diff --git a/.planning/phases/08-test-and-constant-cleanup/08-CONTEXT.md b/.planning/phases/08-test-and-constant-cleanup/08-CONTEXT.md
new file mode 100644
index 0000000..53fcd1c
--- /dev/null
+++ b/.planning/phases/08-test-and-constant-cleanup/08-CONTEXT.md
@@ -0,0 +1,88 @@
+# Phase 8: Test and Constant Cleanup - Context
+
+**Gathered:** 2026-03-27
+**Status:** Ready for planning
+
+<domain>
+## Phase Boundary
+
+Remove stale exported constants (`NumLayers`, `GainPerLayer`) from the synth package and update hardcoded test assertions (`TestFrequenciesInRange`) so that subsequent v1.2 phases can add new traffic classes and frequencies without triggering false CI failures. This is pure cleanup — no new features, no new protocols.
+
+</domain>
+
+<decisions>
+## Implementation Decisions
+
+### Constant Removal Strategy
+- **D-01:** Delete `NumLayers` and `GainPerLayer` constants entirely from `synth/config.go`. They are dead code — `NewBank` already computes `gainPerLayer` dynamically as `1.0 / float64(len(cfgs))` (bank.go:21). No external callers reference either constant outside the test file.
+
+### Frequency Range Test Bounds
+- **D-02:** Replace the hardcoded `[60, 1100]` bounds in `TestFrequenciesInRange` with dynamic validation — derive the valid range from the `ClassFreqConfigs` data itself (e.g., check that all frequencies are positive and below Nyquist) rather than hardcoding a new magic number that would need manual updating when Phase 9/10 add classes above 1100 Hz. The specific approach (positive+Nyquist check, or a generous static bound like `[20, 8000]`) is at Claude's discretion — the key constraint is that adding a new class in the 1100-4000 Hz range must not require editing this test.
+
+### Test Naming
+- **D-03:** Rename `TestNumLayersMatchesAllClasses` to `TestClassFreqConfigsMatchAllClasses` (or similar) to reflect the actual invariant being tested after `NumLayers` removal. The test body already uses `len(synth.ClassFreqConfigs)` and `len(classify.AllClasses())` — only the name references the deleted constant.
+
+### Claude's Discretion
+- Whether to use a generous static upper bound vs a computed Nyquist-based bound for D-02 — either approach satisfies the constraint
+- Whether `WhisperFloor` or other constants in config.go need any adjustment (they don't reference NumLayers, so likely no)
+- Whether `TestClassFreqConfigsComplete` (line 53) should be consolidated with the renamed test since both verify the same invariant
+
+### Folded Todos
+- **"Expand Traffic Classes"** (from `.planning/todos/pending/001-expand-traffic-classes.md`) — This todo requests adding protocols like IMAP, POP3, SNMP, FTP and researching common traffic classes. Phase 8 enables this work by removing the test/constant blockers, but the actual protocol additions are Phase 10's scope. Folded here as context, not as direct Phase 8 work.
+
+</decisions>
+
+<canonical_refs>
+## Canonical References
+
+**Downstream agents MUST read these before planning or implementing.**
+
+### Synth Package (primary targets)
+- `synth/config.go` — Contains `NumLayers` and `GainPerLayer` constants to remove (lines 9-10)
+- `synth/config_test.go` — Contains `TestFrequenciesInRange` (lines 18-24), `TestNumLayersMatchesAllClasses` (lines 61-66), and `TestClassFreqConfigsComplete` (lines 53-59)
+- `synth/bank.go` — `NewBank` already computes `gainPerLayer` dynamically (line 21) — confirms constants are dead code
+
+### Research Context
+- `.planning/research/PITFALLS.md` — Pitfall C4 documents `TestFrequenciesInRange` hardcoding issue
+- `.planning/research/ARCHITECTURE.md` — Lines 414+ document NumLayers/ClassFreqConfigs mismatch risk
+- `.planning/research/SUMMARY.md` — Lines 69, 85, 100 describe Phase 8 cleanup scope
+
+</canonical_refs>
+
+<code_context>
+## Existing Code Insights
+
+### Reusable Assets
+- `synth/bank.go:NewBank` already has the correct dynamic gain computation — no new code needed for gain behavior
+
+### Established Patterns
+- Test file `synth/config_test.go` uses table-driven validation against `ClassFreqConfigs` map and `classify.AllClasses()` — new/renamed tests should follow this pattern
+- `GainPerLayer` constant at line 10 has a comment referencing "D-10" — cleanup should not leave orphan decision references
+
+### Integration Points
+- Only `synth/config.go` and `synth/config_test.go` are modified — no downstream package changes expected
+- `go test ./...` is the verification gate — must pass with zero new failures
+
+</code_context>
+
+<specifics>
+## Specific Ideas
+
+No specific requirements — this is a straightforward cleanup phase with clear targets identified in research.
