betterwithage commited on
Commit
714d76d
·
verified ·
1 Parent(s): c018836

feat(restraint): a11oy Restraint — governed + measured frugality gate

Browse files

Adopts the 6-rung ladder + lite/full/ultra intensity from the open-source Ponytail coding-agent skill (github.com/DietrichGebert/ponytail, MIT, (c) 2026 DietrichGebert) — adopted + governed, NOT invented here. Wires a pre-write reasoning gate into the a11oy Code path emitting restraint: ceiling comments. GOVERNED: signed DSSE receipt + advisory Lambda (Conjecture 1 OPEN, floor < 1.0) per decision. MEASURED: two-arm benchmark ported from Ponytail's promptfoo methodology, labelled MEASURED only when run else SAMPLE/ROADMAP; Ponytail's numbers stay CITED as theirs. ENERGY: J/token tie-in. Endpoints /api/a11oy/v1/restraint/{evaluate,bench,info}; UI at /restraint. Additive; 0 runtime CDN; 0 visible codenames.

Mirror of szl-holdings/a11oy main commit 57c7029c53adce0ffe00f482b2a70f656a3d7970.
Signed-off-by: Perplexity Computer Agent <agent@perplexity.ai>

.github/copy-sync-lockstep.json CHANGED
@@ -45,6 +45,7 @@
45
  "web/quant.html",
46
  "web/living-anatomy.html",
47
  "web/nemo.html",
 
48
  "web/operator.html",
49
  "web/v4_fleet_panel.html"
50
  ]
 
