Skip to content

WONT-DO: add ADR-vocabulary Status line per entry #39

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
AceHack merged 2 commits into from
Apr 21, 2026
Merged

WONT-DO: add ADR-vocabulary Status line per entry #39

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
980a2ef
WONT-DO: add explicit ADR-vocabulary Status line per entry
AceHack Apr 21, 2026
4b1aec0
fix(WONT-DO): MD001 heading increment — statuses-mean section H3 -> H2
AceHack Apr 21, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
74 changes: 73 additions & 1 deletion docs/WONT-DO.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -10,17 +10,42 @@ this file (to BACKLOG, ROADMAP, or delete entirely).

Entries are grouped by area. Each entry has:

- **Status** (one of `Rejected`, `Declined`, `Deprecated`,
`Superseded`) — matches ADR (Nygard) vocabulary.
- **Decision date** (for the record)
- **Proposal** (what was suggested)
- **Why not** (one or two sentences)
- **Revisit criteria** (what would flip the decision)
Comment on lines 15 to 18
Copy link

Copilot AI Apr 21, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

P2: The entry-field list says "Decision date" and "Revisit criteria", but the actual entry keys used throughout the document are "Decision:" and "Revisit when:". Please align the preamble wording with the actual field names (or rename the per-entry keys) so readers don’t have to guess which labels are canonical.

Suggested change
- **Decision date** (for the record)
- **Proposal** (what was suggested)
- **Why not** (one or two sentences)
- **Revisit criteria** (what would flip the decision)
- **Decision:** (for the record)
- **Proposal** (what was suggested)
- **Why not** (one or two sentences)
- **Revisit when:** (what would flip the decision)

Copilot uses AI. Check for mistakes.

## What the statuses mean

- **`Rejected`** — considered and turned down for a durable
architectural, values-level, correctness, or security
reason. Revisit-when is typically "never" or "only if the
Copy link

Copilot AI Apr 21, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

P2: In the status definitions, "Revisit-when" is hyphenated, but the per-entry field is consistently written as "Revisit when". For consistency (and easier searchability), standardize on one form (ideally the exact field label used in entries).

Suggested change
reason. Revisit-when is typically "never" or "only if the
reason. Revisit when is typically "never" or "only if the

Copilot uses AI. Check for mistakes.
underlying constraint moves". Most entries in this file.
- **`Declined`** — turn-down with a plausibly-reachable
revisit threshold — usually evidence, hardware /
ecosystem maturation, or an explicit scope extension. No
architectural mismatch; the answer is "not now", not
"not ever".
- **`Deprecated`** — was once accepted, now being phased
out. Pointer to the replacement should appear in the
entry.
- **`Superseded`** — replaced by a different decision.
Pointer to the superseding ADR or entry required.

The filename `WONT-DO.md` echoes the Jira / Bugzilla
`WONTFIX` / `Won't Do` lineage and is kept as the recognisable
slug; status verbs per entry match ADR convention for
precision about the reason shape.

---

## Algorithms / operators

### Cuckoo / Morton filter as a replacement for counting Bloom

- **Status:** Rejected
- **Decision:** 2026-04-17
- **Proposal:** Use cuckoo / Morton filters — smaller, faster,
support native deletes.
Expand All @@ -36,6 +61,7 @@ Entries are grouped by area. Each entry has:

### DRed (Delete and Re-derive) for retraction-safe recursion

- **Status:** Rejected
- **Decision:** 2026-04-17
- **Proposal:** Implement Gupta-Mumick-Subrahmanian DRed to make
`RecursiveSemiNaive` retraction-safe.
Expand All @@ -50,6 +76,7 @@ Entries are grouped by area. Each entry has:

### Plain / Sandwiched / Partitioned Learned Bloom Filters

- **Status:** Rejected
- **Decision:** 2026-04-17
- **Proposal:** Add learned-Bloom variants (Kraska 2018 / Mitzenmacher 2018 / Vaidya 2021).
- **Why not:** Classifier retraining per retraction batch breaks
Expand All @@ -62,6 +89,7 @@ Entries are grouped by area. Each entry has:

### Deletable Bloom (Rothenberg 2010)

- **Status:** Rejected
- **Decision:** 2026-04-17
- **Proposal:** Use deletable Bloom for a cheaper counting variant.
- **Why not:** Only deletes non-collided bits; produces silent
Expand All @@ -74,6 +102,7 @@ Entries are grouped by area. Each entry has:

