settled-reach/docs/workshops/test-architecture/round-3-notes.md

title, description, type, status, workshop, agent, round, created

title	description	type	status	workshop	agent	round	created
QA Strategy & Test Architecture Workshop — Round 3 Notes	Qatux's synthesis notes from round 3 prioritization and final specifications	workshop	archived	test-architecture		3	2026-02-17

QA Strategy & Test Architecture Workshop — Round 3 Notes

Workshop: QA Strategy & Test Architecture Round: 3 (Prioritization & Final Specs) Date: 2026-02-17 Documenter: Qatux Participants: Tyre, Dudley, Stig, Hoshe, Justine, Gestalt, Ozzie

---

Executive Summary

Round 3 produced build-ready specifications across all tracks. Every agent delivered implementation-level detail: Dudley's determinism patches are copy-pasteable, Justine's Makefile targets are ready for make, Stig's 38 client tests have function names and assertion logic, Hoshe's 59-item prioritized backlog has effort estimates and dependency chains, Tyre's Sprint 8 plan has a critical path and dependency graph, Gestalt's Gauntlet map has 48 entities with coordinate positions, and Ozzie's human tester walkthrough covers start-to-finish with terminal mockups.

Key outcomes:

• Sprint 8 scope: 10 implementation items, ~7-8 team-days. Infrastructure only — no Gauntlet rooms, no anti-tedium features. Critical path: --test-mode (0.5d) -> test client binary (2-3d) -> Layer 3 test (0.5d).
• Gauntlet map: 7 rooms + Central Hub, 48 entities, 2 cross-cuts, hub-and-spoke topology. Designed for Sprint 8, room content built Sprint 9+.
• All Round 2 open questions resolved. 11 questions answered: SetTickRate while paused = bug (reject), entity index recycling = safe (generation counter), canonical room ordering = required, blocked_entities = feasible Sprint 9, test client at tooling/test-client/ (lead override), 0 cross-cuts Sprint 8 (infrastructure only), content-ron in pre-pr = yes, fixed spawn order in setup function, kill+relaunch for full restart.
• One lead override during Round 3: Test client binary moved from server/src/bin/test_client.rs to tooling/test-client/ as a separate workspace crate. Dudley verified bridge types are already pub-exported — no server changes needed.

Scope tension resolved: Tyre scopes Sprint 8 as infrastructure-only. Dudley designs a 5-room Gauntlet loader MVP. Gestalt designs the full 7+Hub map. These are compatible: Tyre's Sprint 8 ships the pipes (flags, binary, test harness), Gestalt's map is the design document, Dudley's loader is the Sprint 9 implementation target. The map design happens now; room building happens Sprint 9.

---

1. Sprint 8 Implementation Plan (Tyre)

Dependency Graph

S8-1: Determinism fixes ────────────────────────────┐
                                                      ├─> S8-5: make pre-pr
S8-2: Content cross-ref validation ─────────────────┤
                                                      │
S8-3: --test-mode + --port 0 ──> S8-4: Test client ──┤
                                  binary MVP           │
                                       │               │
                                       └─> S8-6: Layer 3 test wiring
                                                       │
S8-7: Pause guard tests ────────────────────────────┘ (parallel, no deps)
S8-8: Determinism regression tests ── (after S8-1)
S8-9: EntityRegistry lifecycle tests ── (parallel, no deps)
S8-10: Fixture staleness check ── (after S8-5)

Items

#	Item	Owner	Effort	Depends On
S8-1	Determinism fixes (A, B, D)	Dudley	0.5d	Nothing
S8-2	Content cross-reference validation (9 checks)	Justine	1d	Nothing
S8-3	--test-mode + --port 0 server flags	Dudley	0.5d	Nothing
S8-4	Test client binary MVP	Dudley	2-3d	S8-3
S8-5	make pre-pr chain	Justine	0.5d	S8-1, S8-2
S8-6	Layer 3 test wiring	Dudley	0.5d	S8-3, S8-4
S8-7	Pause guard tests (6 tests)	Dudley	0.5d	Nothing
S8-8	Determinism regression tests	Dudley	0.5d	S8-1
S8-9	EntityRegistry lifecycle tests (3 tests)	Dudley	0.25d	Nothing
S8-10	Fixture staleness check	Justine	0.25d	S8-5

Critical path: S8-3 (0.5d) -> S8-4 (2-3d) -> S8-6 (0.5d) = 3-4 days. Total: Server (Dudley) ~5-6d, Tooling (Justine) ~1.75d = ~7-8 team-days.