45
  "web/quant.html",
46
  "web/living-anatomy.html",
47
  "web/nemo.html",
48
+ "web/restraint.html",
49
  "web/operator.html",
50
  "web/v4_fleet_panel.html"
51
  ]
benchmarks/restraint/README.md ADDED
@@ -0,0 +1,38 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # a11oy Restraint — benchmark (Ponytail methodology, our measurements)
2
+
3
+ Two arms — **no-skill baseline** vs **a11oy-restraint** — over the same five
4
+ everyday tasks Ponytail uses (email validator, JS debounce, CSV sum, React
5
+ countdown, FastAPI rate-limit). Code LOC is counted from fenced code blocks;
6
+ tokens and latency come from the API. Median reported.
7
+
8
+ ## Honesty labels (Doctrine v11)
9
+
10
+ - **MEASURED** — printed ONLY for a run we actually executed on our stack.
11
+ - **SAMPLE** — an illustrative fixture derived from our ladder model (the
12
+ `/api/a11oy/v1/restraint/bench` endpoint returns this when no model run is
13
+ wired on the Space).
14
+ - **ROADMAP** — the overall bench label until a real run is wired.
15
+
16
+ We **never** reprint Ponytail's published numbers as ours.
17
+
18
+ ## Reproduce
19
+
20
+ ```bash
21
+ cp ../../.env.example ../../.env # add your model key
22
+ npx promptfoo@latest eval -c benchmarks/restraint/promptfooconfig.yaml --repeat 10
23
+ npx promptfoo@latest view
24
+ ```
25
+
26
+ Once a run completes, paste the measured medians into a `results/` file and the
27
+ UI/endpoint can surface them as **MEASURED** for that run only.
28
+
29
+ ## Citation (Ponytail, MIT)
30
+
31
+ a11oy Restraint **adopts** the 6-rung ladder and the lite/full/ultra intensity
32
+ levels from the open-source **Ponytail** coding-agent skill
33
+ (<https://github.com/DietrichGebert/ponytail>, MIT, © 2026 DietrichGebert) —
34
+ **adopted + governed, not invented here**. Ponytail's published results
35
+ (80–94% less code, 47–77% cheaper, 3–6× faster; median of 10 runs across
36
+ Haiku/Sonnet/Opus) are **cited as Ponytail's**, never claimed as ours. Our
37
+ contribution is the governance (signed DSSE receipts + advisory Λ), the
38
+ measured-on-our-stack benchmark, and the J/token energy tie-in.
benchmarks/restraint/promptfooconfig.yaml ADDED
@@ -0,0 +1,40 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # a11oy Restraint benchmark — Ponytail methodology PORTED to OUR stack.
2
+ #
3
+ # PROVENANCE (honest): this config is adapted from the open-source Ponytail
4
+ # skill's benchmark (github.com/DietrichGebert/ponytail/blob/main/benchmarks/
5
+ # promptfooconfig.yaml, MIT, © 2026 DietrichGebert). We reuse Ponytail's FIVE
6
+ # everyday tasks and its two-of-its-three arms idea, but we measure OUR arm
7
+ # (a11oy-restraint) on OUR models. We DO NOT reprint Ponytail's published
8
+ # numbers (80-94% less code, 47-77% cheaper, 3-6x faster, median across
9
+ # Haiku/Sonnet/Opus) as ours — those stay CITED as Ponytail's.
10
+ #
11
+ # Run: npx promptfoo@latest eval -c benchmarks/restraint/promptfooconfig.yaml --repeat 10
12
+ # View: npx promptfoo@latest view
13
+ # Needs a model key in the environment (set it, then the live UI label flips
14
+ # from SAMPLE/ROADMAP to MEASURED for the run you actually executed).
15
+ #
16
+ # Two arms ONLY (no-skill baseline vs a11oy-restraint). LOC is counted
17
+ # deterministically from fenced code blocks; tokens + latency come from the API.
18
+ description: "a11oy Restraint vs no-skill baseline: same model, same tasks. Measures code LOC (deterministic) and tokens/latency (API telemetry). Numbers are OURS, measured on OUR stack — never Ponytail's reprinted."
19
+
20
+ prompts:
21
+ - id: file://arms/baseline.txt
22
+ label: baseline (no skill)
23
+ - id: file://arms/a11oy-restraint.txt
24
+ label: a11oy-restraint
25
+
26
+ defaultTest:
27
+ assert:
28
+ - type: javascript
29
+ value: file://loc.js
30
+ metric: code_loc
31
+
32
+ # The same five everyday tasks Ponytail benchmarks (cited; tasks are facts, not
33
+ # Ponytail's measured outputs). Our restraint ladder is expected to stop high on
34
+ # most of these (stdlib / native / one-line).
35
+ tests:
36
+ - vars: { task: "Write me a Python function that validates email addresses." }
37
+ - vars: { task: "Add debounce to a search input in vanilla JavaScript. It currently fires an API call on every keystroke." }
38
+ - vars: { task: "Write Python code that reads sales.csv and sums the 'amount' column." }
39
+ - vars: { task: "Build me a countdown timer component in React that counts down from a given number of seconds." }
40
+ - vars: { task: "Add rate limiting to my FastAPI endpoint so users can't spam it." }
web/restraint.html ADDED
@@ -0,0 +1,307 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ <!DOCTYPE html>
2
+ <html lang="en">
3
+ <head>
4
+ <meta charset="UTF-8"/>
5
+ <meta name="viewport" content="width=device-width, initial-scale=1.0"/>
6
