Reamer Server changelog

Format: Keep a Changelog. Every entry below describes a customer-visible change to the Reamer Server product — the ABI, the shipped reference material, or the kit contents — not internal refactors. Updated at release, alongside the tag.

[4.5.0] - 2026-10-09

ABI bump: REAMER_ABI_VERSION 3 → 4. Additive. Every ABI 3 field keeps its type, name and offset, the vtables are unchanged, and a strategy or relay that sends ABI 3 frames needs no change. Rebuild your gate and connector against the ABI 4 header to read the new fields.

Added

• Order fields on ReamerIntent, appended after expire_time: attached exits (take_profit, stop_loss, stop_loss_limit, stop_loss_trailing_offset), trailing_offset, offset_type (price, basis points, ticks), limit_offset, trigger_type (last, bid/ask, mid, mark, index), display_quantity (iceberg), min_quantity, exec_flags (post-only, reduce-only, hidden, all-or-none), contingency_type (OCO, OTO, OUO) with linked_intent_id, account and destination.
• User-defined payload on ReamerIntent: payload_schema, payload_len and payload, up to REAMER_MAX_PAYLOAD_LEN (1024) bytes that core passes to your gate and connector untouched.
• ReamerReplaceRequest gains new_take_profit, new_stop_loss and new_trailing_offset.
• REAMER_SIDE_SELL_SHORT; order types TRAILING_STOP, TRAILING_STOP_LIMIT, MARKET_IF_TOUCHED, LIMIT_IF_TOUCHED, MARKET_TO_LIMIT; time in force AT_THE_OPEN and AT_THE_CLOSE.
• Exit tracking. An accepted order with take_profit or stop_loss set makes core track its exits as orders <intent_id>:tp and <intent_id>:sl, so the connector's reports for them reach the strategy and the strategy can cancel or replace them by id. The connector places the exits at the venue.
• Both strategy wire protocols carry the new fields in an optional extension block at the end of the NewOrder and Replace frames. A frame without it decodes as before. EXTENSION_PROTOCOL.md has the layout.
• Reference relay: JSON fields for every new field (payload as base64) and the new side, order type and time-in-force spellings. The reference gate rejects an order that uses a field the reference FIX broker does not implement.
• Rust skeleton: ABI 4 types.

This release also carries the Reamer Research ABI 7 changes (see the Reamer Research changelog), including reamer_relay_*, a strategy relay that sends research orders, brackets included, to this server's local-mode strategy socket.

[4.4.0] - 2026-10-09

No change to the Reamer Server binary or ABI. reamer_server_abi.h remains at ABI 3. The version moves with the shared release number; this release carries the Reamer Research ABI 6 changes (see the Reamer Research changelog).

[4.3.2] - 2026-10-03

Fixed

• license.expiring fired every 30 seconds in the last 14 days of a term. The run loop remembered only the last threshold it emitted, so once two thresholds had been crossed it alternated between them on every recheck (30, 14, 30, 14, …). Each license_warning_days threshold now fires exactly once, a process started inside several thresholds fires only the smallest, and the thresholds may be listed in any order, as CONFIG.md and MONITORING.md already described.

Added

• LICENSING.md "Renew on the same machine": deactivate the old key, then activate the new one, and the one license.expired a running server may emit between the two.

[4.3.1] - 2026-09-17

No change to the Reamer Server binary or ABI. reamer_server_abi.h remains at ABI 3 and the core library is byte-identical to 4.3.0. This release corrects documentation figures against the raw EPYC 9575F artifacts and adds a versioning policy document.

Fixed

• BENCHMARK.md figures reconciled with the raw EPYC 9575F data. The concurrent-strategy sweep's peak-throughput row was labeled N=32; the 219,652 orders/sec peak is at N=34 (true N=32 is 218,116). The pinning section's run-to-run spread is corrected to 1.28x unpinned collapsing to 1.02x pinned (was stated approximately as 1.30x to 1.01x), the raw N=62 runs are the measured values, and the cpu-migration reduction is 118 to 63 (47 percent, was stated as ~40 percent). The median pinning gain is reframed as load-dependent and not bankable, with the variance collapse as the reliable claim.

Added

• A written ABI versioning policy for the server ABI. It states the contract: the surface only grows, and a header and library version mismatch is caught at compile time by the abi_layout_guard static_asserts rather than by a runtime fallback. It records the git-evidenced version history and sets support to the latest version only, communicated through this changelog at release.

[4.3.0] - 2026-09-13

No change to the Reamer Server binary or ABI. reamer_server_abi.h remains at ABI 3 and the core library is byte-identical to 4.2.0. This release carries new benchmark evidence and the documentation built on it; the version moves with the shared release number (kReamerVersion, one number across both products).

