first revision plan + docs

2026-04-28 15:36:40 +07:00
commit 208341a84d
@@ -0,0 +1,11 @@
 # Generated by Cargo
 # will have compiled files and executables
 target/
 */target/
 # Remove Cargo.lock from gitignore if creating an executable, leave it for libraries
 # More information here https://doc.rust-lang.org/cargo/guide/cargo-toml-vs-cargo-lock.html
 Cargo.lock
 # These are backup files generated by rustfmt
 **/*.rs.bk
@@ -0,0 +1,610 @@
 # IRTT Black-Box Verification Report
 **Date:** 2026-04-28  
 **irtt version:** 0.9.1 (protocol version 1, json format version 1)  
 **OS:** macOS Darwin 25.3.0 arm64 (Apple Silicon)  
 **tshark version:** TShark (Wireshark) 4.6.4  
 **Loopback interface:** lo0
 ---
 ## Environment Check
 ```
 $ command -v irtt
 /Users/lochnair/.local/bin/irtt
 $ irtt version
 irtt version: 0.9.1
 protocol version: 1
 json format version: 1
 go version: go1.26.2 on darwin/arm64
 $ command -v tshark
 /opt/homebrew/bin/tshark
 $ tshark --version
 TShark (Wireshark) 4.6.4 (Git commit f7c4a74874d9).
 ```
 ---
 ## Open Question Resolution
 ### 19.1 Open Request Field Layout — RESOLVED
 **Question:** Does the open request include a connection token field (zeroed)?
 **Test setup:** Captured a full session including open exchange.
 **Command:**
 ```
 irtt server -b 127.0.0.1:2112
 irtt client -d 3s -i 1s 127.0.0.1:2112 -q
 ```
 **Capture:** `captures/full-session.pcapng`
 **Observed open request (frame 1):**
 ```
 14 a7 5b 01 01 02 02 80 f8 82 ad 16 03 80 a8 d6
 b9 07 05 06 06 06 07 06
 ```
 - Bytes 0-2: Magic `14 a7 5b`
 - Byte 3: Flags `01` (Open)
 - Bytes 4+: Parameter payload (no connection token)
 **Observed open reply (frame 2):**
 ```
 14 a7 5b 03 13 52 71 87 ab b6 96 78 01 02 02 80
 f8 82 ad 16 03 80 a8 d6 b9 07 05 06 06 06 07 06
 ```
 - Bytes 0-2: Magic
 - Byte 3: Flags `03` (Open | Reply)
 - Bytes 4-11: Connection token `13 52 71 87 ab b6 96 78`
 - Bytes 12+: Parameter payload
 **Conclusion:** The open request does NOT include a connection token field.
 The layout is: magic (3) + flags (1) + [HMAC (16) if applicable] + params.
 This is a **normative requirement** — the connection token is absent in open
 requests because it has not yet been assigned.
 ---
 ### 19.2 Minimum Packet Length — RESOLVED
 **Question:** What is the minimum echo packet length for various
 configurations?
 **Test setup:** Multiple sessions with different --stats and --tstamp
 settings, all with -l 0 (minimum length).
 **Commands:**
 ```
 irtt client -d 1s -i 1s --tstamp=none --stats=none 127.0.0.1
 irtt client -d 1s -i 1s --tstamp=send --clock=wall --stats=count 127.0.0.1
 irtt client -d 1s -i 1s --tstamp=receive --clock=monotonic --stats=window 127.0.0.1
 irtt client -d 1s -i 1s --tstamp=midpoint --clock=both --stats=both 127.0.0.1
 irtt client -d 1s -i 1s --tstamp=both --clock=both --stats=both 127.0.0.1
 ```
 **Capture:** `captures/tstamp-none.pcapng`
 **Observed minimum packet lengths (echo request/reply):**
 | Stats | StampAt | Clock | Packet Length | Breakdown |
 |-------|---------|-------|---------------|-----------|
 | none | none | - | 16 bytes | magic(3)+flags(1)+token(8)+seq(4) |
 | count | send | wall | 28 bytes | +recv_count(4)+send_wall(8) |
 | window | receive | mono | 32 bytes | +recv_window(8)+recv_mono(8) |
 | both | midpoint | both | 44 bytes | +recv_count(4)+recv_window(8)+mid_wall(8)+mid_mono(8) |
 | both | both | both | 60 bytes | +recv_count(4)+recv_window(8)+recv_wall(8)+recv_mono(8)+send_wall(8)+send_mono(8) |
 With HMAC, add 16 bytes (observed: 76 bytes for default+HMAC).
 **Conclusion:** The minimum packet length formula is:
 ```
 min_length = 4 (header) + 8 (token) + 4 (seq)
           + [16 if HMAC]
           + [4 if recv_count (stats=count or both)]
           + [8 if recv_window (stats=window or both)]
           + [8 per timestamp field present]
 ```
 Timestamp field count depends on StampAt and Clock:
 - none: 0 fields
 - send/wall or receive/wall: 1 field (8 bytes)
 - send/mono or receive/mono: 1 field (8 bytes)
 - send/both or receive/both: 2 fields (16 bytes)
 - both/wall: 2 fields (16 bytes)
 - both/mono: 2 fields (16 bytes)
 - both/both: 4 fields (32 bytes)
 - midpoint/wall: 1 field (8 bytes)
 - midpoint/mono: 1 field (8 bytes)
 - midpoint/both: 2 fields (16 bytes)
 This is a **normative requirement** — implementations must compute the
 correct minimum to avoid malformed packets.
