diff --git a/Doxyfile b/Doxyfile
index 4715b0d..2c871b0 100644
--- a/Doxyfile
+++ b/Doxyfile
@@ -962,7 +962,9 @@ INPUT                  = ./include \
                          ./docs/OtlpHttpLogger.md \
                          ./docs/PrometheusLogger.md \
                          ./docs/TaskExecutor.md \
-                         ./docs/backpressure.md
+                         ./docs/backpressure.md \
+                         ./docs/future-plans.md \
+                         ./docs/adr
 
 # This tag can be used to specify the character encoding of the source files
 # that doxygen parses. Internally doxygen uses the UTF-8 encoding. Doxygen uses
diff --git a/README-RU.md b/README-RU.md
index 8a4ff05..a94128c 100644
--- a/README-RU.md
+++ b/README-RU.md
@@ -61,6 +61,8 @@ scope-замер.
 - [`docs/PrometheusLogger.md`](docs/PrometheusLogger.md) — payload/server-бэкенды, registry, scrape и ограничения.
 - [`docs/TaskExecutor.md`](docs/TaskExecutor.md) — варианты очереди, политики переполнения, hot resize и lifecycle.
 - [`docs/backpressure.md`](docs/backpressure.md) — настройка очереди и счётчики отброшенных задач.
+- [`docs/future-plans.md`](docs/future-plans.md) — чеклист roadmap с разделением готовых, планируемых и отложенных задач.
+- [`docs/adr/README.md`](docs/adr/README.md) — архитектурные решения, границы публичного API и записи о совместимости.
 
 ## Примеры макросов
 
diff --git a/README.md b/README.md
index f41aaa5..9553ac4 100644
--- a/README.md
+++ b/README.md
@@ -77,6 +77,8 @@ Detailed guides and documentation map:
 - [`docs/PrometheusLogger.md`](docs/PrometheusLogger.md) — payload/server backends, registry metrics, scrape configuration, and limitations.
 - [`docs/TaskExecutor.md`](docs/TaskExecutor.md) — queue variants, overflow policies, hot resize, and lifecycle guarantees.
 - [`docs/backpressure.md`](docs/backpressure.md) — application-facing queue tuning and drop counters.
+- [`docs/future-plans.md`](docs/future-plans.md) — checked roadmap separating delivered, planned, and deferred work.
+- [`docs/adr/README.md`](docs/adr/README.md) — architecture decisions, public-boundary rationale, and compatibility records.
 
