…lete (#548)

* docs: rewrite the README, and gate the drift that produced it

777 lines down to 180. The old one buried what a reader decides on
under a release archive that stopped at v4.13.0, twelve releases
behind, and a badge whose alt text said 4.17.1 while the badge itself
said 4.20.0.

The rewrite keeps what a reader needs to decide and to install: the
problem, the install, what is automatic versus what you call, one
benchmark table with its provenance, the storage trade-off as a table
rather than three paragraphs, and the limits stated before the reader
hits them. The cortex-viz identity migration is preserved in full,
folded into a details block, because two contract tests pin those exact
commands and they are right to.

Every link in the file was resolved against the tree; two I first wrote
did not exist and were replaced with real targets.

The systemic half: the badge SVG is generated and gated but its README
alt text is hand-written prose that nothing read, which is how it
drifted three minor versions. It is now the 17th version surface, and a
test drives the desync rather than asserting the count alone.

Co-Authored-By: Claude Opus 5 (1M context) <[email protected]>

* docs: write the README around what makes Cortex different

The first cut compressed the old sections instead of deciding what the
file should say. It kept the generic claim (stores what happened,
brings it back, which is what every memory tool claims) and deleted the
two parts that made the case: the concrete Monday/Thursday scenario and
the real tool output.

What differentiates Cortex is quality control on memory, and none of it
was above the fold. The predictive-coding gate that merges a
near-duplicate instead of filing it beside the original. Provenance
grading at write time, locally: a memory whose file path no longer
resolves is stored as unverifiable rather than returned later as fact.
Supersession, so a correction demotes what it replaces and the chain
stays readable. Thermodynamic decay, so the store curates itself.

The file now opens on the problem those solve (a store that accumulates
gets worse as it grows and you cannot tell what to trust), shows each
mechanism with output taken from a real run, and only then explains
what it feels like to use. The scenario is restored, tightened.

229 lines, from 777.

Co-Authored-By: Claude Opus 5 (1M context) <[email protected]>

* docs: make README welcoming while preserving evidence and limits

* docs: layer the three positions and marketplace sections onto the welcoming pass

Base is 0571343, the owner's tone pass, kept verbatim: install first,
the hedged wording on consolidation and grading, the illustrative
example in place of internal ids, and the benchmark table labelled as a
historical v4.14.1 run. Nothing from that pass is rewritten; the merge
adds to it.

Added, because the marketplace and the MCPB directory require them and
both earlier cuts in this PR had dropped them: the not-an-Anthropic-
product affiliation disclaimer under the banner, Privacy Policy and
Support sections matching the privacy_policies and support fields
manifest.json already declares, the License section with its
independent-work statement and algorithm attribution, and the Citation
block. All restored from the pre-rewrite README, word for word.

Added, for the three positions the project holds: the Sovereign /
Cross-platform / Eco-responsible block after the header, the full
what-works-where host matrix with the uvx, Gemini extension and Codex
launch commands, and the Green software engineering section with every
hedge intact (no CO2 or energy figure is published; the SCI tooling's
own README says what it does not establish; the shipped efficiency PRs
are listed; the hook-boot constant is cited to auto_recall.py).

The one paragraph that moved rather than stayed put is the
never-downgraded PostgreSQL note, from Storage into Install, where the
behaviour it describes happens. The Project link row is absorbed into
Development. 464 lines.

Every relative link resolves against the tree. The codex contract
tests and the version-surface tests pass unchanged.

Co-Authored-By: Claude Fable 5.1 <[email protected]>

---------

Co-authored-by: Claude Opus 5 (1M context) <[email protected]>

Removed the disclaimer about the project's independence from Anthropic.

….0 (#549)

Follow-up to #548, cut from main at 5e94667. README.md only.

A fresh-context review of the merged head found that the blocks the
PR had described as restored from v4.20.0 were not. The Green software
engineering section had lost 23 lines, including two hedges that are
the point of the section: that the operator supplies the region,
observation period, lifecycle assessment and reservation assumptions,
and the closing paragraph stating that no token-savings or CO2 figure
is published for end-to-end sessions because none has been measured.
Security had been folded into Privacy Policy and both paraphrased;
Support and the License opener carried small wording changes.

