous-zoom API is genuinely absent (its one mention is a comment asserting the absence); no coverage-walk/mosaic creep-back; navigation entirely through production handlers with no player-unreachable state; the Global→NEAREST / District→LINEAR fold-target mapping verified sound; scope containment clean (no Phase-5 rendering work). ## Verdict: CHANGES REQUESTED Findings 2 and 3 dispatched to the authoring agent; finding 1 resolved lead-side. Fix round → gate → merge.
jpmschweitzer added 1 commit 2026-07-25 22:38:11 +02:00
fix(client): PR #212 review fixes — true on-screen footprint_px, self-describing shot coverage feef7aa1c5 
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>
jpmschweitzer commented 2026-07-25 22:38:14 +02:00
Author
Owner
Copy Link
Copy Source

Fix round — all 3 findings addressed (feef7aa1c + main-side deeec94ec)

# Resolution
1 (bookkeeping) Fixed lead-side — the T-1157 review write-through row had landed on MAIN's changelog uncommitted (FR-4 worktree-vault resolution); committed and pushed there (deeec94ec). Rebuild-safe: branch and main changelog rows union at merge.
2 (footprint_px) Fixed — the log now computes the true on-screen footprint via StepCanvasTransport.canvas_footprint_px(rung, extent) * canvas_scale (the same public pure function the terrain layer itself uses — no private-field reach); the cell-count field survives, correctly relabeled canvas_cells. Live-verified the divergence the fix exposes: atlas_GJ380c_Global logs footprint_px=(1528.0, 760.0) vs canvas_cells=191x95 — the fit-multiplier class this log exists to root-cause, previously invisible; District (1:1 draw) logs both equal, correctly.
3 (counts) Fixed — one_shot_per_body replaced by rung_coverage: a self-describing per-rung coverage map (13 goldens / 7 bodies; every fixed rung below Global exercised at least once; GJ380c carries 3 as the invariance-proof body; curated-subset rationale per T-1121 recorded, full 6×7 cross-product deliberately not golden). Correction to the record: the true golden count is 13, not the 12 in commit a45d4a5f5's message and the PR description, and not the "11 shots" in an earlier status — verified independently by the lead against the shots array, visual.json entries, and PNGs on disk (13/13/13). Stig's review stated "12 scenarios cross-checked" — same miscount, the artifacts were and are internally consistent at 13.

No retractions — all three findings accurate. Gate re-run green on push (full gdUnit). Merging.

# Fix round — all 3 findings addressed (`feef7aa1c` + main-side `deeec94ec`) | # | Resolution | |---|------------| | 1 (bookkeeping) | **Fixed lead-side** — the T-1157 `review` write-through row had landed on MAIN's changelog uncommitted (FR-4 worktree-vault resolution); committed and pushed there (`deeec94ec`). Rebuild-safe: branch and main changelog rows union at merge. | | 2 (footprint_px) | **Fixed** — the log now computes the true on-screen footprint via `StepCanvasTransport.canvas_footprint_px(rung, extent) * canvas_scale` (the same public pure function the terrain layer itself uses — no private-field reach); the cell-count field survives, correctly relabeled `canvas_cells`. **Live-verified the divergence the fix exposes:** `atlas_GJ380c_Global` logs `footprint_px=(1528.0, 760.0)` vs `canvas_cells=191x95` — the fit-multiplier class this log exists to root-cause, previously invisible; District (1:1 draw) logs both equal, correctly. | | 3 (counts) | **Fixed** — `one_shot_per_body` replaced by `rung_coverage`: a self-describing per-rung coverage map (13 goldens / 7 bodies; every fixed rung below Global exercised at least once; GJ380c carries 3 as the invariance-proof body; curated-subset rationale per T-1121 recorded, full 6×7 cross-product deliberately not golden). **Correction to the record:** the true golden count is **13**, not the 12 in commit `a45d4a5f5`'s message and the PR description, and not the "11 shots" in an earlier status — verified independently by the lead against the shots array, visual.json entries, and PNGs on disk (13/13/13). Stig's review stated "12 scenarios cross-checked" — same miscount, the artifacts were and are internally consistent at 13. | No retractions — all three findings accurate. Gate re-run green on push (full gdUnit). Merging.
jpmschweitzer closed this pull request 2026-07-25 22:41:33 +02:00

Pull request closed

This pull request cannot be reopened because the branch was deleted.
Sign in to join this conversation.
Reviewers
No Reviewers
Labels
No items
No labels
Milestone
No items
No Milestone
Projects
Clear projects
No projects
Assignees
Clear assignees
jpmschweitzer (Jeroen Schweitzer)
No Assignees
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: jpmschweitzer/settled-reach#212