docs(10): capture phase context

This commit is contained in:
2026-03-27 15:36:08 +01:00
parent a91b1a854c
commit 9761dcb9e5
2 changed files with 151 additions and 0 deletions
@@ -0,0 +1,104 @@
# Phase 10: Classification Layer - Context
**Gathered:** 2026-03-27
**Status:** Ready for planning
<domain>
## Phase Boundary
Add ~21 new TrafficClass constants and port-matching rules covering all new protocol families (Mail, File Transfer, Remote Access, Database, Discovery, VoIP, Web extension, Infrastructure extension). All existing 10 protocol classifications remain unchanged — no regression. `AllClasses()` covers all new classes and `DefaultRules` maps all new ports.
</domain>
<decisions>
## Implementation Decisions
### PROTO-08 Frequency Strategy
- **D-01:** LDAP, Kerberos, and Syslog get built-in TrafficClass constants and classification rules, but their synthesis frequencies are handled by `autoAssignFreq` (FNV hash in [2500, 4000] Hz) rather than designed table slots. No Phase 9 frequency table redesign needed.
- **D-02:** These 3 classes do NOT get entries in `ClassFreqConfigs` in this phase — they are treated like user-defined custom classes for frequency purposes. Phase 11 will add their `ClassFreqConfigs` entries using `autoAssignFreq`-compatible Hz values.
### Plain vs TLS Port Handling
- **D-03:** Plaintext and TLS/secure variants of the same protocol share a single TrafficClass. Multiple ports map to the same class constant: IMAP (143) + IMAPS (993) → ClassIMAP, POP3 (110) + POP3S (995) → ClassPOP3, LDAP (389) + LDAPS (636) → ClassLDAP, SIP (5060) + SIPS (5061) → ClassSIP, FTP data (20) + FTP control (21) → ClassFTP.
- **D-04:** Exception: SMTP (port 25) and SMTP-submission (port 587) remain separate classes (ClassSMTP and ClassSMTPSub) because Phase 9 designed distinct frequency slots for them (214 Hz and 305 Hz respectively).
### Naming Convention
- **D-05:** Follow existing uppercase convention for TrafficClass string values: "IMAP", "POP3", "RDP", "MySQL", etc. Matches existing "ICMP", "DNS", "HTTPS" pattern. SMTP-submission uses "SMTP-sub" to match the Phase 9 frequency table label.
### Rule Ordering
- **D-06:** New port-specific rules insert before the catch-all `{tcp, 0, ClassOtherTCP}` and `{udp, 0, ClassOtherUDP}` entries. Catch-alls remain last. First-match-wins semantics preserved.
### Claude's Discretion
- Exact ordering of new rules within the specific-port section (before catch-alls)
- Test structure — whether to extend existing TestClassify or add new test functions
- How to update `TestAllClassesCount` (hardcoded to 14) — update the count or make it dynamic
- Whether to group rules by family in `DefaultRules` with comments, or keep flat
</decisions>
<canonical_refs>
## Canonical References
**Downstream agents MUST read these before planning or implementing.**
### Classification Package (primary modification target)
- `classify/types.go` — TrafficClass constants, `AllClasses()` function (currently 14 classes)
- `classify/rules.go``DefaultRules` slice (currently 12 rules, first-match-wins)
- `classify/classifier.go``Classifier.Classify()` method, `hashBucket()` function
- `classify/classifier_test.go` — Tests for all existing classes, `TestAllClassesCount` (hardcoded to 14)
### Synth Package (reference — not modified in Phase 10)
- `synth/config.go` — Frequency allocation table comment (lines 74-110), `ClassFreqConfigs` map, `FreqConfig` struct with Group field
### Requirements
- `.planning/REQUIREMENTS.md` — PROTO-01 through PROTO-09
- `.planning/ROADMAP.md` — Phase 10 success criteria
### Research
- `.planning/research/FEATURES.md` — Complete protocol list with ports, families, priorities
- `.planning/research/STACK.md` — gopacket layer availability (port-based classification confirmed for all new protocols)
- `.planning/research/PITFALLS.md` — Rule count scaling (linear scan ~12→~40), PROTO-08 gap
- `.planning/phases/09-frequency-design-and-group-architecture/09-RESEARCH.md` — Pitfall 6: PROTO-08 classes not in table