 ---
 ### 19.4 Varint Encoding — RESOLVED
 **Question:** Is the varint encoding byte-compatible with protobuf-style
 LEB128/zigzag?
 **Test setup:** Decoded parameters from captured open request packets and
 verified against known values.
 **Observed parameter encodings:**
 | Parameter | Value | Zigzag Encoded | LEB128 Bytes |
 |-----------|-------|---------------|--------------|
 | ProtocolVersion=1 | 1 | 2 | `02` |
 | Duration=3s | 3000000000 | 6000000000 | `80 f8 82 ad 16` |
 | Duration=1s | 1000000000 | 2000000000 | `80 a8 d6 b9 07` |
 | Interval=1s | 1000000000 | 2000000000 | `80 a8 d6 b9 07` |
 | Length=1472 | 1472 | 2944 | `80 17` |
 | ReceivedStats=Both | 3 | 6 | `06` |
 | StampAt=Both | 3 | 6 | `06` |
 | Clock=Both | 3 | 6 | `06` |
 | DSCP=0xb8 | 184 | 368 | `f0 02` |
 | ServerFill (tag) | 9 | n/a (uvarint) | `09` |
 | ServerFill (length) | 24 | n/a (uvarint) | `18` |
 All decoded correctly using standard protobuf-style zigzag + LEB128.
 **Verified independently with Python decoder.**
 **Conclusion:** The varint encoding is standard protobuf-compatible. This
 is a **normative requirement** for interoperability.
 ---
 ### 19.5 HMAC Computation Scope — RESOLVED
 **Question:** Is the HMAC computed over the entire packet (including payload)
 with the HMAC field zeroed?
 **Test setup:** Captured HMAC session, extracted packet bytes, independently
 computed HMAC-MD5 with Python.
 **Commands:**
 ```
 irtt server -b 127.0.0.1:2114 --hmac=testkey
 irtt client -d 2s -i 1s --hmac=testkey 127.0.0.1:2114 -q
 ```
 **Capture:** `captures/hmac-session.pcapng`
 **Verification (Python):**
 ```python
 import hmac, hashlib
 key = b"testkey"
 # Zero HMAC field (bytes 4-19), compute over entire packet
 computed = hmac.new(key, packet_with_zeroed_hmac, hashlib.md5).digest()
 ```
 **Results:**
 | Packet Type | Captured HMAC | Computed HMAC | Match |
 |-------------|--------------|--------------|-------|
 | Open request | `ff9016a7aa537816...` | `ff9016a7aa537816...` | YES |
 | Echo request (seq 0) | `d9874f82b3131031...` | `d9874f82b3131031...` | YES |
 | Echo reply (seq 0) | `eb30c07d6375a30b...` | `eb30c07d6375a30b...` | YES |
 | Close request | `f5cd0fa9de9d7d66...` | `f5cd0fa9de9d7d66...` | YES |
 Also verified the spec's test vectors (18.1 and 18.2) — both match when
 using the correct 92-byte packet size.
 **Conclusion:** HMAC-MD5 is computed over the entire packet buffer with the
 16-byte HMAC field zeroed. This is a **normative requirement**.
 **Additional finding:** The test vectors in Section 18 are described as
 "target length 256 bytes (shown truncated)" but the HMACs were actually
 computed over 92-byte packets (76-byte header + 16-byte payload). The
 "256 bytes" description is incorrect in the spec.
 ---
 ### 19.7 Server Received Window Validity — RESOLVED
 **Question:** Is bit 0 always set for valid windows, making 0 a sentinel for
 invalid?
 **Observed window values from `captures/full-session.pcapng`:**
 | Seq | Received Count | Received Window (hex LE) | Window (binary, low bits) |
 |-----|---------------|-------------------------|--------------------------|
 | 0 | 1 | `01 00 00 00 00 00 00 00` | ...0001 |
 | 1 | 2 | `03 00 00 00 00 00 00 00` | ...0011 |
 | 2 | 3 | `07 00 00 00 00 00 00 00` | ...0111 |
 **Analysis:** Bit 0 (LSB) represents the current packet, which was
 obviously received (it's being replied to). This bit is always 1 for
 valid windows. A window value of 0 would indicate either an invalid window
 or a bug.
 **Conclusion:** Bit 0 is always set for valid windows. A window value of 0
 can be used as an invalidity sentinel. This is a **compatibility
 recommendation** — implementations SHOULD treat window=0 as invalid.
 ---
 ### 19.8 HMAC Mismatch Behavior — RESOLVED
 **Question:** What happens when HMAC keys don't match?
 **Commands:**
 ```
 irtt server -b 127.0.0.1:2120 --hmac=correctkey
 irtt client --hmac=wrongkey --timeouts=500ms 127.0.0.1:2120
 irtt client --timeouts=500ms 127.0.0.1:2120  # no HMAC
 ```
 **Observed behavior:**
 | Scenario | Server Log | Client Result |
 |----------|-----------|--------------|
 | Wrong HMAC key | `[Drop] [BadHMAC] invalid HMAC: ...` | `[OpenTimeout] no reply from server` (exit 1) |
 | No HMAC to HMAC server | `[Drop] [NoHMAC] no HMAC present` | `[OpenTimeout] no reply from server` (exit 1) |
 **Conclusion:** The server silently drops packets with missing or invalid
 HMAC. The client times out. This is **observed behavior** — the protocol does
 not define an explicit HMAC error response.
 ---
 ### 19.9 Minimum Restricted Interval Safety Floor — PARTIALLY RESOLVED
 **Question:** What safety floor does the client enforce for server-restricted
 intervals?