What Explicitly Does NOT Ship Sprint 8

• Gauntlet room content (rooms 1-14)
• Anti-tedium features (room reset, hub teleport, WRONG button)
• Crossterm live terminal display
• CI automation (Gitea Actions)
• Performance baselines
• Encoding asymmetry cross-language tests
• Client tests (Stig's 38)
• Cross-room transition scenarios

---

2. Determinism Fixes — Final Patch Specs (Dudley)

Three fixes shipping Sprint 8. Fix C was already done (self-corrected in Round 2). Total: ~22 lines production code, 4 regression tests.

Fix A: BTreeSet for visible_positions + sort visible_tiles

Files: server/src/perception/query.rs, server/src/perception/observer/mod.rs

• visible_positions: HashSet<(i32, i32)> -> BTreeSet<(i32, i32)> in VisibilityGeometry
• visible_ids: HashSet<u64> -> BTreeSet<u64> in filter_visible_entities
• Add visible_tiles.sort_by_key(|t| (t.x, t.y)) after collection
• Update collect_remembered_entities signature to accept &BTreeSet
• sector_lookup: HashMap stays as-is (point-lookup only, confirmed safe per OQ-4)

Regression tests: snapshot_visible_tiles_are_sorted, sprint_anomaly_picks_lowest_stable_id

Fix B: Sort visible entities by entity_id

File: server/src/perception/observer/mod.rs

Insert entities.sort_by_key(|e| e.entity_id); after collect_remembered_entities.

Regression test: snapshot_entities_sorted_by_id

Fix D: Sort movers in validate_movement

File: server/src/simulation/movement.rs

Collect movers into Vec, sort by entity.to_bits(), then process. Deterministic collision resolution — entity with lower to_bits() wins contested tiles.

Regression test: validate_movement_deterministic_collision_winner

---

3. Server --test-mode Final Spec (Dudley)

Complete main.rs replacement provided. Key design:

Aspect	Behavior
Stdout	LISTENING:{port} after TCP bind (test client parses this)
Stderr	All tracing via tracing_subscriber::fmt::layer().with_writer(std::io::stderr)
Content	Gauntlet content pack (when ready), fallback to proof room
Seed	--seed 42 default in test-mode
Shutdown	Exits after first client disconnect
Port	--port 0 for OS-assigned port (uses existing accept_on(listener))

setup_proof_room() extracted as reusable function for both modes.

---

4. Test Client Binary Final Spec (Tyre + Dudley)

Lead Override: Crate Location

During Round 3, the lead overruled the Round 2 location. Test client moves from server/src/bin/test_client.rs to tooling/test-client/ as a separate workspace crate.

Note: Tyre's Round 3 spec was written before this override and still references server/src/bin/. Dudley's addendum confirms the override and verifies all bridge types are already pub-exported — no server changes needed.

CLI (Sprint 8 MVP)

settled-reach-test-client [OPTIONS]

Connection:
  --connect <host:port>     (default: 127.0.0.1:9876)

Input:
  --replay <file.jsonl>     JSONL input file (one JSON array per tick)

Output (mutually exclusive):
  --text                    Structured text to stdout (default)
  --json                    JSON to stdout (for golden files)
  --quiet                   No output (CI assertions only)

Assertions:
  --golden <file.json>      Compare final snapshot, exit 1 on diff
  --ticks <N>               Disconnect after N ticks

Text Output Format

=== Tick 42 | Player (15,10) facing East | Stance: Walk | TickRate: Full ===
Game time: Day 0, 04:12 (Morning)
Room: Occlusion Corridor
Entities (5):
  npc:100     (18,10) Forward  rel:Neutral     vis:Visible      d=3
  ...
Tiles: 31 visible
Interactions (2):
  npc:100 [Talk(1), ExamineNpc(2)] distance=3
Inventory: 2/9 [item:300(slot-0), item:301(slot-3)]
Monologue: "Something about this manifest doesn't add up."
===

Entity labels: kind:entity_id. Sorted by distance (nearest first), ties broken by entity_id. Sections with no data omitted.