 ## Macro Examples
 
diff --git a/docs/adr/0000-template.md b/docs/adr/0000-template.md
new file mode 100644
index 0000000..42c08e0
--- /dev/null
+++ b/docs/adr/0000-template.md
@@ -0,0 +1,24 @@
+# ADR NNNN: Short decision title
+
+- Status: Proposed
+- Date: YYYY-MM-DD
+
+## Context
+
+What problem, constraint, or repeated failure mode requires a decision?
+
+## Decision
+
+State the chosen approach and its public contract precisely.
+
+## Consequences
+
+Describe benefits, costs, compatibility impact, and operational trade-offs.
+
+## Alternatives considered
+
+List the credible alternatives and why they were not selected.
+
+## References
+
+Link the implementation, tests, documentation, issues, or superseding ADRs.
diff --git a/docs/adr/0001-aggregate-first-public-headers.md b/docs/adr/0001-aggregate-first-public-headers.md
new file mode 100644
index 0000000..2666675
--- /dev/null
+++ b/docs/adr/0001-aggregate-first-public-headers.md
@@ -0,0 +1,31 @@
+# ADR 0001: Aggregate-first public headers and API boundary
+
+- Status: Accepted
+- Date: 2026-09-14
+
+## Context
+
+The library has many historical leaf headers, including
+`logit/log_macros.hpp`. Treating every header as independently includable made
+include order accidental and made internal implementation details look public.
+
+## Decision
+
+`<logit.hpp>` is the supported application entry point. The module umbrellas
+`<logit/utils.hpp>`, `<logit/formatter.hpp>`, and `<logit/loggers.hpp>` are
+supported focused entry points. Leaf headers are standalone only when their
+documentation and include-contract test explicitly say so. `logit/log_macros.hpp`
+is an aggregate-owned implementation header and has no standalone-inclusion
+guarantee. Public aliases, including `logit::QueuePolicy`, are consumed through
+the umbrella or the relevant module umbrella.
+
+## Consequences
+
+The include contract is predictable and can evolve without promising every
+historical file as an API boundary. Documentation and tests must use the
+nearest supported umbrella, while implementation headers remain free to change.
+
+## Alternatives considered
+
+Making every leaf header self-contained would increase compile cost and freeze
+implementation structure without adding value for normal application code.
diff --git a/docs/adr/0002-async-executor-model.md b/docs/adr/0002-async-executor-model.md
new file mode 100644
index 0000000..2884e75
--- /dev/null
+++ b/docs/adr/0002-async-executor-model.md
@@ -0,0 +1,25 @@
+# ADR 0002: Shared and dedicated asynchronous executors
+
+- Status: Accepted
+- Date: 2026-09-14
+
+## Context
+
+Using one shared asynchronous executor is efficient for the default case, but
+a slow backend can delay unrelated backends. A separate worker per backend
+improves isolation at the cost of a thread and queue per opted-in backend.
+
+## Decision
+
+The shared `TaskExecutor` remains the default. Backends that support it expose
+`Config::use_dedicated_executor`; when enabled for an asynchronous backend, it
+owns a `SingleThreadExecutor` with its own queue, policy, wait, and shutdown
+lifecycle. Platform-specific backends that cannot provide this behavior keep
+their documented limitations.
+
+## Consequences
+
+Applications can choose throughput/resource sharing or isolation per backend.
+The feature is already tested for lifecycle, mixed shared/dedicated operation,
+and macro configuration. Changing the default is a separate compatibility and
+resource-usage decision.
diff --git a/docs/adr/0003-immutable-registry-snapshots.md b/docs/adr/0003-immutable-registry-snapshots.md
new file mode 100644
index 0000000..89d11a9
--- /dev/null
+++ b/docs/adr/0003-immutable-registry-snapshots.md
@@ -0,0 +1,25 @@
+# ADR 0003: Immutable logger registry snapshots
+
+- Status: Accepted
+- Date: 2026-09-14
+
+## Context
+
+Copying the mutable logger registry under a read lock on every log call made
+the normal dispatch path pay synchronization and allocation costs. Logger
+registry mutations and reconfiguration are infrequent compared with logging.
+
+## Decision
+
+Publish an immutable copy-on-write strategy-list snapshot on registry changes.
+The hot path reads it with an atomic `shared_ptr` load. Mutable `enabled` and
+`single_mode` state is atomic. The per-strategy execution mutex remains around
+formatter/backend invocation because custom formatters and backends do not
+have a universal concurrent-invocation contract.
+
+## Consequences
+
+Normal dispatch avoids the registry mutex and temporary vector copy while
+preserving safe lifetime and mutation semantics. A future lock-elision path
+requires an explicit capability contract and separate regression coverage; a
+benchmark sink alone is not evidence that arbitrary backends are thread-safe.
diff --git a/docs/adr/0004-timeshield-compatibility.md b/docs/adr/0004-timeshield-compatibility.md
new file mode 100644
index 0000000..776f277
--- /dev/null
+++ b/docs/adr/0004-timeshield-compatibility.md