 **Test setup:** Server with -i 500ms (minimum interval), client requests
 200ms.
 **Command:**
 ```
 irtt server -b 127.0.0.1:2123 -i 500ms
 irtt client -d 2s -i 200ms 127.0.0.1:2123
 ```
 **Observed:** `[ServerRestriction] server increased interval from 200ms to
 500ms` — client exits with code 1 in strict mode. In loose mode, client
 accepts 500ms interval.
 **Limitation:** This test only confirms that the client rejects server-
 increased intervals in strict mode. It does not confirm the exact safety
 floor value for server-decreased intervals (which would require a server
 that decreases interval below the client's request). The safety floor for
 decreased intervals cannot be easily tested without a cooperative server
 that sends deliberately small intervals.
 **Conclusion:** Strict mode rejects any server parameter change by default.
 The specific safety floor for decreased intervals remains an **open
 question** that requires further testing.
 ---
 ### 19.10 Maximum Parameter Buffer Size — PARTIALLY RESOLVED
 **Question:** Is there a maximum serialized parameter payload size?
 **Test setup:** Open request with all parameters including a 24-character
 ServerFill string.
 **Observed:** Largest serialized parameter payload was 52 bytes (with
 ProtocolVersion, Duration, Interval, Length=1472, ReceivedStats=both,
 StampAt=both, Clock=both, DSCP=0xb8, ServerFill="pattern:aabbccddeeff0011").
 **Analysis:** Even with maximum-value parameters and the longest reasonable
 ServerFill string (32 bytes max), the parameter buffer stays well under
 128 bytes. A rough upper bound:
 - Tag+value overhead per param: ~6 bytes for large values
 - 8 params × 6 bytes = 48 bytes
 - Plus ServerFill: tag(1) + length(1) + string(32) = 34 bytes
 - Total: ~82 bytes maximum
 **Conclusion:** The 128-byte limit stated in the spec appears safe.
 Reclassified as **compatibility recommendation** — implementations should
 allocate at least 128 bytes for the parameter buffer.
 ---
 ### 19.11 Field Ordering Verification — RESOLVED
 **Question:** Is the field ordering in Section 8.1.3 correct?
 **Test setup:** Multiple captures with different field combinations.
 **Verified orderings:**
 1. **No optional fields (stats=none, tstamp=none):**
   Token → Seq. Packet = 16 bytes. ✓
 2. **Count + send wall:**
   Token → Seq → Recv Count → Send Wall. Packet = 28 bytes. ✓
 3. **Window + receive mono:**
   Token → Seq → Recv Window → Recv Mono. Packet = 32 bytes. ✓
 4. **Both + midpoint both:**
   Token → Seq → Recv Count → Recv Window → Midpoint Wall → Midpoint Mono.
   Packet = 44 bytes. ✓
 5. **Both + both both (default):**
   Token → Seq → Recv Count → Recv Window → Recv Wall → Recv Mono →
   Send Wall → Send Mono. Packet = 60 bytes. ✓
 6. **HMAC + both + both both:**
   HMAC → Token → Seq → Recv Count → Recv Window → Recv Wall → Recv Mono →
   Send Wall → Send Mono. Packet = 76 bytes. ✓
 **Conclusion:** The field ordering in Section 8.1.3 is correct. This is a
 **normative requirement** — incorrect ordering breaks interoperability.
 ---
 ### 19.3 Server Close During Test — NOT DIRECTLY TESTED
 **Limitation:** Triggering a server-initiated close mid-test requires
 specific server conditions that are difficult to set up reliably in a
 black-box test. The spec's description (server MAY set Close flag on echo
 reply) has not been verified externally. Reclassified as **open /
 verification required** — the spec now recommends defensive handling
 without asserting that this behavior occurs.
 ---
 ### 19.6 Send Timestamp Capture Timing — NOT TESTABLE
 **Limitation:** The exact timing of client-side timestamp capture relative
 to the send syscall is an implementation-internal detail that cannot be
 verified through black-box testing. Remains an implementation choice.
 ---
 ### 19.12 RTT When Server Processing > Raw Round-Trip — NOT DIRECTLY TESTED
 **Limitation:** This edge case requires server processing time to exceed
 the raw client-measured RTT, which is extremely unlikely on localhost. Would
 require a heavily loaded server or deliberate delays. Remains an
 **open question**.
 ---
 ## Additional Findings
 ### Finding A: Packet Count Formula Correction
 **Spec claim:** Expected packet count = `floor(d / i) + 1`
 **Test setup:** Multiple sessions with various duration/interval combinations.
 **Command:** `irtt client -d <D> -i <I> 127.0.0.1 -q`
 **Results:**
 | Duration | Interval | Observed Packets | floor(d/i)+1 | ceil(d/i) |
 |----------|----------|-----------------|-------------|-----------|
 | 10s | 200ms | 50 | 51 | 50 |
 | 1s | 200ms | 5 | 6 | 5 |
 | 1s | 333ms | 4 | 4 | 4 |
 | 1s | 500ms | 2 | 3 | 2 |
 | 1s | 1s | 1 | 2 | 1 |
 | 2s | 1s | 2 | 3 | 2 |
 | 500ms | 100ms | 5 | 6 | 5 |
 | 100ms | 10ms | 10 | 11 | 10 |
 | 999ms | 1s | 1 | 1 | 1 |
 | 1001ms | 1s | 2 | 2 | 2 |
 **Analysis:** When duration is an exact multiple of interval, the packet at
 exactly `start + duration` is NOT sent (exclusive end). The correct formula
 is:
 ```
 expected_packets = ceil(duration / interval)
 ```
 Or equivalently:
 ```
 if duration % interval == 0:
    expected_packets = duration / interval
 else:
    expected_packets = floor(duration / interval) + 1
 ```
 **Conclusion:** The spec's formula `floor(d / i) + 1` is incorrect. The
 correct formula is `ceil(d / i)`. The spec MUST be updated.