Text Renderer

server/src/bridge/text_renderer.rs — library function format_snapshot_text() callable by the test client crate and integration tests. Pub-exported from server crate. Full implementation (~100 lines) provided by Tyre.

Golden File Comparison

JSON with sorted keys. Recursive diff_json_values() produces field-by-field diff on mismatch. Exit code 1 on diff.

Replay Format

JSONL — one JSON array of PlayerInput per line. Empty array [] = idle tick. Uses rmp_serde::to_vec (NOT to_vec_named) to match GDScript encoding.

Effort Estimate

~2-3 days: binary scaffolding (0.5d), text renderer (0.5-1d), replay loading (0.5d), golden file comparison (0.5d), Layer 3 wiring (0.5d).

---

5. Gauntlet Map Specification (Gestalt)

MVP Room List: 7 rooms + Central Hub

#	Room	Size	Observer	Facing	Entities
0	Central Hub	24x24	(50,58)	--	4 signs
1	Fog Theater	44x32	(56,18)	South	4 (3 NPCs, 1 object)
2	Occlusion Corridor	42x22	(84,58)	East	4 NPCs
3	Inventory Warehouse	30x28	(17,54)	East	11 (10 objects, 1 NPC)
4	Interaction Gallery	24x20	(14,92)	East	5 (2 NPCs, 3 objects)
5	Pause Chamber	16x16	(50,86)	North	1 NPC
6	Dialogue Room	28x20	(50,114)	North	4 NPCs
7	Crowd Plaza	32x32	(96,94)	West	15 NPCs

Total: 48 entities. Map bounds: 0-116 x 0-124 sim tiles.

Topology

Hub-and-spoke with 2 cross-cuts:

• cross-cut-W: Inventory Warehouse <-> Interaction Gallery (T3: Full Inventory Interact)
• cross-cut-E: Occlusion Corridor <-> Crowd Plaza (T1: Sprint Exit)

7 corridors connecting rooms to hub and each other. All 6 tiles wide, 8-14 tiles long.

StableId Assignment Order

Entities receive StableIds in canonical spawn order:

1. Hub entities (signs): StableId 1-4
2. Fog Theater: StableId 5-8
3. Occlusion Corridor: StableId 9-12
4. Inventory Warehouse: StableId 13-23
5. Interaction Gallery: StableId 24-28
6. Pause Chamber: StableId 29
7. Dialogue Room: StableId 30-33
8. Crowd Plaza: StableId 34-48

Additive-only rule: Existing rooms and entities are NEVER reordered. New rooms and entities append. This preserves golden file stability.

Cross-Room Transitions (3 MVP)

#	Transition	Path	Systems Tested
T1	Sprint Exit	Crowd Plaza -> cross-cut-E -> Occlusion	Sprint suppression + LOS recalculation
T3	Full Inventory Interact	Inventory -> cross-cut-W -> Interaction Gallery	Inventory limit + verb computation
T6	Pause Anywhere	Pause -> Hub -> any room	Pause persistence across teleport/room change

Scope Note: Gestalt vs Tyre vs Dudley

Gestalt designs 7+Hub rooms for the full Gauntlet. Tyre scopes Sprint 8 as infrastructure-only (no rooms). Dudley designs a 5-room loader MVP as the Sprint 9 implementation target. These are consistent: the map design is complete, room building follows infrastructure.

---

6. Client Test Suite — Final 38 Tests (Stig)

Deduplicated from 41 candidates (32 original + 6 Tyre additions + 3 Stig R2 additions).

Priority	Count	Sprint	Categories
P0	2	Sprint 8	Monologue overwrite (Bug #5), camera static during pause (Bug #2)
P1	7	Sprint 8-9	Fog shader (4), entity lifecycle (2), pending recognition blob (1)
P2	24	Sprint 9	Camera (5), entity alpha/color (5), UI (12), lerp (1), teleport (1)
P3	5	Sprint 10+	Z-layer (4), lerp target (1)

Each test includes: function name, category, exact assertions, and setup requirements.

