Files
lolympus-tournament-support/docs/plans/double-elimination-connector-routing.md

85 lines
3.9 KiB
Markdown

# Implementation Plan: Double-Elimination Connector Routing
## Overview
Replace per-edge winner rail staggering with deterministic, collision-aware winner
bundles while retaining the existing target-corridor `H-V-H` path model and the
current loser/reset routing behavior.
## Architecture Decisions
- Keep routing as a pure layout calculation; the React renderer consumes unchanged
`BracketConnectorLayout` paths.
- Allocate baseline rails first, then choose one rail for each same-target winner
bundle while leaving loser/reset coordinates unchanged.
- Score complete candidate paths by parallel-stroke clearance, strict crossings,
and distance from the corridor midpoint.
- Permit shared `verticalX` only when the participating vertical intervals do not
overlap; otherwise search the same corridor for the nearest deterministic pair
of separate rails with focused-stroke clearance.
## Task List
### Phase 1: Regression Contract
- [x] Add tests proving quiet winner bundles use the corridor midpoint and all
shareable same-target winner paths remain aligned.
- [x] Replace target-column rail uniqueness with interval-aware segment assertions.
- [x] Cover reversed A/B input where overlapping vertical spans require the
deterministic separate-rail fallback.
- [x] Confirm the focused test fails with the current allocator.
### Checkpoint: RED
- [x] The new behavioral assertion fails for the expected unequal `verticalX`.
- [x] Existing tests remain unchanged except where their invariant is obsolete.
### Phase 2: Bundle-Aware Allocator
- [x] Represent connector routing inputs with precomputed source/target anchors.
- [x] Retain stable independent rails for loser/reset routes.
- [x] Group the two winners entering each target and select the nearest safe shared
rail without card penetration, forbidden overlap, or avoidable added crossings.
- [x] Fall back to the nearest deterministic, stroke-clear separate rails only
when the pair's own vertical intervals overlap and cannot safely share a rail.
### Checkpoint: GREEN
- [x] Focused layout tests pass for four-team, eight-team, reset-final, and shuffled
input fixtures.
- [x] Strict crossing, collinear-overlap, and 5px parallel-clearance safeguards pass.
### Phase 3: Verification and Review
- [x] Run the full web suite, lint, and production build; document the unrelated
pre-existing full-lint failures while verifying changed files separately.
- [x] Inspect the diff for correctness, simplicity, security, performance, and
adherence to the approved spec.
- [x] Perform an independent review and address all required findings.
### Checkpoint: Complete
- [x] All success criteria in the spec are verified.
- [x] No unrelated files are modified.
## Risks and Mitigations
| Risk | Impact | Mitigation |
|---|---|---|
| Blind midpoint alignment adds crossings in busy upper rounds | High | Score complete candidate paths and move the winner bundle together to the nearest safe rail. |
| Shared rails create overlapping vertical segments | High | Check interval overlap before allowing a shared `verticalX`; search deterministic fallback rails with 5px clearance otherwise. |
| Visually parallel rails pass exact-coordinate tests but overlap at stroke width | High | Require 5px clearance across overlapping vertical runs and cover it for 4/8-team fixtures. |
| Routing changes depend on input array order | Medium | Sort routes and bundles by graph display order and retain shuffled-fixture equality tests. |
| A long connector bends through an intermediate card | High | Restrict every candidate to the target corridor and retain card-penetration assertions. |
| Reset-final slot carry-over is accidentally treated as a winner pair | Medium | Bundle only connectors whose computed kind is `winner`; keep reset paths independent. |
## Dependencies
```text
Regression contract -> bundle-aware rail allocator -> full verification -> review
```
## Open Questions
None.