 ---
 ### Finding B: DSCP Parameter Encodes TOS Byte Value
 **Spec claim:** DSCP valid values are "0-63 (6-bit DSCP field)"
 **Test setup:** Captured packets with various --dscp values.
 **Results:**
 | CLI Option | TOS Byte in IP Header | DSCP (6-bit) | Param Value Encoded |
 |------------|----------------------|-------------|-------------------|
 | --dscp=0xb8 | 0xb8 | 46 (EF) | 184 |
 | --dscp=0x2e | 0x2e | 11 | 46 |
 | --dscp=46 | 0x2e | 11 | 46 |
 | --dscp=8 | 0x08 | 2 | 8 |
 | --dscp=0x20 | 0x20 | 8 (CS1) | 32 |
 **Conclusion:** The DSCP parameter encodes the raw value as specified on the
 CLI, which is used directly as the TOS/Traffic Class byte in the IP header.
 The range is 0-255, not 0-63. The spec MUST be updated to reflect that
 the parameter is the TOS byte value, not the 6-bit DSCP field.
 ---
 ### Finding C: DSCP Not Applied to Open/Close Packets
 **Observation:** In all captures, the IP TOS field for open request, open
 reply, and close request packets is always 0x00, regardless of the
 negotiated DSCP value. Only echo request and echo reply packets carry the
 negotiated DSCP/TOS value.
 **Verified with:** `captures/dscp-session.pcapng`
 **Conclusion:** This is **observed behavior**. It may not be a strict
 protocol requirement, but implementations should be aware that DSCP is only
 applied during the active test phase.
 ---
 ### Finding D: No-Test Mode Packet Exchange
 **Test setup:** Captured a no-test session.
 **Command:** `irtt client -n 127.0.0.1:2113`
 **Capture:** `captures/no-test.pcapng`
 **Observed:**
 Open request (frame 1):
 ```
 14 a7 5b 05 [params...]
 ```
 - Flags: `05` = Open (0x01) | Close (0x04)
 - No connection token
 Open reply (frame 2):
 ```
 14 a7 5b 07 00 00 00 00 00 00 00 00 [params...]
 ```
 - Flags: `07` = Open (0x01) | Reply (0x02) | Close (0x04)
 - Connection token: `00 00 00 00 00 00 00 00` (zero)
 **Conclusion:** In no-test mode:
 - Client sets both Open and Close flags in the open request.
 - Server replies with Open|Reply|Close and a **zero** connection token.
 - The client treats this as successful completion, not rejection.
 - The Close flag in the open reply is **not** a rejection when the client
  originally requested Close.
 The spec's description of "An open reply with the close flag set: the
 server rejected the session" needs clarification. The Close flag in the
 reply indicates rejection ONLY when the client did not request Close in its
 open request.
 ---
 ### Finding E: Test Vector Size Correction
 The test vectors in Section 18 are described as "target length 256 bytes
 (shown truncated)." Independent HMAC verification confirms that both test
 vectors (18.1 and 18.2) were computed over 92-byte packets (76-byte header
 + 16-byte payload), not 256-byte packets. The "256 bytes" description is
 incorrect.
 ---
 ### Finding F: Default Server Fill Pattern
 **Observation:** When no server fill is requested, the server fills reply
 payloads with the repeating hex pattern `69 72 74 74` (ASCII "irtt"). This
 is the default server fill as documented in server help:
 `--fill=fill  payload fill if not requested (default pattern:69727474)`.
 ---
 ### Finding G: Close Packet Minimal Format
 **Observed close packet:**
 ```
 14 a7 5b 04 [8-byte connection token]
 ```
 Total: 12 bytes. Contains only magic (3) + flags (1) + connection token (8).
 No sequence number, no payload.
 With HMAC:
 ```
 14 a7 5b 0c [16-byte HMAC] [8-byte connection token]
 ```
 Total: 28 bytes.
 ---
 ### Finding H: Unreachable Server Behavior
 **Command:** `irtt client --timeouts=200ms,200ms 127.0.0.1:2121` (no server)
 **Observed:** `Error: read udp4 ... connection refused` (exit 1)
 On localhost, the OS delivers ICMP port unreachable immediately. On remote
 hosts, the behavior would be a timeout after all retransmissions are
 exhausted.
 ---
 ### Finding I: Parameter Restriction Behavior
 **Commands:**
 ```
 irtt server -b 127.0.0.1 -l 100    # max length 100
 irtt client -l 200 127.0.0.1       # request length 200
 ```
 **Strict mode (default):** `[ServerRestriction] server reduced length from
 200 to 100` — exit code 1.
 **Loose mode (--loose):** `[ServerRestriction] server reduced length from
 200 to 100` — continues with length 100, exit code 0.
