docs: apply row-contract README layout

README.md

@@ -16,27 +16,38 @@ pytest -q
 
 <table width="100%">
 <tr>
-<td colspan="12" width="100.0%" valign="top">
+<td width="100%" valign="top">
 <div><span><b>00 · GNOSIS-MORPH-BENCH</b> · ANCIENT SCRIPT RESEARCH UTILITY</span> <span>RESEARCH-READY · LIVE INDUS BLOCKED</span></div>
       <h1>Toolkitting for ancient script <span>research.</span></h1>
       <p>A shared scoring rig for ancient-script methods &middot; PyPI <em>gnosis-morph-bench</em> v0.1.0 &middot; github.com/Zer0pa/Morph-Bench</p>
       <p>Every researcher studying ancient-script morphology rebuilds the same machinery from scratch: scoring routes, permutation nulls, stability sweeps. The result lives in one paper, then the next lab starts again from nothing. <em>Morph-Bench</em> is the shared rig that ends that cycle — it scores routes, preserves null results, runs stability checks, and records replay so the next study begins from a measured floor. Today <strong>37 tests pass</strong> and runs are byte-identical across macOS and Linux. Live Indus Phase 4 rerun is <strong>blocked</strong> on Phase 3c manifest access.</p>
 </td>
 </tr>
+</table>
+
+<table width="100%">
 <tr>
-<td colspan="12" width="100.0%" align="center" valign="top">
+<td width="100%" valign="top">
 <figure>
         <div><img src="docs/assets/product-page-mechanics.gif" alt="Gnosis-Morph-Bench approved scientific square mechanics diagram showing morphology benchmark-grid mechanics."></div>
         <figcaption><b>Scope:</b> morphology benchmark rig. Scoring, permutation nulls, stability sweeps, and replay pass; live Indus Phase 4 is blocked.</figcaption>
       </figure>
 </td>
 </tr>
+</table>
+
+<table width="100%">
 <tr>
-<td colspan="4" width="33.33%" valign="top">
+<td width="100%" valign="top">
 <div><b>01 · THE GAP</b> <span>NO MEASURE, NO BASELINE</span></div>
       <h2>Every ancient-script researcher rebuilds the same baseline from scratch. The measure is never kept.</h2>
 </td>
-<td colspan="8" width="66.67%" valign="top">
+</tr>
+</table>
+
+<table width="100%">
+<tr>
+<td width="100%" valign="top">
 <div><b>02 · MARKETS</b> <span>ADJACENT FORECASTS</span></div>
       <div>
         <div>
@@ -50,29 +61,38 @@ pytest -q
       <div>Every field on this list has ancient-script work underway. None of it starts from a measured baseline.</div>
 </td>
 </tr>
+</table>
+
+<table width="100%">
 <tr>
-<td colspan="6" width="50.0%" valign="top">
+<td width="50%" valign="top">
 <div><b>03 · VALUE</b></div>
       <div>$8.1<span>B</span></div>
       <div>Cultural-heritage digitization in 2030; research infrastructure for review and reuse, not commercial sale.</div>
 </td>
-<td colspan="6" width="50.0%" valign="top">
+<td width="50%" valign="top">
 <div><b>04 · INSIGHT</b></div>
       <h2>Measure the method once. <span>The next researcher starts from here.</span></h2>
 </td>
 </tr>
+</table>
+
+<table width="100%">
 <tr>
-<td colspan="6" width="50.0%" valign="top">
+<td width="50%" valign="top">
 <div><b>05.1 · CURRENT TECH</b> <span>REBUILT EVERY TIME</span></div>
         <p>Ancient-script research has no standard scoring rig. Each lab writes its own metric, picks its own null, records its own result — then the work moves on. The next researcher starts again from zero.</p>
 </td>
-<td colspan="6" width="50.0%" valign="top">
+<td width="50%" valign="top">
 <div><b>05.2 · OUR TECH</b> <span>THE BASELINE, KEPT</span></div>
         <p><em>Morph-Bench</em> installs from PyPI and gives a researcher a working starting point: a route scorer, a permutation null, a five-mode stability battery, and replayable records that are byte-identical across macOS and Linux Python 3.11. <strong>37 tests</strong> pass on a fresh clone. <strong>9 of 9</strong> adapter clauses hold.</p>
 </td>
 </tr>
+</table>
+
+<table width="100%">
 <tr>
-<td colspan="12" width="100.0%" valign="top">
+<td width="100%" valign="top">
