congruency — tournament submission

Evolved from the 2006 congruency PHP content-management system (PHP - content management). This folder is a code-evolution tournament entry: a self-contained snapshot of branch order-logging @ commit 2a2e4f8c142e…, evolved into a minimal, reliable, exhaustively-tested tagging system.

See submission_00.txt for the entry's location relative to the main codebase and the full protocol details.

What this is

At the core is a custom tagging system — a parser that scans content for <<<Name(arg)(arg)>>> tags and renders it. The entry evolves that core, guided entirely by an oracle (test-first), against seven goals.

Run the oracle

cd tests/parser
../../../congruency-harness/php/php run.php    # exit 0 iff all assertions pass

One command. No database, no web server, zero JavaScript. Prints parser corpus: N/N assertions passed.

Run the live CMS

The 2006 CMS itself is resurrected and served (zero-JavaScript, server-rendered through the <<<tag>>> engine):

python3 tools/serve.py            # http://127.0.0.1:8899/?page=catalog

- Config is data — boot/constants.default.json (defaults) merged with install.json (overrides, e.g. CONGRUENCY_SQLITE). install.json may live outside the install folder via the $CONGRUENCY_CONFIG env var. The unified DB is ~/.jazz/congruency.sqlite. - Database — native PDO — every DAO reaches the database through DataConnection, a thin native-PDO layer (new PDO('sqlite:' . CONGRUENCY_SQLITE)); its query() returns a small result object whose ->rows is an array of associative rows that the DAOs iterate directly. The 2006 code was written against ext/mysql (mysql_*), removed from PHP 7+; that entire API was migrated to PDO — there are now zero mysql_* calls or shims in the tree. boot/shim.php survives only as a one-line get_magic_quotes_gpc() polyfill (that function was removed in PHP 8.0 and the DAO quote() guards still reference it). The layer is driver-agnostic: hand DataConnection a mysql: DSN instead of sqlite: and the same DAOs run against MySQL — SQLite is just this deployment's target. - Pages — ?page= catalog · about · bugs · tickets · memories · pages · tags. Navigation is generated from the Documents table by <<<SiteMap>>>. - Tickets & tags — a full ticket tracker (<<<TicketList>>> + a native-forms submission flow), a memory log (<<<MemoryList>>>, the gate/MCP tool-use log), a page index (<<<PageList>>>), a tag gallery (<<<TagList>>>), and category tagging (Page_Categories + <<<CategoryPages>>>). - REST — generic CRUD over every table: GET/POST/PUT/DELETE ?api=<table> (paginated, allowlisted, column-validated). - Tooling — tools/crawl.py (broken-link spider), tools/tagcheck.py (tag-render harness), tools/gpl_stamp.py (GPL-header check/stamp), tools/deploy_lifecycle_check.py (production-deploy on/off/redeploy gate — drives deploy.py in an isolated git worktree, leaving your branch untouched).

Full write-up: doc/CMS_ADDITIONS.md.

Layout

2a2e4f8c142e…/            this submission (snapshot of the branch @ entry)
├── lib/ invocators/ …    the original CMS under test (unmodified)
├── submission_00.txt     where this entry sits vs. the main codebase
└── tests/parser/         ALL evolved code + tested data
    ├── run.php           the single oracle (4481 green assertions)
    ├── manifest.json     self-describing dataset graph (38 datasets)
    ├── *.php             render / eval / scan / compute / expand / store / flow / …
    ├── generate_*.php    ground-truth generators (re-run + reproduced by the oracle)
    ├── *-fixtures.json   captured/authored test data
    └── README.md         architecture + per-goal coverage map (start here for detail)

The seven mission goals (all implemented, test-first)

1. One state → four formats at the edge — XML/HTML/JSON/YAML, single + collection, bidirectional, plus a 4×4 ingest-any/emit-any interoperability matrix. Format is a registry key, never baked into logic. 2. No magic strings — every key resolves through maps (tag registry, op-table, handler registry); dispatch precedence is op > handler > template. 3. Self-configuring + self-healing — reads its own manifest.json, reconciles the description against reality, and regenerates corrupt datasets (Corpus::heal); a reproducibility guard re-runs every generator. 4. Turing-complete evaluation core — 100+ recursive programs (factorial, Fibonacci, Ackermann, gcd, popcount, isqrt, Collatz, mutual/accumulator recursion) + 100+ AST expressions. <<<add(2)(3)>>> and <<<if(c)(t)(e)>>> drive computation through the real parser. 5. Zero JavaScript — server-side form → server → next-form state machine (reachability-verified); a real tag asserted to emit no <script>. 6. Git as the primary store — file store, git-versioned store (history is the database), and a content-addressed blob store (sha1 + dedup). 7. Modernized dated conventions, guided by tests — killed the Tag_Parser dynamic-property deprecation; scan.php validated against the real Tag_Wrapper::identify_tag via reflection.

Final state

38 datasets · 4481 green oracle assertions · 104 commits since entry. Oracle green every round — zero missed commits, zero oracle failures. Ground truth is captured from the live code; authored expectations are locked as regression baselines.

Detailed architecture and the suite-to-goal map: tests/parser/README.md.