 ---
 ## Captures Created
 | File | Contents |
 |------|----------|
 | `captures/full-session.pcapng` | Complete session: open, 3 echoes, close |
 | `captures/basic-session.pcapng` | Basic session (partial, missed open) |
 | `captures/no-test.pcapng` | No-test mode: open+close only |
 | `captures/hmac-session.pcapng` | HMAC authenticated session |
 | `captures/tstamp-none.pcapng` | Multiple sessions with different timestamp modes |
 | `captures/dscp-session.pcapng` | DSCP=0xb8 session |
 | `captures/dscp-46.pcapng` | DSCP=46 and DSCP=0x2e sessions |
 | `captures/dscp-values.pcapng` | DSCP=8 and DSCP=0x20 sessions |
 | `captures/large-packet.pcapng` | 1472-byte packet session |
 | `captures/long-sfill.pcapng` | Session with long server fill string |
 ## Test Vectors Created
 See `test-vectors/` directory.
 ## Summary
 ### Resolved Open Questions
 - **19.1** Open request field layout (no token)
 - **19.2** Minimum packet length calculation
 - **19.4** Varint encoding compatibility
 - **19.5** HMAC computation scope
 - **19.7** Received window validity (bit 0 sentinel)
 - **19.8** HMAC mismatch behavior
 - **19.11** Field ordering verification
 ### Partially Resolved
 - **19.9** Minimum restricted interval safety floor (strict mode rejects,
  but exact floor for decreased intervals untested)
 - **19.10** Maximum parameter buffer size (observed well under 128 bytes)
 ### Still Unresolved
 - **19.3** Server close during test (hard to trigger)
 - **19.6** Send timestamp capture timing (implementation detail, not testable)
 - **19.12** RTT when server processing exceeds raw RTT (hard to trigger)
 ### Spec Corrections Applied
 1. Packet count formula: `ceil(d/i)` not `floor(d/i)+1`
 2. DSCP parameter: TOS byte value (0-255) not 6-bit DSCP (0-63)
 3. Test vector size: 92 bytes not 256 bytes
 4. No-test mode close flag: clarified in Sections 6.2, 8.3, and 14
 5. Server close during test (19.3): reclassified as open/verification
   required; removed source-derived claims; spec now recommends
   defensive handling without asserting the behavior occurs
 6. Received window late-packet behavior: removed unverified claim about
   late packets producing window=0; retained verified bit-0 sentinel
@@ -0,0 +1,189 @@
 # Clean-Room Implementation Notes
 ## Purpose
 This document records the clean-room boundary for the IRTT client
 reimplementation project. The goal is to ensure that the implementation
 is developed from a behavioral/protocol specification only, without exposure
 to the GPL-licensed upstream source code.
 ## Clean-Room Boundary
 - **Contaminated side:** The specification agent (this side) inspected the
  upstream source code, documentation, tests, CLI behavior, and release
  notes to understand the IRTT protocol.
 - **Clean side:** The implementation agent will receive only the
  `IRTT_CLIENT_PROTOCOL_SPEC.md` document. The implementation agent MUST NOT
  see the upstream source code.
 ## Source Materials Inspected
 The following materials from the upstream IRTT repository (heistp/irtt,
 GPLv2) were inspected to produce the protocol specification:
 1. **Documentation files:**
   - `README.md` — project overview, FAQ, features, limitations
   - `doc/irtt.md` — main man page
   - `doc/irtt-client.md` — client man page with CLI options and JSON output format
   - `doc/irtt-server.md` — server man page (referenced, not fully read)
   - `CHANGES.md` — changelog
 2. **Source files (inspected for protocol behavior only):**
   - Protocol version and constants
   - Packet format, field layout, magic bytes, flags
   - Parameter serialization format (tag-value pairs, varint encoding)
   - Connection lifecycle (open, echo, close handshakes)
   - Client configuration and defaults
   - Server connection handling and parameter restriction logic
   - HMAC computation
   - Timestamp handling and clock types
   - Received stats (count and window) format
   - Round-trip data recording and measurement calculations
   - Result computation (RTT, OWD, IPDV, loss)
   - Error codes and error handling
   - Packet test vectors (serialized bytes for known inputs)
   - Network layer (IP version, DSCP, socket options)
   - Wait/drain behavior
 3. **Test files:**
   - Packet serialization tests (used to derive test vectors only; test
     code itself was not copied)
 ## Statement of Non-Copying
 No upstream source code, comments, or pseudocode were copied into the
 specification document (`IRTT_CLIENT_PROTOCOL_SPEC.md`). The specification
 describes only externally observable protocol behavior, wire format, and
 interoperability requirements.
 Test vectors in the specification were derived from the known inputs and
 expected byte sequences found in the upstream test suite. Only the input
 parameters and resulting byte arrays were used; no test logic or code was
 copied.
 ## Post-Drafting Audit
 After the initial specification was drafted, a clean-room audit was
 performed. The following issues were identified and corrected:
 1. **Upstream private type names leaked:** "ctoken" and "seqno" appeared
   as parenthetical abbreviations in the terminology table. Removed.
 2. **Upstream implementation preference disclosed:** A note about IPv4
   preference in dual-stack resolution revealed an upstream implementation
   choice. Removed.
 3. **Go-specific function references:** References to `binary.PutUvarint`,
   `binary.PutVarint`, and `encoding/binary` package named the upstream
   language's standard library. Replaced with language-neutral descriptions
   (LEB128, protobuf-style zigzag).
 4. **Internal scheduling algorithm described:** The send scheduling section
   described a specific "snap-to-interval" drift-correction algorithm that
   mirrored upstream control flow. Replaced with observable behavior
   description (packets sent at approximately ideal times, drift
   compensation is implementation-defined).
 5. **Upstream timer algorithm named:** "The upstream uses exponential
   averaging of timer errors" revealed an internal implementation choice.