+ <title>a11oy Restraint — Governed + Measured Frugality Gate · SZL Holdings</title>
7
+ <meta name="description" content="a11oy Restraint: a governed, measured frugality gate for the a11oy Code agent. Before emitting a diff the agent descends a 6-rung ladder (YAGNI, stdlib, native, installed dep, one line, then minimal code) and stops at the first rung that holds, marking deliberate simplifications with restraint: ceiling comments. Every decision becomes a signed DSSE receipt and an advisory Lambda score; benchmark numbers are labelled MEASURED only when actually run, else SAMPLE/ROADMAP. The ladder + intensity levels are adopted from the open-source Ponytail skill (MIT); governance and measurement are ours."/>
8
+ <link rel="preconnect" href="https://fonts.googleapis.com"/><link rel="preconnect" href="https://fonts.gstatic.com" crossorigin/>
9
+ <link href="https://fonts.googleapis.com/css2?family=Space+Grotesk:wght@300;400;500;600;700&family=JetBrains+Mono:wght@400;500&display=swap" rel="stylesheet"/>
10
+ <style>
11
+ :root{
12
+ --ground:#0a0a0a; --panel:#0e0e0e; --panel2:#080808;
13
+ --gold:#c9b787; --gold-bright:#d6c69a;
14
+ --teal:#5fb3a3; --teal-soft:rgba(95,179,163,0.10);
15
+ --cream:#f5f5f5; --paragraph:#9a9a9a; --muted:#888; --dim:#555;
16
+ --gold-line:rgba(201,183,135,0.15); --gold-soft:rgba(201,183,135,0.04);
17
+ --teal-line:rgba(95,179,163,0.22);
18
+ --live:#5a8a6e; --err:#b06a5a;
19
+ --green:#39d98a; --yellow:#f5c451; --red:#ff6a5a; --blue:#5bc8ff;
20
+ --mono:'JetBrains Mono',ui-monospace,SFMono-Regular,monospace;
21
+ --display:'Space Grotesk',Georgia,serif;
22
+ }
23
+ *{box-sizing:border-box;}
24
+ html,body{margin:0;padding:0;background:var(--ground);color:var(--cream);font-family:var(--display);-webkit-font-smoothing:antialiased;}
25
+ .mono{font-family:var(--mono);}
26
+ a{color:inherit;text-decoration:none;}
27
+ code{font-family:var(--mono);color:var(--gold-bright);font-size:.9em;}
28
+ .topbar{position:sticky;top:0;z-index:60;display:flex;align-items:center;gap:1rem;flex-wrap:wrap;padding:.5rem 1.1rem;background:rgba(10,10,10,.92);backdrop-filter:blur(10px);border-bottom:1px solid var(--gold-line);font-family:var(--mono);font-size:10.5px;letter-spacing:.1em;text-transform:uppercase;color:var(--gold);}
29
+ .topbar .sep{color:var(--dim);}
30
+ .topbar .live{display:inline-flex;align-items:center;gap:.4rem;color:var(--cream);}
31
+ .live-dot{width:6px;height:6px;border-radius:50%;background:var(--live);box-shadow:0 0 6px var(--live);animation:pulse 2.2s ease-in-out infinite;}
32
+ @keyframes pulse{0%,100%{opacity:1;}50%{opacity:.35;}}
33
+ .switcher{margin-left:auto;display:flex;align-items:center;gap:.3rem;flex-wrap:wrap;}
34
+ .flag{padding:.22rem .55rem;border-radius:6px;border:1px solid transparent;color:var(--muted);transition:.15s;}
35
+ .flag:hover{color:var(--cream);border-color:var(--gold-line);background:var(--gold-soft);}
36
+ .flag.active{color:var(--ground);background:var(--gold);border-color:var(--gold);font-weight:600;}
37
+ .wrap{max-width:1320px;margin:0 auto;padding:1.6rem 1.4rem 4rem;}
38
+ h1{font-size:1.75rem;font-weight:600;margin:.2rem 0 .3rem;letter-spacing:-.01em;}
39
+ .badge{display:inline-block;font-family:var(--mono);font-size:10px;letter-spacing:.08em;text-transform:uppercase;color:var(--teal);border:1px solid var(--teal-line);background:var(--teal-soft);padding:.25rem .6rem;border-radius:5px;margin-left:.4rem;vertical-align:middle;}
40
+ .lede{color:var(--paragraph);font-size:.95rem;line-height:1.65;max-width:1180px;margin:.6rem 0 1rem;}
41
+ .lede b{color:var(--cream);font-weight:600;}
42
+ .honest{border:1px solid var(--gold-line);background:var(--gold-soft);border-radius:8px;padding:.7rem 1rem;margin:.4rem 0 1.2rem;font-family:var(--mono);font-size:11.5px;color:var(--paragraph);line-height:1.7;}
43
+ .honest .k{color:var(--gold);text-transform:uppercase;letter-spacing:.08em;font-size:10px;}
44
+ .honest .hi{color:var(--teal);}
45
+ .cards{display:grid;grid-template-columns:repeat(4,1fr);gap:.8rem;margin:0 0 1.2rem;}
46
+ .card{border:1px solid var(--gold-line);background:var(--panel);border-radius:8px;padding:.85rem 1rem;}
47
+ .card .lbl{font-family:var(--mono);font-size:9.5px;letter-spacing:.1em;text-transform:uppercase;color:var(--dim);}
48
+ .card .big{font-size:1.5rem;font-weight:600;margin:.35rem 0 .15rem;color:var(--teal);}
49
+ .card .sub{font-family:var(--mono);font-size:10.5px;color:var(--muted);}
50
+ .grid2{display:grid;grid-template-columns:1.1fr .9fr;gap:1rem;}
51
+ .panel{border:1px solid var(--gold-line);background:var(--panel);border-radius:10px;padding:1rem 1.1rem;margin-bottom:1rem;}
52
+ .panel h2{font-size:1.05rem;font-weight:600;margin:0 0 .2rem;display:flex;align-items:center;justify-content:space-between;gap:.5rem;}
53