 <div><b>05.3 · BENCHMARKS</b> <span>PUBLIC RECEIPTS</span></div>
       <div>
         <div>
@@ -90,14 +110,15 @@ pytest -q
       <div><b>Status:</b> working today. Live Indus Phase 4 rerun blocked on Phase 3c manifest.</div>
 </td>
 </tr>
+</table>
+
+<table width="100%">
 <tr>
-<td colspan="12" width="100.0%" valign="top">
+<td width="34%" valign="top">
 <div><b>06 · MEASUREMENT</b> <span>METHOD SCORING + REPLAY</span></div>
       <h2>Every method scored. <span>Every null kept. Every blocker named.</span></h2>
 </td>
-</tr>
-<tr>
-<td colspan="12" width="100.0%" valign="top">
+<td width="66%" valign="top">
 <div><b>06.1 · COMPARATIVE PERFORMANCE · BENCHMARK SURFACE</b></div>
       <div>
         <div>
@@ -110,92 +131,155 @@ pytest -q
       <div>Today's measurement is the rig itself: pytest on a fresh Python 3.11 clone, adapter contract coverage, forbidden-pattern lint, and byte-equal replay across macOS and Linux. Live Indus Phase 4 rerun and Cuneiform adapter are <strong>not</strong> in scope yet.</div>
 </td>
 </tr>
+</table>
+
+<table width="100%">
 <tr>
-<td colspan="12" width="100.0%" valign="top">
+<td width="100%" valign="top">
 <div><b>07 · KEY METRICS</b> <span>PUBLIC RECEIPTS</span></div>
 </td>
 </tr>
+</table>
+
+<table width="100%">
 <tr>
-<td width="20%" valign="top">
+<td width="100%" valign="top">
 <div><b>07.1 · PYTEST SURFACE</b></div>
       <div>37<span>/37</span></div>
       <div>Fresh clone &middot; <b>pytest passes end to end</b></div>
 </td>
-<td width="20%" valign="top">
+</tr>
+</table>
+
+<table width="100%">
+<tr>
+<td width="100%" valign="top">
 <div><b>07.2 · ADAPTER INTERFACE</b></div>
       <div>9<span>/9</span></div>
       <div>Adapter v1 &middot; <b>required clauses covered</b></div>
 </td>
-<td width="20%" valign="top">
+</tr>
+</table>
+
+<table width="100%">
+<tr>
+<td width="100%" valign="top">
 <div><b>07.3 · FORBIDDEN-PATTERN LINT</b></div>
       <div>0<span>/6</span></div>
       <div>Repo lint &middot; <b>no forbidden patterns</b></div>
 </td>
-<td width="20%" valign="top">
+</tr>
+</table>
+
+<table width="100%">
+<tr>
+<td width="100%" valign="top">
 <div><b>07.4 · PYPI RELEASE</b></div>
       <div>0.1.0</div>
       <div>Public PyPI &middot; <b>wheel and source ready</b></div>
 </td>
-<td width="20%" valign="top">
+</tr>
+</table>
+
+<table width="100%">
+<tr>
+<td width="100%" valign="top">
 <div><b>07.5 · LIVE INDUS RERUN</b></div>
       <div>null</div>
       <div>Blocked &middot; <b>Phase 3c manifest unblocks it</b></div>
 </td>
 </tr>
+</table>
+
+<table width="100%">
 <tr>
-<td colspan="4" width="33.33%" valign="top">
+<td width="100%" valign="top">
 <div><b>08 · FIXTURE REPLAY</b> <span>BYTE-STABLE SCOPE</span></div>
       <h2>Replay is byte-stable for fixture records, <span>not rights-limited heavy data.</span></h2>
 </td>
-<td colspan="5" width="41.67%" valign="top">
+</tr>
+</table>
+
+<table width="100%">
+<tr>
+<td width="66%" valign="top">
 <div><b>08.1 · WHAT REPLAYS EXACTLY</b> <span>FIXTURE RECORDS ONLY</span></div>
       <p>Smoke and replay outputs are <strong>byte-identical across documented macOS and Linux Python 3.11 environments</strong>, verified end-to-end through fresh-clone install. Replay records and SHA-256 reference-freeze helpers ship as part of the public package.</p>
       <p>That does not prove deterministic live Indus replay from repo custody, undisclosed heavy data, or cultural-heritage imagery — and it does not prove any text recovery. The unit of byte-exactness is <em>benchmark-fixture replay</em>, not domain-result reproduction.</p>
 </td>
-<td colspan="3" width="25.0%" valign="top">
+<td width="34%" valign="top">
 <div><b>08.2 · HONEST BLOCKER</b></div>
       <span>Honest Blocker &middot;</span>