Fog Constants

Added to client/scripts/autoloads/fog_state.gd:

• VIS_HIDDEN = 0, VIS_PERIPHERAL = 180, VIS_FORWARD = 255
• EXP_UNEXPLORED = 0, EXP_EXPLORED = 128, EXP_VISIBLE = 255

5 magic number replacements in fog_state.gd. No shader changes needed.

---

7. Prioritized Test Backlog — 59 Items (Hoshe)

Summary by Sprint

Sprint	P0 items	P1 items	Combined effort
Sprint 8	10 (~7d)	15 (~10.75d)	~17.75d
Sprint 9	10 (~7.5d)	12 (~12.5d)	~20d
Sprint 10+	--	12 (~18.75d)	~18.75d

Sprint 8 P0 (10 items, ~7d)

1. Determinism Fix A (0.5d)
2. Determinism Fix B (0.25d)
3. Determinism Fix D (0.25d)
4. Server --test-mode + --port 0 (1d)
5. make pre-pr target (0.5d) 6-8. Pause guard tests: movement_discarded, unpause_accepted, roundtrip (0.75d total)
6. Content cross-reference validation (1.5d)
7. Fixture staleness check (0.25d)

Bug Catalogue Coverage

Every Sprint 6-7 bug has Sprint 8 P0/P1 coverage:

• Bug #1 (snapshot delivery): #4 (--test-mode), #26 (Layer 3)
• Bug #2 (camera startup): #21 (client P0: camera static during pause)
• Bug #3 (movement while paused): #6, #7, #8, #13 (pause guard suite)
• Bug #4 (MessagePack -128): #15, #16, #17, #18 (boundary value matrix)
• Bug #5 (monologue overwrite): #21 (client P0: monologue not lost)
• Bug #6 (snapshot spam): #1, #2 (determinism), #29 (golden files)

---

8. Tooling Specifications (Justine)

make pre-pr — Final

6-step chain: lint -> build -> test -> validate-content -> check-fact-ids -> fixtures-check. ~90s incremental, ~8min clean cache. Fixture staleness is BLOCKER (not WARNING).

Branch-specific variants: pre-pr-server, pre-pr-client, pre-pr-content.

Full Makefile provided with failure output examples.

make perf-baseline

• Baseline file: tests/perf/baseline.json (committed)
• Median of 5 runs, thresholds: <15% PASS, 15-30% WARN, >30% FAIL
• Machine tag prevents cross-machine comparison
• tooling/perf-compare and tooling/perf-update scripts provided (full bash code)
• Blocked on Gauntlet — shadowcast bench is the only available benchmark until rooms ship

Golden File Workflow

• Generator: server/tests/gauntlet_golden_gen.rs (generates JSON at ticks 0, 10, 100)
• Comparator: server/tests/gauntlet_golden.rs (field-by-field diff on mismatch)
• JSON format: sorted keys, pretty-printed, deterministic
• make golden-diff / make golden-update targets provided
• Full Rust code for both (~180 lines combined)

CI Pipeline Design (Deferred)

3-tier pipeline documented and ready for when lead greenlights:

• Commit tier (<2min): lint + validate-content + check-fact-ids
• PR tier (<15min): build + test + fixtures staleness (BLOCKER)
• Nightly tier (<30min): Layer 3 + golden files + perf benchmarks + content scaling

Complete .gitea/workflows/ci.yaml provided (~100 lines). Self-hosted runner required.

---

9. Content Validation — 9 Checks (Hoshe + Justine merged)

Extends tooling/validate-content (Python) with a second pass after schema validation.

#	Check	Severity	Source
1	canonical_id_uniqueness	ERROR	Hoshe
2	relationship_target_resolution	ERROR	Both
3	location_slug_resolution	ERROR	Both
4	dialogue_location_resolution	ERROR	Both
5	fact_id_resolution	ERROR	Both (absorbs check-fact-ids)
6	triangle_membership_resolution	ERROR	Both
7	npc_count_accuracy	WARNING	Hoshe
8	dialogue_line_id_uniqueness	ERROR	Hoshe
9	bidirectional_relationship_consistency	WARNING	Justine

ContentIndex Python class skeleton provided. Phased rollout: Phase 1 (NPC/triangle/district) Sprint 8, Phase 2 (dialogue/fact_ids) Sprint 8, Phase 3 (warnings) Sprint 9.