   Removed.
 6. **RTT edge case speculated:** The spec asserted that server processing
   time exceeding raw RTT should cause subtraction to be skipped — a guard
   the upstream does not actually implement. Moved to Open Questions.
 7. **Received window validity asserted normatively:** The behavior of
   window=0 as an invalidity sentinel was stated as fact when it should be
   verification-required. Softened with reference to Open Questions.
 8. **Go concurrency term leaked:** "goroutine" appeared in the result model
   section. Replaced with "operation."
 9. **Upstream test design referenced:** Section 18.2 described what "the
   upstream test" was testing internally. Rewritten to describe the test
   vector neutrally.
 10. **Upstream constants asserted normatively:** The minimum restricted
    interval (1 second) and maximum parameter buffer size (128 bytes) are
    internal constants that were stated as protocol requirements. Moved to
    Open Questions with verification suggestions.
 11. **Timestamp capture order stated as MUST:** Implementation-level timing
    advice was stated as a protocol requirement. Changed to SHOULD with
    measurement-accuracy rationale.
 12. **Internal recording structure described:** Section 19.6 described a
    "pre-send timestamp" used for send-call-time measurement, which is an
    internal recording detail. Rewritten to focus on observable behavior.
 13. **Self-contradictory correction note:** Section 8.2 contained an inline
    "Correction:" note that was confusing and revealed the drafting process.
    Cleaned up.
 14. **Field ordering lacked verification note:** The critical field order
    in Section 8.1.3 was derived from source and stated without any
    verification caveat. Added verification note and new Open Question 19.11.
 ## Clean-Room Compliance Checklist
 - [x] No upstream source code included.
 - [x] No upstream comments quoted.
 - [x] No upstream private function names included.
 - [x] No upstream private type names included.
 - [x] No upstream file/module layout described.
 - [x] No pseudocode derived from upstream implementation.
 - [x] No implementation-language API design included.
 - [x] No Go-specific language references remain.
 - [x] No upstream implementation preferences or algorithm choices disclosed.
 - [x] Protocol behavior described independently.
 - [x] Source-derived constants marked for verification.
 - [x] Ambiguous behavior marked for verification (Section 19).
 - [x] Black-box tests proposed where needed (Section 17).
 - [x] Post-drafting audit completed and documented (this section).
 - [x] Second-pass scrub completed (see below).
 ## Second-Pass Clean-Room Scrub
 A second independent review was performed on 2026-04-28 by an agent that
 had not seen the upstream source. The following issues were identified and
 corrected:
 1. **Implementation architecture directive in scope:** Section 2 stated
   "Library-first implementation target," which prescribes internal
   architecture. Removed; the spec now says only that CLI design is not
   prescribed.
 2. **Concurrency architecture leaked:** Section 6.3 described "a separate
   receive path processes incoming reply packets concurrently," which
   prescribes an internal threading/task model. Rewritten to state the
   observable requirement (send and receive occur concurrently) without
   prescribing how.
 3. **Source-derived design rationale in echo request:** Section 8.4
   explained that zeroed fields exist to "use the same field layout that
   appears in echo replies," which reveals internal design reasoning.
   Simplified to state only the observable purpose (reaching the
   negotiated packet length).
 4. **Source-exhaustive knowledge parenthetical:** Section 8.6 said
   "(currently only ServerFill)" for string-typed values, revealing
   knowledge of the complete parameter set from source inspection.
   Removed the parenthetical.
 5. **Upstream CLI syntax leaked:** Section 10.10 used `NxD`/`NrD`/`D`
   notation for wait variants, which mirrors upstream CLI option syntax.
   Replaced with plain-language descriptions.
 6. **Algorithm name disclosed:** Section 12.9 named "Welford's online
   algorithm" as a MAY recommendation. This is implementation guidance.
   Replaced with "the method of computing running statistics is
   implementation-defined."
 7. **Implementation-specific error category:** Section 14 listed
   "Allocate results buffer failure," implying a pre-allocated buffer
   strategy. Generalized to "Insufficient resources for test parameters."
 8. **Language-specific references in CLEANROOM_NOTES.md:** The Purpose
   section named "the Rust implementation" and the checklist mentioned
   "No Rust API design." Both replaced with language-neutral phrasing.
@@ -0,0 +1,304 @@
 # IRTT Packet Test Vectors
 **Source:** Black-box captures from irtt 0.9.1 on macOS Darwin 25.3.0 arm64  
 **Date:** 2026-04-28
 All vectors are from actual network captures (not constructed).
 ---
 ## Vector 1: Open Request (no HMAC)
 **Scenario:** Client opens session with default parameters (duration=3s,
 interval=1s, stats=both, tstamp=both, clock=both).  