@@ -0,0 +1,25 @@
+# ADR 0004: TimeShield compatibility and dependency reuse
+
+- Status: Accepted
+- Date: 2026-09-14
+
+## Context
+
+The current TimeShield release is `v2.0.0`, whose CMake package uses
+`SameMinorVersion` compatibility. Consumers may provide dependency targets
+from a parent project or sibling build instead of installing another package
+configuration.
+
+## Decision
+
+Pin the bundled submodule to the `v2.0.0` release and request
+`TimeShield 2.0.0`, which expresses the supported `2.0.x` line. Before package
+discovery, reuse an existing `time_shield::time_shield` target; installed
+dependency discovery follows the same target-reuse rule for optional packages.
+
+## Consequences
+
+Build-tree and installed-package consumers remain composable, while the
+minimum supported TimeShield API is explicit. Moving to another minor line or
+changing the compatibility policy requires a new compatibility review and
+release note.
diff --git a/docs/adr/0005-benchmark-methodology.md b/docs/adr/0005-benchmark-methodology.md
new file mode 100644
index 0000000..7205164
--- /dev/null
+++ b/docs/adr/0005-benchmark-methodology.md
@@ -0,0 +1,27 @@
+# ADR 0005: Benchmark evidence and comparison methodology
+
+- Status: Accepted
+- Date: 2026-09-14
+
+## Context
+
+Prepared records, the public macro path, formatted output, and external
+library adapters measure different work. Mixing them into one number produces
+misleading claims, and short CI runners are not a substitute for a fixed
+publication machine.
+
+## Decision
+
+Keep separate scenarios for prepared-record dispatch, the real public
+`LOGIT_INFO(...)` path, formatting, and external-library comparisons. The
+public macro smoke benchmark may use a passthrough formatter when it is
+explicitly documented as record-construction/dispatch coverage. Report
+absolute timings only with compiler, platform, commit, queue, producer, and
+flush settings; treat CI runs as regression smoke unless the environment is
+fixed.
+
+## Consequences
+
+Benchmark documentation remains comparable and honest across changes. New
+scenarios require their own workload contract and should not silently replace
+historical measurements.
diff --git a/docs/adr/README.md b/docs/adr/README.md
new file mode 100644
index 0000000..71cb586
--- /dev/null
+++ b/docs/adr/README.md
@@ -0,0 +1,30 @@
+# Architecture Decision Records
+
+Architecture Decision Records (ADRs) capture decisions that should remain
+discoverable after the implementation or the original discussion changes.
+They complement the roadmap: the roadmap says **what is next**, while an ADR
+explains **why the current boundary or trade-off exists**.
+
+## Statuses
+
+- **Proposed** — under discussion; do not treat it as a public contract.
+- **Accepted** — the current implementation and documentation should follow it.
+- **Superseded** — replaced by a newer ADR; retain it for historical context.
+- **Rejected** — considered and intentionally not adopted.
+
+## Creating an ADR
+
+1. Copy [`0000-template.md`](0000-template.md) to the next zero-padded number.
+2. Use a short, stable title and set the status/date.
+3. Describe the context, decision, consequences, and rejected alternatives.
+4. Update the implementation and user-facing documentation in the same change
+   when the decision changes a public contract.
+5. Link a superseding ADR instead of rewriting an accepted historical record.
+
+## Accepted decisions
+
+- [0001 — Aggregate-first public headers and API boundary](0001-aggregate-first-public-headers.md)
+- [0002 — Shared and dedicated asynchronous executors](0002-async-executor-model.md)
+- [0003 — Immutable logger registry snapshots](0003-immutable-registry-snapshots.md)
+- [0004 — TimeShield compatibility and dependency reuse](0004-timeshield-compatibility.md)
+- [0005 — Benchmark evidence and comparison methodology](0005-benchmark-methodology.md)
diff --git a/docs/future-plans.md b/docs/future-plans.md
new file mode 100644
index 0000000..b1b895e
--- /dev/null
+++ b/docs/future-plans.md
@@ -0,0 +1,81 @@
+# Roadmap and delivery checklist — LogIt++
+
+This file is the maintained roadmap, not a changelog. It deliberately separates
+delivered capabilities from planned work and from ideas that are intentionally
+delegated to user-defined `ILogger` implementations.
+
+Legend:
+
+- `[x]` delivered and covered by code/docs/tests;
+- `[~]` available, but opt-in or platform-limited;
+- `[ ]` planned work;
+- `[-]` intentionally deferred, with no current commitment to implement it in
+  the core library.
+
+## Delivered
+
+- [x] Public aggregate-first header policy: `<logit.hpp>` is the supported
+  application entry point; module umbrellas and leaf-header contracts are
+  documented.
+- [x] Public `logit::QueuePolicy` API and queue policy macros.
+- [x] Immutable copy-on-write logger registry snapshots on the production hot
+  path, with atomic logger state and the existing execution mutex retained for