       <p>Live Indus Phase 4 rerun is blocked on Phase 3c manifest access. Heavy-data and image-bearing release policy is open. Cuneiform adapter is deferred to a separate contract once the first live Indus replay lands. PyPI: <code>gnosis-morph-bench==0.1.0</code>; no <code>0.1.1</code> until a receipt. Non-claims: <em>no text recovery, no source identification, no data permission, no hosted service</em>.</p>
 </td>
 </tr>
+</table>
+
+<table width="100%">
 <tr>
-<td colspan="6" width="50.0%" valign="top">
+<td width="33%" valign="top">
 <div><b>09</b> </div>
       <h2>ONE BASELINE THE FIELD <span>CAN INHERIT.</span></h2>
 </td>
-<td colspan="6" width="50.0%" valign="top">
+<td width="67%" valign="top">
 <div><b>09.1 · THE AMBITION</b></div>
       <p>An ancient-script field that loses its scoring rig with every postdoc loses a generation of measurement. <em>Morph-Bench</em> is the rig the next lab installs instead of writing — a public floor for method comparison that outlives the researcher who first ran it.</p>
 </td>
 </tr>
+</table>
+
+<table width="100%">
 <tr>
-<td colspan="6" width="50.0%" valign="top">
+<td width="33%" valign="top">
 <div><b>09.2 · WHAT WORKS NOW</b></div>
         <h2>Working today: 37 tests pass, adapter contract holds, fixture replay is byte-stable, public PyPI release available.</h2>
 </td>
-<td colspan="6" width="50.0%" valign="top">
+<td width="67%" valign="top">
 <div><b>09.3 · WHAT'S STILL OPEN</b></div>
         <h2>Still blocked: live Indus Phase 4 rerun, heavy-data release policy, and a future Cuneiform adapter.</h2>
 </td>
 </tr>
+</table>
+
+<table width="100%">
 <tr>
-<td width="20%" valign="top">
+<td width="100%" valign="top">
 <div><b>09.4</b> &middot; LABS · NEAR-TERM (12–24 MO)</div>
       <div>One installable baseline, not three rebuilds</div><div>A second lab studying Indus morphology can install the same scoring rig their predecessor used, run it on their candidate method, and compare results that have a common floor. The conversation moves from "whose code" to "whose finding".</div>
 </td>
-<td width="20%" valign="top">
+</tr>
+</table>
+
+<table width="100%">
+<tr>
+<td width="100%" valign="top">
 <div><b>09.5</b> &middot; REVIEW · NEAR-TERM (12–24 MO)</div>
       <div>Reviewers see the null beside the result</div><div>A journal reviewer reading a morphology paper can ask for the permutation null and the stability sweep alongside the headline number, knowing both were produced by a public harness, not the author's private script. Skeptical review gets cheaper.</div>
 </td>
-<td width="20%" valign="top">
+</tr>
+</table>
+
+<table width="100%">
+<tr>
+<td width="100%" valign="top">
 <div><b>09.6</b> &middot; MUSEUMS · MID-TERM (24–48 MO)</div>
       <div>Heritage archives keep the method, not just the picture</div><div>When a museum funds a script-morphology study, the resulting paper can ship with a replayable run — same scoring rig, same null check, same stability battery — so the archive's investment outlives the postdoc who did the work. The method becomes part of the collection.</div>
 </td>
-<td width="20%" valign="top">
+</tr>
+</table>
+
+<table width="100%">
+<tr>
+<td width="100%" valign="top">
 <div><b>09.7</b> &middot; CROSS-SCRIPT · MID-TERM (24–48 MO)</div>
       <div>Indus and Cuneiform share a floor</div><div>Once a Cuneiform scorer sits beside the Indus one, two distant script communities can argue about results on the same scoring floor. Disagreement narrows to the data and the model — the scoring rig stays the same, so the argument finally becomes worth having.</div>
 </td>
-<td width="20%" valign="top">
+</tr>
+</table>
+
+<table width="100%">
+<tr>
+<td width="100%" valign="top">
 <div><b>09.8</b> &middot; CONVENTION · PARADIGM (48 MO+)</div>
       <div>A standard the next generation inherits</div><div>A graduate student starting in ancient-script morphology in 2030 finds the same baseline their advisor used in 2026 — installable, documented, with null results preserved. The field stops losing a generation of measurement every time the tools rot. One kept harness, many researchers.</div>
 </td>