 **Direction:** Client → Server  
 **irtt version:** 0.9.1  
 **Command:** `irtt client -d 3s -i 1s 127.0.0.1:2112`
 **Packet (24 bytes):**
 ```
 14 a7 5b 01 01 02 02 80 f8 82 ad 16 03 80 a8 d6
 b9 07 05 06 06 06 07 06
 ```
 **Field breakdown:**
 - Bytes 0-2: Magic `14 a7 5b`
 - Byte 3: Flags `01` (Open)
 - Bytes 4-23: Parameter payload:
  - Tag 1 (ProtocolVersion): value `02` → zigzag → 1
  - Tag 2 (Duration): value `80 f8 82 ad 16` → zigzag → 3000000000 ns (3s)
  - Tag 3 (Interval): value `80 a8 d6 b9 07` → zigzag → 1000000000 ns (1s)
  - Tag 5 (ReceivedStats): value `06` → zigzag → 3 (Both)
  - Tag 6 (StampAt): value `06` → zigzag → 3 (Both)
  - Tag 7 (Clock): value `06` → zigzag → 3 (Both)
 **Notes:**
 - No connection token field (token not yet assigned)
 - Tags 4 (Length), 8 (DSCP) omitted because values are 0
 - Tag 9 (ServerFill) omitted because not specified
 ---
 ## Vector 2: Open Reply (no HMAC)
 **Scenario:** Server accepts session, returns token and echoed parameters.  
 **Direction:** Server → Client  
 **Token:** `0x7896b6ab87715213`
 **Packet (32 bytes):**
 ```
 14 a7 5b 03 13 52 71 87 ab b6 96 78 01 02 02 80
 f8 82 ad 16 03 80 a8 d6 b9 07 05 06 06 06 07 06
 ```
 **Field breakdown:**
 - Bytes 0-2: Magic `14 a7 5b`
 - Byte 3: Flags `03` (Open | Reply)
 - Bytes 4-11: Connection token `13 52 71 87 ab b6 96 78` (LE: 0x7896b6ab87715213)
 - Bytes 12-31: Parameter payload (same as request — no restrictions)
 ---
 ## Vector 3: Echo Request (no HMAC, default stats+timestamps)
 **Scenario:** First test packet, sequence 0, default settings.  
 **Direction:** Client → Server  
 **Token:** `0x7896b6ab87715213`, Seq: 0
 **Packet (60 bytes):**
 ```
 14 a7 5b 00 13 52 71 87 ab b6 96 78 00 00 00 00
 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
 00 00 00 00 00 00 00 00 00 00 00 00
 ```
 **Field breakdown:**
 - Bytes 0-2: Magic `14 a7 5b`
 - Byte 3: Flags `00` (no flags — echo request from client)
 - Bytes 4-11: Connection token
 - Bytes 12-15: Sequence number `00 00 00 00` (LE: 0)
 - Bytes 16-19: Received count (zeroed placeholder)
 - Bytes 20-27: Received window (zeroed placeholder)
 - Bytes 28-35: Receive wall timestamp (zeroed placeholder)
 - Bytes 36-43: Receive mono timestamp (zeroed placeholder)
 - Bytes 44-51: Send wall timestamp (zeroed placeholder)
 - Bytes 52-59: Send mono timestamp (zeroed placeholder)
 **Notes:** All stats and timestamp fields are zeroed in echo requests.
 They serve as placeholders to reach the negotiated packet length.
 ---
 ## Vector 4: Echo Reply (no HMAC, default stats+timestamps)
 **Scenario:** Server reply to seq 2, third packet in session.  
 **Direction:** Server → Client  
 **Token:** `0x7896b6ab87715213`, Seq: 2, Recv Count: 3, Window: 0x07
 **Packet (60 bytes):**
 ```
 14 a7 5b 02 13 52 71 87 ab b6 96 78 02 00 00 00
 03 00 00 00 07 00 00 00 00 00 00 00 b8 1a 33 0c
 86 6d aa 18 de 26 35 95 00 00 00 00 80 4d 33 0c
 86 6d aa 18 b2 57 35 95 00 00 00 00
 ```
 **Field breakdown:**
 - Bytes 0-2: Magic `14 a7 5b`
 - Byte 3: Flags `02` (Reply)
 - Bytes 4-11: Connection token
 - Bytes 12-15: Sequence number `02 00 00 00` (LE: 2)
 - Bytes 16-19: Received count `03 00 00 00` (LE: 3)
 - Bytes 20-27: Received window `07 00 00 00 00 00 00 00` (LE: 0x07, bits 0-2 set)
 - Bytes 28-35: Receive wall timestamp `b8 1a 33 0c 86 6d aa 18` (LE int64)
 - Bytes 36-43: Receive mono timestamp `de 26 35 95 00 00 00 00` (LE int64)
 - Bytes 44-51: Send wall timestamp `80 4d 33 0c 86 6d aa 18` (LE int64)
 - Bytes 52-59: Send mono timestamp `b2 57 35 95 00 00 00 00` (LE int64)
 **Window interpretation (for seq 2):**
 - Bit 0 (0x01): Seq 2 received ✓
 - Bit 1 (0x02): Seq 1 received ✓
 - Bit 2 (0x04): Seq 0 received ✓
 ---
 ## Vector 5: Close Request (no HMAC)
 **Scenario:** Client sends close after test.  
 **Direction:** Client → Server  
 **Token:** `0x7896b6ab87715213`
 **Packet (12 bytes):**
 ```
 14 a7 5b 04 13 52 71 87 ab b6 96 78
 ```
 **Field breakdown:**
 - Bytes 0-2: Magic `14 a7 5b`
 - Byte 3: Flags `04` (Close)
