Feature Placement PR Checklist (League / Lab / Dimensions)¶

Last updated: 2026-02-28

Use this checklist for every product/UI PR that introduces a new page, panel, chart, model output, or workflow.

Canonical framework: League / Lab / Dimensions Brief

Review Goal¶

Each feature must have one primary station: - League: observed state - Lab: experimental analysis - Dimensions: counterfactual/possible state exploration

If a feature spans multiple stations, split it into separate surfaces with explicit transitions.

Hard Gates (Pass/Fail)¶

All gates must pass before merge.

1. Primary station is explicit
2. PR states one owner station (League, Lab, or Dimensions).

3. Fail if station is omitted or multi-owned by default.

4. Primary user question is explicit

5. PR includes one dominant user question the feature answers.

6. Fail if question is vague ("insights", "explore data") without a concrete ask.

7. Evidence type matches station

8. League uses observed/validated data.
9. Lab uses controlled experiments with tunable parameters.
10. Dimensions uses scenario/counterfactual generation.

11. Fail if evidence type conflicts with station.

12. Uncertainty contract is correct

13. League: no speculative claims presented as fact.
14. Lab: conditional outputs are labeled as model-dependent.
15. Dimensions: outputs are probabilistic/calibrated, not deterministic claims.

16. Fail on certainty mismatch.

17. Interaction contract is correct

18. League: read/monitor/compare interactions.
19. Lab: test/fit/adjust interactions.
20. Dimensions: simulate/branch/compare outcomes interactions.

21. Fail if dominant user action belongs to another station.

22. Loss function alignment is documented

23. PR states how the feature reduces the owner station's loss.

24. Fail if no explicit loss framing.

25. Cross-station leakage is controlled

26. Any secondary capability links out to the correct station.
27. Fail if one page tries to be all three stations.

Station-Specific Loss Checks¶

League Loss¶

L_league = inaccuracy + staleness + ambiguity + missing_context + friction

Must show: - Source/provenance clarity - Freshness handling (timestamp, season, or data scope) - Clear factual language

Hard fail conditions: - Predictive/simulated values shown as current fact - Missing time scope on key stats

Lab Loss¶

L_lab = spurious_confidence + non_reproducibility + poor_explainability + friction + overfit_to_view

Must show: - Reproducible controls (axes, model choice, params, seed when relevant) - Model assumptions and caveat language - Comparability across configurations

Hard fail conditions: - No control state visibility - "Best model" claims without diagnostics context

Dimensions Loss¶

L_dimensions = miscalibration + state_incoherence + false_precision + opaque_assumptions + narrow_coverage

Must show: - Scenario assumptions surfaced to user - Distribution/range outputs, not single-point certainty - Ability to compare alternate branches

Hard fail conditions: - Single deterministic future presented as truth - Hidden assumptions that materially change outcomes

PR Description Template (Required)¶

Paste this into every relevant PR:

## Feature Placement
- Primary station: `League | Lab | Dimensions`
- Feature summary: <one sentence>
- Primary user question: <one sentence>
- Dominant user verb: `see/verify` | `test/compare` | `simulate/project`

## Why This Station
- Why it belongs here:
  - <bullet>
  - <bullet>
- Why it does NOT belong in other stations:
  - <bullet>
  - <bullet>

## Loss Function Impact
- Station loss terms reduced:
  - <term>: <how>
  - <term>: <how>
- New risks introduced:
  - <risk + mitigation>

## Uncertainty Contract
- Claim type: `factual` | `conditional` | `probabilistic`
- How uncertainty is communicated: <copy/UI>

Scoring Rubric (Secondary Check)¶

Score each item 0-2: - 2: clear and correct - 1: partial/ambiguous - 0: missing/incorrect

Criteria: 1. Station clarity 2. User-question clarity 3. Evidence-station fit 4. Uncertainty fit 5. Interaction fit 6. Loss-function articulation 7. Leakage control

Threshold: - 12+ = acceptable (if all hard gates pass) - <12 = revise

Note: hard gates override score. A hard-gate fail blocks merge even with a high score.

Imagined Feature Placement Examples¶

Placement Anti-Patterns¶

• League page contains speculative model output without caveats.
• Lab page hides model parameters and cannot be reproduced.
• Dimensions page offers only one "official" predicted future.
• One page attempts full read + experiment + simulate in a single screen.

Tie-Break Rule¶

If a feature is genuinely ambiguous, classify by the dominant user action: - user mostly reads/verifies -> League - user mostly tunes/tests -> Lab - user mostly branches/simulates -> Dimensions

If still ambiguous, split into: - factual baseline block in League - experiment block in Lab - scenario block in Dimensions