Every one of those blocks is now the v4.20.0 text, plus the
Verification section that follows the green section there (the
per-mechanism ablation evidence and both paper PDFs), which had been
dropped in an earlier cut. Each block was compared to
`git show v4.20.0:README.md` at word level after whitespace
normalisation inside the splicing script; all comparisons print True.

The cross-platform positioning paragraph no longer lists hosts on its
own, since that list did not match the table it pointed to; it names
Claude Code, the Claude Desktop bundle and Claude Cowork and defers
every other host to the table.

Deliberately not reintroduced: the independent-project disclaimer the
owner removed on main in 5e94667.

Gates on this head: link and cited-path resolution, no duplicated
heading, check_doc_claims.py, check_version_surfaces.py (17 sites),
and the codex-contract and version-surface tests (33 pass).

Co-authored-by: Claude Fable 5.1 <[email protected]>

…runs nothing (#550)

--only was compared token for token against longmemeval, locomo, beam
and decision-ids, while the script itself prints the artifact names
longmemeval-s and beam-100K. Passing one of those, or any typo, matched
no benchmark: the script built the package, started and stopped an
ephemeral container, wrote MANIFEST.json and START_SNAPSHOT.json,
measured nothing and exited 0. Reproduced on main before the change.

A sourced library, benchmarks/lib/bench_only.sh, maps the two aliases
to the tokens want_bench matches and fails closed on an unknown or
empty token with exit 2 and the accepted list, at the end of
parse_args, before any uv run or container start. Empty tokens are
detected on the whole string first, because read -a drops a trailing
empty field.

Eleven tests pin the alias table, the rejection cases and the script's
end-to-end behaviour, which now needs neither Docker nor a synced
environment. README row and CHANGELOG entry added.

Co-authored-by: Claude Fable 5.1 <[email protected]>

…th its artifacts (#551)

* docs: reflect the v4.20.0 LongMemEval-S measurement in the README

The retrieval table cited only the v4.14.1 pre-tag run. It now shows
v4.20.0 beside it: Recall@10 97.8%, MRR 0.9047, n=500, measured on the
exact tagged tree (86251ab) with reproduce.sh --only longmemeval
--no-ablation in an isolated ephemeral pgvector pg16 container,
reranker loaded, consolidation disabled. The artifacts are committed
under benchmarks/results/repro/20260909-v4.20.0-longmemeval-s/:
longmemeval-s.json, MANIFEST.json and START_SNAPSHOT.json.

The provenance paragraph states the delta against v4.14.1 (0.4 points
of Recall@10, 0.012 of MRR) and the run's own floor check as the script
reports it: Recall@10 within the 0.005 tolerance of the July floor,
MRR 0.0093 below its floor, non-blocking by design, with the pointer to
docs/agent-guidance.md on why main no longer clears those floors. The
Verification lead-in is scoped to the v4.14.1 figures, since the
v4.20.0 run has no ablation campaign behind it.

The earlier full run on the same tree produced the same numbers before
the host killed it and its JSON was lost; its log cannot be committed
(.gitignore excludes benchmarks/results/**/*.log) and an absent
artifact cannot be cited, so the leg was re-measured.

Co-Authored-By: Claude Fable 5.1 <[email protected]>

* docs: state the run's dirty flag from the artifact's embedded manifest

The JSON's manifest.repro.git_dirty is false; the standalone MANIFEST.json has no such field, which is why the first cut said none existed. Found by the fresh-context review.

Co-Authored-By: Claude Fable 5.1 <[email protected]>

---------

Co-authored-by: Claude Fable 5.1 <[email protected]>

…nesia-mcp-viz v3.1.2 (#552)

Moves the two GitHub-sourced pins to the releases cut today:
zetetic-team-subagents v2.40.0 (657795b87d884f465d218492cc368ba25525307d,
release run 34403584553 green, assets attested) and hypermnesia-mcp-viz
v3.1.2 (cd471aa9750c356ced8048394c9ac28fff0da605, the squash commit of
cdeust/cortex-viz#155). Until this lands, both releases reach zero
installs (docs/agent-guidance.md, Releasing; Cortex #179).