---

10. Anti-Tedium Specifications (Ozzie + Gestalt + Stig + Dudley)

Sprint 8 Priorities (Ozzie)

Priority	Feature	Effort	Justification
P1	Room Reset Triggers	1-1.5d	Can't re-test without it
P2	WRONG Button MVP	1d	Can't report bugs without it
P3 (Sprint 9)	Hub Teleport	0.5d	Convenience, not necessity
P4 (Sprint 9)	Timer + Checklist	1.5-2d	Engagement, not necessity

Note: Tyre's Sprint 8 plan does NOT include anti-tedium features (infrastructure-only scope). Anti-tedium ships Sprint 9 per Tyre's roadmap, Sprint 8 per Ozzie/Gestalt's priority.

Room Reset — Final Spec

• Trigger: Step on ResetPlate + Interact (NOT automatic)
• Server: RoomResetTrigger component, RoomSnapshots resource, 10-tick debounce
• Resets: entity positions, KG, fog (room tiles), room-sourced inventory, dialogue state
• Does NOT reset: other rooms, player position, session timer, SimRng state
• Test-mode only: entities only added with --test-mode

WRONG Button MVP — Final Spec

• Hotkey: F12
• Captures: current ObserverSnapshot (JSON), text output, tick/room/seed, tester description
• Output: tests/bug-reports/gauntlet-t{tick}-{timestamp}/ (4 files: report.md, snapshot.json, text_output.txt, description.txt)
• Zero server changes needed for MVP
• Full version (Sprint 9+): 60-tick ring buffer, input history, replay seed

Hub Teleport

• Home key -> PlayerAction::TeleportToHub
• Instant camera snap + 0.3s fade-to-black
• Gauntlet-only (server rejects in non-Gauntlet maps)

Client-Side UI (Stig)

• bug_report.gd autoload: ~80 lines GDScript, F12 handler, ring buffer, modal prompt
• InsertOverlay/FlashRect: shared by reset flash and teleport fade
• GauntletProgress overlay: room name, run counter, timer, progress bar (Sprint 9)
• Fog constants migration: 6 named constants, 5 line replacements

---

11. Human Tester Workflow (Ozzie)

6-Phase Walkthrough

1. Navigate to a Room — walk from Hub, room name updates, timer starts, checklist loads
2. Execute the Checklist — printed checklist or terminal display, auto-tracking for type: auto items
3. Encounter a Bug — F12 -> pause -> one-line prompt -> 4 files saved -> resume
4. Reset and Re-Test — step on reset plate -> room reverts to tick-0 state -> retry
5. Move to Another Room — Home -> Hub -> walk to next room
6. End Session — Ctrl+C -> session summary (rooms tested, coverage %, times, bug reports)

Quick-Test Developer Workflow

Target: 65 seconds from fix to verification. Build (10s) -> start server+client (3s) -> navigate (5s) -> test checklist items (45s) -> exit (2s).

Entity Visibility Symbols

Symbol	State	Meaning
●	VISIBLE	In clear vision cone
◐	REMEMBERED	In fog, previously seen
◌	FOGGED	Detected but not recognized
✕	BLOCKED	LOS blocked by wall (debug, requires blocked_entities)
⚡	RECOGNIZING	Mid-cognitive-delay

Test Client Display

10 sections: Header, Player, Entities, Fog, Sound (Sprint 9+), Cognition (Sprint 9+), Interactions, Monologue/Dialogue, Inventory, Status. MVP ships sections 1-4, 7-10.

---

12. Checklist YAML Schema (Stig + Ozzie merged)

Per-Room Format

# content/gauntlet/rooms/{room_id}/checklist.yaml
room: occlusion_corridor
checks:
  - id: occ_01_hidden_not_visible
    description: "NPC behind wall is NOT visible"
    type: auto          # auto | manual
    step: "Stand at (45,3), face East"
    condition:
      player_near: [45, 3]
      entity: hidden-1
      expected: blocked
    if_wrong: |
      LOS leaking through wall. Check shadowcast.rs.

Condition Grammar (7 types)

player_near, player_facing, entity + expected (blocked/visible/remembered/recognizing), expected_sector, perception_mode, fog_visible_count_min/max, inventory_count, dialogue_active, monologue_contains.