+
+</specifics>
+
+<deferred>
+## Deferred Ideas
+
+### Reviewed Todos (not folded)
+None — the matched todo was folded as milestone context.
+
+None — discussion stayed within phase scope.
+
+</deferred>
+
+---
+
+*Phase: 08-test-and-constant-cleanup*
+*Context gathered: 2026-03-27*
diff --git a/.planning/phases/08-test-and-constant-cleanup/08-DISCUSSION-LOG.md b/.planning/phases/08-test-and-constant-cleanup/08-DISCUSSION-LOG.md
new file mode 100644
index 0000000..7c50d9c
--- /dev/null
+++ b/.planning/phases/08-test-and-constant-cleanup/08-DISCUSSION-LOG.md
@@ -0,0 +1,60 @@
+# Phase 8: Test and Constant Cleanup - Discussion Log
+
+> **Audit trail only.** Do not use as input to planning, research, or execution agents.
+> Decisions are captured in CONTEXT.md — this log preserves the alternatives considered.
+
+**Date:** 2026-03-27
+**Phase:** 08-test-and-constant-cleanup
+**Areas discussed:** Constant removal strategy, Frequency range bound, Test naming
+**Mode:** Auto (all decisions auto-selected)
+
+---
+
+## Constant Removal Strategy
+
+| Option | Description | Selected |
+|--------|-------------|----------|
+| Delete entirely | Remove NumLayers and GainPerLayer from config.go — bank.go already computes dynamically | ✓ |
+| Deprecate with comment | Keep but mark as deprecated for backward compatibility | |
+| Replace with function | Convert to a function that returns len(ClassFreqConfigs) | |
+
+**User's choice:** [auto] Delete entirely (recommended default)
+**Notes:** NewBank already computes gainPerLayer as 1.0/len(cfgs). No external callers reference either constant.
+
+---
+
+## Frequency Range Test Bound
+
+| Option | Description | Selected |
+|--------|-------------|----------|
+| Dynamic validation | Derive valid range from data (positive + below Nyquist) — no magic numbers | ✓ |
+| Generous static bound | Replace 1100 with e.g. 8000 Hz — simple but still hardcoded | |
+| Remove range test | Delete TestFrequenciesInRange entirely — other tests cover correctness | |
+
+**User's choice:** [auto] Dynamic validation (recommended default)
+**Notes:** Key constraint: adding a class in 1100-4000 Hz range must not require editing this test.
+
+---
+
+## Test Naming
+
+| Option | Description | Selected |
+|--------|-------------|----------|
+| Rename to TestClassFreqConfigsMatchAllClasses | Reflects actual invariant after NumLayers removal | ✓ |
+| Keep current name | Leave as-is despite referencing deleted constant | |
+| Delete test | TestClassFreqConfigsComplete already covers same invariant | |
+
+**User's choice:** [auto] Rename to TestClassFreqConfigsMatchAllClasses (recommended default)
+**Notes:** Test body already uses len() comparisons, only the name references NumLayers.
+
+---
+
+## Claude's Discretion
+
+- Specific approach for dynamic frequency range validation (Nyquist-based vs generous static bound)
+- Whether to consolidate TestClassFreqConfigsComplete with renamed test
+- Whether WhisperFloor needs adjustment (likely not)
+
+## Deferred Ideas
+
+None — discussion stayed within phase scope.