### Lock-free `Circuit.Register` via CAS-on-record

- **Status:** Declined
- **Decision:** 2026-04-17
- **Proposal:** Replace the `lock registerLock` in `Circuit.Register`
with a CAS-on-record loop.
Expand All @@ -89,6 +118,7 @@ Entries are grouped by area. Each entry has:

### Centralised feature-flag server (LaunchDarkly / Unleash / GrowthBook)

- **Status:** Rejected
- **Decision:** 2026-04-17
- **Proposal:** Use a SaaS feature-flag service.
- **Why not:** Library consumers must be able to run offline.
Expand All @@ -100,6 +130,7 @@ Entries are grouped by area. Each entry has:

### CXL persistent memory integration

- **Status:** Declined
- **Decision:** 2026-04-17 (early decision)
- **Proposal:** Target CXL-backed persistent memory for the
storage tier.
Expand All @@ -111,6 +142,7 @@ Entries are grouped by area. Each entry has:

### io_uring as the I/O primitive

- **Status:** Declined
- **Decision:** early decision (undated)
- **Proposal:** Use `io_uring` for disk I/O in the Linux path.
- **Why not:** No first-class .NET support; P/Invoke only, which
Expand All @@ -121,6 +153,7 @@ Entries are grouped by area. Each entry has:

### Microsoft Threat Modeling Tool for the threat model

- **Status:** Declined
- **Decision:** early decision (undated)
- **Proposal:** Use Microsoft's GUI Threat Modeling Tool.
- **Why not:** Windows-only; Parallels-only workflow on macOS.
Expand All @@ -130,6 +163,7 @@ Entries are grouped by area. Each entry has:

### Sakana AI Scientist / Karpathy autoresearch for research-pipeline

- **Status:** Rejected
- **Decision:** early decision (undated)
- **Proposal:** Adopt one of these autoresearch frameworks.
- **Why not:** Sakana is RAIL-licensed + GPU-only; workshop-tier
Expand All @@ -141,6 +175,7 @@ Entries are grouped by area. Each entry has:

### git-crypt for secrets management

- **Status:** Rejected
- **Decision:** 2026-04-21
- **Proposal:** Adopt `git-crypt` (GPG / symmetric, AES-256-CTR
with SHA-1-HMAC-derived IV) for secrets-at-rest inside the
Expand Down Expand Up @@ -184,6 +219,7 @@ Entries are grouped by area. Each entry has:

### Archive completed changes into `openspec/changes/archive/`

- **Status:** Rejected
- **Decision:** 2026-04-17
- **Proposal:** Use the upstream OpenSpec change-history
archive flow.
Expand All @@ -197,6 +233,7 @@ Entries are grouped by area. Each entry has:

### BinaryFormatter / NetDataContractSerializer / SoapFormatter

- **Status:** Rejected
- **Decision:** 2026-04-17
- **Proposal:** Use one of these for `.NET`-native serialisation.
- **Why not:** Known-unsafe under untrusted input (arbitrary
Expand All @@ -209,6 +246,7 @@ Entries are grouped by area. Each entry has:

### Round-named test files (`Round17Tests.fs`)

- **Status:** Rejected
- **Decision:** 2026-04-17
- **Proposal:** Keep ad-hoc `RoundNTests.fs` files as a chronological
grab-bag.
Expand All @@ -222,6 +260,7 @@ Entries are grouped by area. Each entry has:

### Automatic skill self-modification without git visibility

- **Status:** Rejected
- **Decision:** 2026-04-17
- **Proposal:** Let skills rewrite their own `SKILL.md` freely.
- **Why not:** A skill that edits its own prompt is effectively
Expand All @@ -234,6 +273,7 @@ Entries are grouped by area. Each entry has:

### Fetching adversarial prompt-injection corpora for pen-testing

- **Status:** Declined
- **Decision:** 2026-04-17
- **Proposal:** Fetch the elder-plinius repos (`L1B3RT4S`,
`OBLITERATUS`, `G0DM0D3`, `ST3GG`) for in-repo pen-tests.
Expand All @@ -257,6 +297,7 @@ library in F#. Declining them here saves review cycles.

### SQL tokenizer / parser / binder / logical planner / optimizer

- **Status:** Rejected
- **Decision:** 2026-04-17
- **Proposal:** Build a full SQL front-end pipeline (parser → binder
→ logical plan → optimizer → physical plan → executor).
Expand All @@ -271,6 +312,7 @@ library in F#. Declining them here saves review cycles.