Cross-Room Checklist

content/gauntlet/cross_room_checks.yaml at Gauntlet root. Transition scenarios with multi-room paths and structured conditions.

Auto-Tracking Split

Feature	Godot Client	Test Client
Room name + timer	Yes	Yes
Progress bar (X/Y)	Yes (total from YAML)	Yes
Auto-evaluate conditions	No	Yes (Rust, type-safe)
Per-item display	No	Yes

---

13. Boundary Value Tests — 41 Values (Hoshe)

Encoding Asymmetry (Resolved)

GDScript uses int_16 for values 256-32767; Rust uses uint_16. Both are spec-valid. rmp_serde accepts int_16-encoded positive values for u64 fields (Dudley traced through rmp-serde 1.3.1 in Round 2).

4-Direction Tests

Direction	What	Location	When
Rust -> GDScript (fixture)	Snapshots with overlap-zone ticks	test_msgpack_boundaries.gd	Every PR
GDScript -> Rust (fixture)	Inputs with overlap-zone ticks	serialization.rs	Every PR
Rust -> GDScript (raw bytes)	Hand-crafted uint_16/uint_32 bytes	test_msgpack_boundaries.gd	Every commit
GDScript -> Rust (raw bytes)	Hand-crafted int_16/int_32 bytes	serialization.rs	Every commit

Full test code provided for all 4 directions.

---

14. Layer 3 Test — Final Spec (Hoshe)

server/tests/layer3.rs, #[test] #[ignore], <10 seconds.

• Build server binary
• Launch as subprocess with --test-mode --port 0
• Parse LISTENING:{port} from stdout (5-second timeout)
• Connect via TCP, send PlayerInput via rmp_serde::to_vec (not to_vec_named)
• Receive ObserverSnapshot, assert version, tick, entity count, player entity kind
• ServerGuard drop pattern for cleanup (kill on drop)
• Tolerates initial tick=0 snapshot before input processed (Dudley R2 adjustment)

Full Rust code (~80 lines) provided.

---

15. Resolved Questions

New Answers from Round 3

ID	Question	Answer	By
Q1 (R2-OQ-09)	Canonical room ordering for StableId?	Yes — append-only rule. YAML room order = spawn order = StableId order.	Dudley, Gestalt
Q2 (R2-OQ-10)	Room reset vs full restart?	Both. Room reset for iteration, kill+relaunch for determinism.	Gestalt
Q3	Cross-cuts in Sprint 8?	0 cross-cuts Sprint 8 (infrastructure only). 2 in map design. Build first 2 Sprint 10.	Tyre, Gestalt
Q4 (R2-OQ-05)	blocked_entities feasibility?	Feasible, ~300 tile lookups/tick. Sprint 9 scope, gated behind --test-mode.	Dudley
Q5 (R2-OQ-07)	Test client binary location?	tooling/test-client/ (lead override). Bridge types already pub-exported.	Tyre, Dudley
Q6 (R2-OQ-08)	Cross-room checklist location?	content/gauntlet/cross_room_checks.yaml at Gauntlet root.	Gestalt
Q7 (R2-OQ-01)	SetTickRate while paused?	Bug. Reject. Pause exits only via explicit Unpause. 4-line fix in input.rs.	Dudley
Q8 (R2-OQ-02)	Entity index recycling safe?	Yes (bevy generation counter). Discipline: call unregister() on despawn.	Dudley
R2-OQ-03	content-ron in pre-pr?	Yes, in full pre-pr only. Not in pre-pr-server.	Tyre
R2-OQ-06	Checklist overlay in Godot?	Lightweight overlay (room name + timer). Full tracking test-client-only.	Stig
R2-OQ-11 (cross-cuts)	4 cross-cuts too many?	Keep all 4 in design. Build 2 for Sprint 8 map. Don't consolidate Sound Lab.	Tyre, Gestalt

Questions Deferred to Implementation

ID	Question	Assigned To	When
R2-OQ-04	Fixture staleness separate target?	Justine	Resolved: No. 10-15s cost negligible.
OQ-11	Client tests headless stability	Stig	Before making client tests a CI merge gate

---

16. Sprint 9+ Roadmap (Tyre)

Tier 1: Sprint 9 (~10-12 team-days)

R-01: Gauntlet rooms 1-4 (3-4d), R-02: Room reset trigger (1.5d), R-03: Hub teleport (0.5d), R-04: Client tests P0-P1 (2-3d), R-05: Determinism golden files (1d), R-06: Fog byte constants (0.25d), R-07: Encoding asymmetry tests (1.5d).