</canonical_refs>
<code_context>
## Existing Code Insights
### Reusable Assets
- `buildTCPPacket(t, dstPort)` and `buildUDPPacket(t, dstPort)` test helpers — reuse for all new protocol tests
- `Rule` struct with Protocol/DstPort/Class — same structure works for all new rules
- `hashBucket()` function — unchanged, still handles unmatched packets
### Established Patterns
- TrafficClass is a `string` type with `const` declarations — add new constants following same pattern
- `AllClasses()` returns a hand-maintained slice — must be updated with all new classes
- `DefaultRules` is a `[]Rule` literal — new rules append before catch-alls
- Tests use table-driven subtests within `TestClassify` — follow same pattern for new protocols
### Integration Points
- `AllClasses()` is used by `synth/config_test.go` `TestClassFreqConfigsMatchAllClasses` to verify every class has a FreqConfig entry — new classes added here will fail that test until Phase 11 adds their ClassFreqConfigs entries
- `TestAllClassesCount` hardcodes `14` — must be updated to new count
- Phase 11 depends on these constants existing to add ClassFreqConfigs entries
</code_context>
<specifics>
## Specific Ideas
- The complete protocol list with ports is documented in `.planning/research/FEATURES.md` (lines 297-311)
- Phase 9 frequency table comment in `synth/config.go` (lines 74-110) shows exactly which classes are expected with their family groupings
- The `TestClassFreqConfigsMatchAllClasses` cross-check will break when new classes are added without ClassFreqConfigs entries — Phase 10 planner needs to account for this (skip or temporarily adjust the test)
</specifics>
<deferred>
## Deferred Ideas
None — discussion stayed within phase scope.
</deferred>
---
*Phase: 10-classification-layer*
*Context gathered: 2026-03-27*
@@ -0,0 +1,47 @@
# Phase 10: Classification Layer - 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:** 10-classification-layer
**Areas discussed:** PROTO-08 frequency gap, Plain vs TLS port handling
---
## PROTO-08 Frequency Gap
| Option | Description | Selected |
|--------|-------------|----------|
| Auto-assign range (Recommended) | Let autoAssignFreq handle them (FNV hash in [2500, 4000] Hz). Built-in constants + rules, but frequencies behave like user-defined classes. Simplest, no table redesign. | ✓ |
| Extend the table | Add 3 more slots above SIP at 2449 Hz. Proper built-in ClassFreqConfigs entries. Pushes into auto-assign territory, requires range shift. | |
| Squeeze into Infrastructure band | Insert 3 slots between existing Infrastructure entries (65-133 Hz). Most musically coherent but touches Phase 9's locked design. | |
**User's choice:** Auto-assign range (Recommended)
**Notes:** LDAP, Kerberos, Syslog get TrafficClass constants and rules but their Hz values come from autoAssignFreq rather than designed table slots. No Phase 9 table redesign needed.
---
## Plain vs TLS Port Handling
| Option | Description | Selected |
|--------|-------------|----------|
| Same class (Recommended) | IMAP 143 and IMAPS 993 both classify as ClassIMAP. Simpler, fewer classes. Exception: keep SMTP/SMTP-sub separate per Phase 9 design. | ✓ |
| Separate classes | IMAP and IMAPS are distinct TrafficClasses with distinct sounds. More granular but doubles class count and needs more frequency slots. | |
| Same class, no SMTP exception | Merge everything including SMTP + SMTP-sub into single ClassSMTP. Wastes Phase 9 SMTP-sub slot at 305 Hz. | |
**User's choice:** Same class (Recommended)
**Notes:** Plain/TLS variants share one class. SMTP (25) and SMTP-submission (587) remain separate because Phase 9 designed distinct frequency slots for them.
---
## Claude's Discretion
- Rule ordering within specific-port section
- Test structure for new protocols
- How to update TestAllClassesCount
- Whether to group rules by family with comments
## Deferred Ideas
None — discussion stayed within phase scope.