Changed

• BENCHMARK.md now includes a many-core verification section measured on a 64-core AMD EPYC 9575F. It confirms the concurrent-strategy latency elbow tracks physical core count — previously stated as an expectation to verify, now backed by data (clean-latency region and throughput plateau extend to roughly one strategy per physical core; peak plateau ~200–220k orders/s). The ABI-only end-to-end P99 on this hardware is ~12µs, so the earlier laptop-only "P99 occasionally exceeds 100µs under burst" caveat does not reproduce there.
• New tuning guidance from the same run: the event ring buffer is a cache-fit knob (the shipped 16 MB default sits at the measured throughput peak; sizing it up costs throughput, it does not add it), and pinning strategy threads to cores is a run-to-run repeatability lever (collapses unpinned throughput spread from ~1.3× to ~1.0×), not a throughput lever.

[4.2.1] - 2026-09-12

No change to the Reamer Server binary. The version moves with the shared release number (kReamerVersion, one number across both products) for a fix that is entirely inside Reamer Research's Python reference binding — see that product's changelog. reamer_server_abi.h remains at ABI 3, the core library is byte-identical to 4.2.0, and only the kit's version stamp and the server-kit-<version> filename change.

Changed

• The kit now ships SECURITY_HARDENING.md in place of the security-review statement of work. The statement of work was a forward-looking scope for an independent external firm review; the hardening log is the record of the internal adversarial pass already performed and the memory-safety and license-path defects it found and fixed, each with sanitizer and regression-test verification.

[4.2.0] - 2026-09-09

The kit now ships reference/, a working reference spine, alongside the existing shape-only skeletons. It is a FIX 4.4 session, a broker connector, a strategy relay and an event-bus reader, exercised by its own suite (96 tests, 0 failures) against ABI 3. demo-trade.sh takes the extracted kit to a filled order over real FIX in about 13 seconds, including the Go build.

The two tiers answer different questions: the skeletons confirm the shape of reamer_server_run() wiring; the spine shows the whole path once it closes. Neither is required to reach a running server, and neither is a venue adapter for your venue — the header plus EXTENSION_PROTOCOL.md remain the complete integration contract.

Fixed in the reference connector: poll_published_events returned events read off the shared event bus, which core correctly rejected as echoes (connector.echoed_event). That callback is inbound only. It now drains a bounded queue of the connector's own transport diagnostics, and a FIX session drop — invisible to core, since it happens on a socket below the ABI — is surfaced as SESSION_DISCONNECTED on the bus. To consume core's stream instead, attach a reader directly, as cmd/event-tail shows.

reamer_server_abi.h remains at ABI 3, and the core library is unchanged.

[4.1.14] - 2026-09-06

Reamer Research and Reamer Server are licensed separately. A licence key now activates only its own product's binaries: the product is compiled into each kit, carried inside the signed grant, and checked against the minted record at activation, so a Research key cannot activate a Server build.

Activation is per machine and per product. ~/.config/reamer/activation.json holds one grant per product, so activating Server on a machine that already runs Research leaves the Research grant intact. A machine running both consumes one seat of each.

Kits built before 4.1.14 cannot activate against the current licence server, and 4.1.14 kits cannot use a pre-4.1.14 activation file. This is a coordinated cutover: re-cut both kits from this tag and re-activate each machine. Any existing activation.json must be discarded — the old flat format is treated as absent, not migrated.

DEPLOYMENT.md no longer documents copying an activation.json between hosts to license container replicas. Every licensed machine runs activate itself; a copied grant fails on any host that does not reproduce the original's fingerprint. Containerised deployments are documented as a single licensed machine.

reamer_server_abi.h remains at ABI 3, and the core library is unchanged.

[4.1.13] - 2026-09-05

reamer_server_abi.h remains at its own, independently versioned ABI 3, and the core library and its reference material are unchanged.

Added

• LICENSE.txt ships in the kit. The archive previously carried no licence terms at all. The file sits in the kit root and is covered by SHA256SUMS.

• Activation requires accepting the terms. reamer-license activate now displays LICENSE.txt and requires an explicit accept before it contacts the licence server. Declining consumes no seat, and the acceptance record is written only after activation succeeds. Because activation was then machine-global, accepting once covered every Reamer Labs product on that machine (activation became per product in 4.1.14).

  For unattended provisioning set REAMER_ACCEPT_LICENSE=accept. Activation with no terminal and no such variable fails rather than defaulting to acceptance, so existing non-interactive provisioning that activates a licence must set it.

  reamer-license reads LICENSE.txt from the kit root, one level above bin/. Moving bin/ out of the extracted tree makes activation fail with a message naming the cause.

Earlier releases, 3.0.0 through 4.1.12, are listed in CHANGELOG.md in the kit.