### SQL standards-conformance feature matrix

- **Status:** Rejected
- **Decision:** 2026-04-17
- **Proposal:** Track per-feature SQL:2023 conformance in a generated
matrix modelled on PostgreSQL's `sql_features` table.
Expand All @@ -281,6 +323,7 @@ library in F#. Declining them here saves review cycles.

### Vendor-dialect adapters (PostgreSQL / SQLite / MySQL / T-SQL)

- **Status:** Rejected
- **Decision:** 2026-04-17
- **Proposal:** Build interface-layer dialect adapters so one engine
serves many SQL surfaces.
Expand All @@ -290,6 +333,7 @@ library in F#. Declining them here saves review cycles.

### B-tree page parsing, varint decoding, overflow-page reconstruction

- **Status:** Rejected
- **Decision:** 2026-04-17
- **Proposal:** Build B-tree page parsers, varint encoders, overflow
pages — the SQLite-style storage-reader primitives.
Expand All @@ -302,6 +346,7 @@ library in F#. Declining them here saves review cycles.

### ACID transactions, MVCC snapshots, crash recovery as engine core

- **Status:** Rejected
- **Decision:** 2026-04-17
- **Proposal:** Build the standard single-node ACID engine (WAL,
MVCC, crash recovery, isolation-level semantics) inside
Expand All @@ -317,6 +362,7 @@ library in F#. Declining them here saves review cycles.

### Root-catalog discovery / catalog snapshots / table metadata parsing

- **Status:** Rejected
- **Decision:** 2026-04-17
- **Proposal:** Parse a root catalog, maintain immutable catalog
snapshots, resolve table/index metadata on open.
Expand All @@ -327,6 +373,7 @@ library in F#. Declining them here saves review cycles.

### Typed CLR materialization with constructor-binding

- **Status:** Rejected
- **Decision:** 2026-04-17
- **Proposal:** Ship a reflection-compiled materializer that maps
table rows to immutable CLR types with `required init` binding.
Expand All @@ -339,6 +386,7 @@ library in F#. Declining them here saves review cycles.

### JDBC-like driver / DB-API / `IQueryable` provider

- **Status:** Rejected
- **Decision:** 2026-04-17
- **Proposal:** Ship an `IQueryable<T>` + `IQueryProvider` public
surface so LINQ queries translate into our operators.
Expand All @@ -349,12 +397,25 @@ library in F#. Declining them here saves review cycles.
async step without contorting the algebra. `IQbservable`
(P2, Reaqtor / Bonsai slim-IR) is the direction for persistable
queries.
- **Scope clarification:** today the only Rx-family surface we ship
is `IObservable<'T>` via `RxAdapter.asObservable` in
`src/Core/Rx.fs` — push, no expression tree. We ship neither
`IQbservable<'T>` nor `IAsyncQueryable<'T>`. Both of those carry
`System.Linq.Expressions` machinery — the same substrate
`IQueryable` owns — so "expose async LINQ but not sync LINQ" is
not a clean split: picking up the expression-tree substrate
re-opens the `IQueryable` question even if we only wanted the
async variant. What's rejected here is the pull-LINQ contract,
not the expression-tree substrate.
Comment on lines +401 to +409
Copy link

Copilot AI Apr 21, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

P2: The scope clarification says the only Rx-family surface shipped is IObservable<'T> via RxAdapter.asObservable, but src/Core/Rx.fs also publicly exposes RxAdapter.asObservableForCount (also returning IObservable<'T>). Consider rephrasing to refer to the RxAdapter module’s IObservable surface (or list both functions) to avoid implying there is only one entry point.

Suggested change
is `IObservable<'T>` via `RxAdapter.asObservable` in
`src/Core/Rx.fs` — push, no expression tree. We ship neither
`IQbservable<'T>` nor `IAsyncQueryable<'T>`. Both of those carry
`System.Linq.Expressions` machinery — the same substrate
`IQueryable` owns — so "expose async LINQ but not sync LINQ" is
not a clean split: picking up the expression-tree substrate
re-opens the `IQueryable` question even if we only wanted the
async variant. What's rejected here is the pull-LINQ contract,
not the expression-tree substrate.
is the `RxAdapter` module's `IObservable<'T>` surface in
`src/Core/Rx.fs` (`RxAdapter.asObservable` and
`RxAdapter.asObservableForCount`) — push, no expression tree. We
ship neither `IQbservable<'T>` nor `IAsyncQueryable<'T>`. Both of
those carry `System.Linq.Expressions` machinery — the same
substrate `IQueryable` owns — so "expose async LINQ but not sync
LINQ" is not a clean split: picking up the expression-tree
substrate re-opens the `IQueryable` question even if we only
wanted the async variant. What's rejected here is the pull-LINQ
contract, not the expression-tree substrate.