Co-authored-by: Claude Fable 5.1 <[email protected]>

Moves the zetetic-team-subagents pin to v2.40.1
(586e439e575e9759c3224b840768d15c3604ac2f): the worktree sweep now
recognises squash merges and <repo>/.claude/worktrees, the Linux age probe
is fixed, and `worktree-manager.sh inventory` reports every worktree with
the verdict sweep will apply (zetetic-team-subagents #135). Until this pin
moves, installs keep the sweep that removed nothing for two months.

Co-authored-by: Claude Fable 5.1 <[email protected]>

Moves the zetetic-team-subagents pin to v2.41.0
(77b4f78503e6842928039b75440e20ea670584c8): every state file the plugin
owns now lives under ~/.claude/zetetic/, an existing flat install is
migrated once without overwriting anything, and setup.sh ends with a
layout report (zetetic-team-subagents #137, owner issue #136).

Co-authored-by: Claude Fable 5.1 <[email protected]>

…ader (ADR-1060 revision, no exception) (#555)

* fix(hooks): the decision gate judges every language, test file and header

Owner ruling of 2026-09-10 on ADR-1060: no exception. Six lines of ///
rationale in a tests/*.rs file passed the gate because the language was
deferred "until measured", the path was a test root and the run sat before
anything executable. Each exemption sufficed alone.

- _decision_gate_lex.py: one character-level scanner for the //, /* */, --
  and ; families (string literals, Rust raw strings, backtick strings and
  block comments tracked), covering .rs .js .jsx .ts .tsx .go .java .c .h
  .cpp .hpp .cc .swift .kt .kts .cs .scala .m .mm .dart .php .sql .lua .hs
  .el .clj .lisp, plus the full # family (.py via tokenize, the shell
  family via the heredoc skip, .pl .r .toml .yaml .yml .cfg .ini). The
  suffix table and the scanner table are one table (LANGUAGES).
- decision_gate.py: is_test_path, _TEST_DIRS, _TEST_FILE_RE, _is_header_run
  and the header boundary are deleted; the refusal states that no header,
  test file or language is exempt. PROSE_RUN_LIMIT, grandfathering, the
  source: pointer, fail-open and CORTEX_DECISION_GATE=off stay.
- Python: a comment token after code on the same line no longer counts as
  a comment line, as the scanner's docstring already claimed.
- ADR-1060 wiki page: revision section with the leak, what is removed,
  what stays and the tracked-tree measurement at 8d71210 (1479 files in
  judged suffixes, 10 with a header run of 8+, all grandfathered); mirror
  regenerated with scripts/check_project_wiki.py --write.

Co-Authored-By: Claude Fable 5.1 <[email protected]>
Claude-Session: https://claude.ai/code/session_01LSTGMnuSuUXWs7ZBofW1AW

* docs(changelog): record the ADR-1060 no-exception revision of the decision gate (#555)

Co-Authored-By: Claude Fable 5.1 <[email protected]>
Claude-Session: https://claude.ai/code/session_01LSTGMnuSuUXWs7ZBofW1AW

---------

Co-authored-by: Claude Fable 5.1 <[email protected]>

The decision gate judges every language, test file and header (ADR-1060
revision 2026-09-10, no exception, #555): // and /* */, --, ; and # comment
families, no test-path or header exemption, PROSE_RUN_LIMIT unchanged at 8,
grandfathering and the explicit override kept. Every version surface moves
to 4.21.0 (pyproject, plugin and marketplace manifests, server.json,
manifest.json, package.json and lock, gemini and codex manifests, badge,
uv.lock, README alt text) and the hypermnesia-mcp-viz pin moves to v3.2.0
(2482563172c0f02a1c022b82cb4e38299b7748b4, cortex-viz #156: the Wiki view
lists project wikis, where project_root ADRs actually live).


Claude-Session: https://claude.ai/code/session_01LSTGMnuSuUXWs7ZBofW1AW

Co-authored-by: Claude Fable 5.1 <[email protected]>