+  formatter/backend safety.
+- [~] Dedicated per-backend worker threads via
+  `Config::use_dedicated_executor=true`. The default remains the shared
+  `TaskExecutor`; unsupported platform backends keep their documented
+  limitations.
+- [x] Queue capacity, overflow policies, drop counters, resize behavior, and
+  flush/shutdown coverage.
+- [x] MDC/NDC context, structured records, stored-log readers/subscribers, and
+  runtime log-level controls.
+- [x] TimeShield `2.0.x` compatibility (minimum `2.0.0`) and dependency-target
+  reuse for build-tree and installed-package consumers.
+- [x] Benchmark methodology and regression infrastructure: prepared-record hot
+  path A/B, public `LOGIT_INFO(...)` smoke benchmark, queue/CSV validation,
+  delayed-sink flush regression, and CI coverage.
+- [x] Doxygen generation, GitHub Pages publication, header/layout smoke tests,
+  and the comparison guide for spdlog, Quill, Boost.Log, glog, and IceCream-Cpp.
+
+## Next planned work
+
+- [ ] **Release 1.0.2** — prepare changelog/release notes, verify package
+  overlays, generate the non-`-dev` documentation, tag the release, and verify
+  the published Pages site.
+- [ ] **Binary logging research** — choose a versioned binary record format,
+  define compatibility/versioning rules, and prototype a reader before adding
+  a production backend.
+- [ ] **Transparent compressed-file reads** — make `read_log_file()` and
+  `read_log_files()` read `.gz`/`.zst` entries when the corresponding feature is
+  enabled, with platform-specific tests.
+- [ ] **Configuration loading** — design a versioned JSON/properties mapping to
+  the existing backend configuration. Treat file watching/hot reload as a
+  follow-up, not part of the first configuration API.
+- [ ] **Extended filtering** — evaluate source/file, message, tag/MDC, and
+  range filters; define their cost and ordering before adding public API.
+- [ ] **Benchmark follow-up** — add a genuinely formatted public-macro scenario,
+  a versioned fixture containing compiler/toolchain/commit/queue/flush metadata,
+  and a broader 1/4/16/32-producer matrix. Keep publication numbers tied to a
+  fixed machine and toolchain.
+- [ ] **Concurrency fast-path research** — only after documenting a formal
+  thread-safety capability for formatters/backends. Do not remove `exec_mx`
+  based on benchmark results alone.
+
+## Intentionally deferred
+
+- [-] **Database/SQLite appender** — currently better implemented as a custom
+  `ILogger`; revisit if a common portable schema and maintenance owner emerge.
+- [-] **Telnet/TCP appender** — application-specific transport and access
+  policy; implement as a custom backend unless a supported protocol is agreed.
+- [-] **Main-thread GUI appender** — event-loop integration is framework and
+  platform specific; revisit when LogIt++ ships a GUI viewer.
+
+## How to update this document
+
+Move an item to `[x]` only when the implementation, tests, and user-facing
+documentation agree. Use an ADR in [`docs/adr`](https://github.com/LimiNode/log-it-cpp/tree/main/docs/adr) for architectural
+decisions or compatibility changes; keep this file focused on status and next
+steps. Record measurements with their environment instead of turning one local
+run into a universal performance claim.
+
+Last reviewed: 2026-09-14.
diff --git a/docs/mainpage.dox b/docs/mainpage.dox
index 21366c5..cf6d5ad 100644
--- a/docs/mainpage.dox
+++ b/docs/mainpage.dox
@@ -31,6 +31,7 @@ int main() {
 - \ref benchmarks — benchmark methodology and historical snapshot.
 - \ref api_reference — detailed macro, formatting, configuration, and extension reference.
 - \ref comparison — comparison with other C++ logging libraries and trade-offs.
+- Architecture decisions — the source repository's [`docs/adr/`](https://github.com/LimiNode/log-it-cpp/tree/main/docs/adr) records the rationale for public boundaries, executors, snapshots, and compatibility.
 
 \section feature_summary Feature summary
 
diff --git a/docs/quickstart.md b/docs/quickstart.md
index d238f35..2075768 100644
--- a/docs/quickstart.md
+++ b/docs/quickstart.md
@@ -40,6 +40,11 @@ dependency requirements are described in the [`Backend matrix`](backends.html).
   the historical snapshot disclaimer.
 - **Choosing a logging library** — [`Comparison with other C++ logging libraries`](comparison.html),
   including trade-offs and the scope of the comparison.
+- **Architecture decisions** — [`Architecture Decision Records`](https://github.com/LimiNode/log-it-cpp/tree/main/docs/adr),
+  including public API boundaries, executor ownership, snapshots, and
+  dependency compatibility.
+- **Roadmap** — [`delivery checklist`](https://github.com/LimiNode/log-it-cpp/blob/main/docs/future-plans.md),
+  separating delivered, planned, and intentionally deferred work.
 - **Examples** — browse the `examples/` directory in the source repository;
   each optional example states the feature macro it requires.