Tier 2: Sprint 10 (~15-17 team-days)

R-08: WRONG button MVP (2d), R-09: Gauntlet rooms 5-8 (3-4d), R-10: Room timer + PBs (1d), R-11: Checklist auto-tracking (2d), R-12: Performance baselines (1.5d), R-13: Client tests P2 (3d), R-14: Enhanced test client terminal (2d).

Tier 3: Sprint 11+ (build when needed)

Gauntlet rooms 9-14, cross-room transitions, CI automation, content scaling stress, blocked_entities, client test headless, client tests P3, additional pause guard tests, WRONG button full capture, bidirectional relationship warnings.

Risk Register (5 risks)

Risk	Likelihood	Impact	Mitigation
Test client takes >3 days	Medium	Delays Layer 3	MVP scope intentionally minimal
Proof room too simple for golden files	Low	Golden files don't catch real bugs until rooms exist	Proves toolchain; value comes Sprint 9
Content validation finds many existing errors	Medium	Sprint 8 effort bloat	Run as WARNING first sprint
Client test headless blocks CI	Low	Can't automate client tests	Manual make pre-pr-client interim
Fixture BLOCKER frustrates devs	Low	Devs skip pre-pr	Education; false positives are worse

---

17. Points of Agreement (consensus)

1. Sprint 8 is infrastructure. All agents accept Tyre's framing: build the pipes, fill them Sprint 9+.
2. Determinism fixes are complete. Copy-pasteable code, 4 regression tests, ~22 lines total.
3. make pre-pr is the developer discipline tool. Replaces CI until CI is greenlighted. BLOCKER on fixture staleness.
4. Gauntlet map design is complete. 7+Hub rooms, 48 entities, 2 cross-cuts. Additive-only constraint.
5. Test client binary architecture is settled. tooling/test-client/ crate (lead override), text renderer in server library, golden file JSON comparison.
6. 38 client tests are prioritized and build-ready. Function names, assertions, setup — ready for Stig to implement.
7. Anti-tedium is fully specified. All 4 features have implementation details. Priority ranked.
8. All open questions resolved. Zero questions carry forward to implementation.

18. Points of Tension (resolved or minor)

1. Fixture staleness: BLOCKER. Tyre's argument accepted by all. Justine updated spec.
2. Sprint 8 anti-tedium: Ozzie/Gestalt prioritize room reset + WRONG button for Sprint 8. Tyre defers both to Sprint 9 (infrastructure-only scope). The implementation plan (Tyre) is authoritative for Sprint 8 scope; anti-tedium ships Sprint 9.
3. Gauntlet room count for Sprint 8: Dudley (5 rooms), Gestalt (7+Hub). Tyre (0 rooms — infrastructure only). Resolution: map design has 7+Hub, room building is Sprint 9.
4. Test client location: Tyre R3 says server/src/bin/. Lead overrides to tooling/test-client/. Dudley confirms no server changes needed. Lead override is authoritative.
5. F3 debug overlay: Deferred indefinitely per Stig. WRONG button captures same data. Ozzie accepts.

---

19. Gaps Remaining

1. Gauntlet content authoring. Room YAML, entity placement, checklist definitions need writing. No specific agent assigned for content creation.
2. Client test headless stability (OQ-11). Still unverified. Must be tested before making client tests a CI merge gate.
3. Sound events not in ObserverSnapshot. Required for full test client display (sections 5-6). Not yet designed server-side.
4. Cognitive delay visual not implemented. Tyre's tests #33-34 (blob rendering, recognition transition) depend on D-060 client implementation.
5. Gauntlet map vs Dudley's loader room count. Gestalt designs 8 rooms (7+Hub). Dudley's loader MVP has 5 rooms. Reconciliation needed during Sprint 9 implementation (which rooms build first).