Copilot uses AI. Check for mistakes.
- **Revisit when:** A user workload specifically needs `IQueryable`
compatibility and the cost of implementing it honestly doesn't
distort the async circuit step.
distort the async circuit step; or when the IQbservable /
IAsyncQueryable work begins and the expression-tree substrate
needs to be re-adjudicated.

### `WITHOUT ROWID` tables / alternative row-identity schemes

- **Status:** Rejected
- **Decision:** 2026-04-17
- **Proposal:** Support SQLite-style `WITHOUT ROWID` or other
row-identity modes.
Expand All @@ -363,6 +424,7 @@ library in F#. Declining them here saves review cycles.

### Networked single-node service shell

- **Status:** Rejected
- **Decision:** 2026-04-17
- **Proposal:** Wrap the embedded engine in a management + query
service with versioned wire contracts, gRPC/JSON/etc.
Expand All @@ -377,6 +439,7 @@ library in F#. Declining them here saves review cycles.

executor layering as a package

- **Status:** Rejected
- **Decision:** 2026-04-17
- **Proposal:** Adopt the mature analytical-DB layering.
- **Why not:** Same reason — no SQL compiler to layer. The DBSP
Expand All @@ -389,6 +452,7 @@ executor layering as a package

### EventStore-style server-side JavaScript projections

- **Status:** Rejected
- **Decision:** 2026-04-17
- **Proposal:** Adopt the EventStoreDB projection model where
projections are JS code stored on the server.
Expand All @@ -400,6 +464,7 @@ executor layering as a package

### PostgreSQL-style `sql_features` conformance table

- **Status:** Rejected
- **Decision:** 2026-04-17
- **Proposal:** Publish per-feature SQL standards conformance.
- **Why not:** Same reason as the matrix entry above. The
Expand All @@ -410,6 +475,7 @@ executor layering as a package

### MariaDB-style pluggable storage engines

- **Status:** Rejected
- **Decision:** 2026-04-17
- **Proposal:** Support multiple storage engines behind one SQL
surface.
Expand All @@ -422,6 +488,7 @@ executor layering as a package

ClickHouse / Iceberg / Delta Lake)

- **Status:** Rejected
- **Decision:** 2026-04-17
- **Proposal:** Build a columnar analytical projection engine
alongside the row/event engine for HTAP workloads.
Expand All @@ -437,6 +504,7 @@ ClickHouse / Iceberg / Delta Lake)

parser research

- **Status:** Rejected
- **Decision:** 2026-04-17
- **Proposal:** Adopt the layered "SQL front-ends normalise into the
same logical model" architecture.
Expand All @@ -449,6 +517,7 @@ parser research

module manifests

- **Status:** Declined
- **Decision:** 2026-04-17
- **Proposal:** Build a PostgreSQL-style extension lifecycle with
versioned module manifests, upgrade scripts, and DI composition.
Expand All @@ -462,6 +531,7 @@ module manifests

### "Log-only durability" as a repository-level posture

- **Status:** Rejected
- **Decision:** 2026-04-17
- **Proposal:** Ban any non-log durability mode at the repo level.
- **Why not:** Storage ownership belongs to the sink, not the
Expand All @@ -477,6 +547,7 @@ module manifests

### Emulating a deceased family member of a maintainer without the authorized surviving-consent-holders' agreement

- **Status:** Rejected
- **Decision:** 2026-04-19
- **Proposal:** Build an agent, skill, persona, research
artifact, training dataset, fictional backstory, or composite
Expand Down Expand Up @@ -519,6 +590,7 @@ module manifests

### Generalising a single-instance consent rule by analogy

- **Status:** Declined
- **Decision:** 2026-04-19
- **Proposal:** Extend the Elisabeth-specific consent rule by
analogy to other deceased individuals without explicit
Expand Down
Loading