 - Bytes 4-11: Connection token
 ---
 ## Vector 6: HMAC Open Request
 **Scenario:** Client opens session with HMAC key "testkey".  
 **Direction:** Client → Server  
 **HMAC key:** ASCII "testkey" (7 bytes)
 **Packet (40 bytes):**
 ```
 14 a7 5b 09 ff 90 16 a7 aa 53 78 16 9e c3 a2 d5
 54 dc 30 36 01 02 02 80 d0 ac f3 0e 03 80 a8 d6
 b9 07 05 06 06 06 07 06
 ```
 **Field breakdown:**
 - Bytes 0-2: Magic `14 a7 5b`
 - Byte 3: Flags `09` (Open | HMAC)
 - Bytes 4-19: HMAC `ff 90 16 a7 aa 53 78 16 9e c3 a2 d5 54 dc 30 36`
 - Bytes 20-39: Parameter payload
 **HMAC verified independently with Python (HMAC-MD5 over entire packet
 with HMAC field zeroed).**
 ---
 ## Vector 7: HMAC Echo Request
 **Scenario:** Echo request with HMAC, seq 0.  
 **Direction:** Client → Server  
 **Token:** `0x4387e9eb5d3fca59`, HMAC key: "testkey"
 **Packet (76 bytes):**
 ```
 14 a7 5b 08 d9 87 4f 82 b3 13 10 31 59 ad 2c 8f
 6b da ef ff 59 ca 3f 5d eb e9 87 43 00 00 00 00
 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
 00 00 00 00 00 00 00 00 00 00 00 00
 ```
 **Field breakdown:**
 - Bytes 0-2: Magic
 - Byte 3: Flags `08` (HMAC)
 - Bytes 4-19: HMAC
 - Bytes 20-27: Connection token `59 ca 3f 5d eb e9 87 43`
 - Bytes 28-31: Sequence number `00 00 00 00`
 - Bytes 32-75: Stats + timestamp placeholders (zeroed)
 ---
 ## Vector 8: HMAC Close Request
 **Scenario:** Close with HMAC.  
 **Direction:** Client → Server  
 **Token:** `0x4387e9eb5d3fca59`, HMAC key: "testkey"
 **Packet (28 bytes):**
 ```
 14 a7 5b 0c f5 cd 0f a9 de 9d 7d 66 8d 91 a5 32
 48 0e 42 e0 59 ca 3f 5d eb e9 87 43
 ```
 **Field breakdown:**
 - Bytes 0-2: Magic
 - Byte 3: Flags `0c` (Close | HMAC)
 - Bytes 4-19: HMAC
 - Bytes 20-27: Connection token
 ---
 ## Vector 9: No-Test Mode (Open+Close Request)
 **Scenario:** Client opens and immediately closes.  
 **Direction:** Client → Server
 **Packet (25 bytes):**
 ```
 14 a7 5b 05 01 02 02 80 e0 ba 84 bf 03 03 80 a8
 d6 b9 07 05 06 06 06 07 06
 ```
 **Field breakdown:**
 - Bytes 0-2: Magic
 - Byte 3: Flags `05` (Open | Close)
 - Bytes 4-24: Parameter payload
  - Tag 1: ProtocolVersion = 1
  - Tag 2: Duration = 60000000000 ns (60s, default)
  - Tag 3: Interval = 1000000000 ns (1s, default)
  - Tag 5: ReceivedStats = 3 (Both)
  - Tag 6: StampAt = 3 (Both)
  - Tag 7: Clock = 3 (Both)
 ---
 ## Vector 10: No-Test Mode (Open+Close Reply)
 **Scenario:** Server acknowledges no-test open+close.  
 **Direction:** Server → Client
 **Packet (33 bytes):**
 ```
 14 a7 5b 07 00 00 00 00 00 00 00 00 01 02 02 80
 e0 ba 84 bf 03 03 80 a8 d6 b9 07 05 06 06 06 07
 06
 ```
 **Field breakdown:**
 - Bytes 0-2: Magic
 - Byte 3: Flags `07` (Open | Reply | Close)
 - Bytes 4-11: Connection token `00 00 00 00 00 00 00 00` (ZERO)
 - Bytes 12-32: Parameter payload
 **Notes:** In no-test mode, the server returns a zero connection token.
 The Close flag in this context means "acknowledged close" not "rejected."
 ---
 ## Vector 11: Minimal Echo Packet (no stats, no timestamps)
 **Scenario:** Echo request with stats=none, tstamp=none.  
 **Direction:** Client → Server  
 **Token:** `0xa0316fa25c61154e`, Seq: 0
 **Packet (16 bytes):**
 ```
 14 a7 5b 00 4e 15 61 5c a2 6f 31 a0 00 00 00 00
 ```
 **Field breakdown:**
 - Bytes 0-2: Magic
 - Byte 3: Flags `00`
 - Bytes 4-11: Connection token
 - Bytes 12-15: Sequence number
 This is the absolute minimum echo packet size.
 ---
 ## Vector 12: Midpoint Timestamps (both clocks)
 **Scenario:** Echo reply with stats=both, tstamp=midpoint, clock=both.  
 **Direction:** Server → Client
 **Packet (44 bytes):**
 ```
 14 a7 5b 02 91 a4 1e 7a f0 14 f6 62 00 00 00 00
 01 00 00 00 01 00 00 00 00 00 00 00 f7 0e 2c 93
 a2 6d aa 18 5a 5f 14 47 03 00 00 00
 ```
 **Field breakdown:**
 - Bytes 0-2: Magic
 - Byte 3: Flags `02` (Reply)
 - Bytes 4-11: Token
 - Bytes 12-15: Seq `00 00 00 00` (0)
 - Bytes 16-19: Recv count `01 00 00 00` (1)
 - Bytes 20-27: Recv window `01 00 00 00 00 00 00 00`
 - Bytes 28-35: Midpoint wall timestamp
 - Bytes 36-43: Midpoint mono timestamp
 **Notes:** Only midpoint timestamps present. No receive/send timestamp
 fields. Confirms midpoint is mutually exclusive with receive+send.