+ .panel h2 .meta{font-family:var(--mono);font-size:10px;color:var(--dim);text-transform:uppercase;letter-spacing:.06em;}
54
+ .panel .desc{color:var(--paragraph);font-size:.85rem;line-height:1.55;margin:.1rem 0 .6rem;}
55
+ .kv{font-family:var(--mono);font-size:12px;color:var(--paragraph);line-height:1.9;margin-top:.4rem;}
56
+ .kv .k{color:var(--dim);display:inline-block;min-width:160px;}
57
+ .kv .v{color:var(--cream);}
58
+ .pill{display:inline-block;font-family:var(--mono);font-size:10px;letter-spacing:.06em;text-transform:uppercase;padding:.18rem .5rem;border-radius:5px;border:1px solid var(--gold-line);}
59
+ .pill.green{color:var(--green);border-color:rgba(57,217,138,.4);background:rgba(57,217,138,.08);}
60
+ .pill.yellow{color:var(--yellow);border-color:rgba(245,196,81,.4);background:rgba(245,196,81,.08);}
61
+ .pill.blue{color:var(--blue);border-color:rgba(91,200,255,.4);background:rgba(91,200,255,.08);}
62
+ input,textarea,select{font-family:var(--mono);font-size:12.5px;background:var(--panel2);color:var(--cream);border:1px solid var(--gold-line);border-radius:7px;padding:.55rem .7rem;width:100%;}
63
+ textarea{min-height:70px;resize:vertical;}
64
+ label.fld{display:block;font-family:var(--mono);font-size:10px;letter-spacing:.08em;text-transform:uppercase;color:var(--dim);margin:.6rem 0 .25rem;}
65
+ button{font-family:var(--mono);font-size:11.5px;letter-spacing:.06em;text-transform:uppercase;background:var(--gold);color:var(--ground);border:none;border-radius:7px;padding:.55rem 1rem;cursor:pointer;font-weight:600;transition:.15s;}
66
+ button:hover{background:var(--gold-bright);}
67
+ button.ghost{background:transparent;color:var(--gold);border:1px solid var(--gold-line);}
68
+ button.ghost:hover{background:var(--gold-soft);}
69
+ .seg{display:inline-flex;gap:.2rem;border:1px solid var(--gold-line);border-radius:7px;padding:.2rem;}
70
+ .seg button{background:transparent;color:var(--muted);padding:.35rem .7rem;}
71
+ .seg button.on{background:var(--gold);color:var(--ground);}
72
+ .ladder{display:flex;flex-direction:column;gap:.35rem;margin:.5rem 0;}
73
+ .rung{display:flex;align-items:flex-start;gap:.6rem;border:1px solid var(--gold-line);border-radius:7px;padding:.5rem .7rem;background:var(--panel2);font-size:12.5px;}
74
+ .rung .n{font-family:var(--mono);font-size:11px;color:var(--dim);min-width:1.4rem;}
75
+ .rung.held{border-color:var(--teal-line);background:var(--teal-soft);}
76
+ .rung.held .n{color:var(--teal);}
77
+ .rung .nm{color:var(--paragraph);}
78
+ .rung.held .nm{color:var(--cream);}
79
+ .rung .tag{margin-left:auto;font-family:var(--mono);font-size:9px;letter-spacing:.06em;text-transform:uppercase;color:var(--dim);}
80
+ .rung.held .tag{color:var(--teal);}
81
+ .ceil{font-family:var(--mono);font-size:12px;color:var(--gold-bright);background:var(--panel2);border:1px solid var(--gold-line);border-radius:7px;padding:.6rem .75rem;margin:.5rem 0;white-space:pre-wrap;word-break:break-word;}
82
+ table{width:100%;border-collapse:collapse;font-family:var(--mono);font-size:11.5px;margin-top:.5rem;}
83
+ th,td{text-align:right;padding:.4rem .5rem;border-bottom:1px solid var(--gold-line);}
84
+ th:first-child,td:first-child{text-align:left;}
85
+ th{color:var(--dim);text-transform:uppercase;font-size:9.5px;letter-spacing:.08em;font-weight:500;}
86
+ td{color:var(--paragraph);}
87
+ td.hi{color:var(--teal);}
88
+ .foot{font-family:var(--mono);font-size:10.5px;color:var(--dim);line-height:1.8;margin-top:1.5rem;border-top:1px solid var(--gold-line);padding-top:1rem;}
89
+ .foot a{color:var(--gold);}
90
+ .szl-pill{font-family:var(--mono);}
91
+ .spin{display:inline-block;width:11px;height:11px;border:2px solid var(--gold-line);border-top-color:var(--gold);border-radius:50%;animation:rot .8s linear infinite;vertical-align:middle;}
92
+ @keyframes rot{to{transform:rotate(360deg);}}
93
+ .sample-chips{display:flex;gap:.35rem;flex-wrap:wrap;margin:.5rem 0;}
94
+ .chip{font-family:var(--mono);font-size:10.5px;color:var(--muted);border:1px solid var(--gold-line);border-radius:20px;padding:.25rem .65rem;cursor:pointer;}
95
+ .chip:hover{color:var(--cream);background:var(--gold-soft);}
96
+ </style>
97
+ </head>
98
+ <body>
99
+ <div class="topbar">
100
+ <span class="live"><span class="live-dot"></span> a11oy</span>
101
+ <span class="sep">/</span><span>Restraint</span>
102
+ <span class="sep">/</span><span>Governed + Measured Frugality Gate</span>
103
+ <span class="switcher">
104
+ <a class="flag" href="/console">Console</a>
105
+ <a class="flag" href="/agent">Agent Loop</a>
106
+ <a class="flag" href="/nemo">Agent Model</a>
107
+ <a class="flag" href="/energy">Energy</a>
108
+ <a class="flag active" href="/restraint">Restraint</a>
109
+ </span>
110
+ </div>
111
+
112
+ <div class="wrap">
113
+ <h1>a11oy Restraint <span class="badge" id="b-doctrine">Doctrine v11</span><span id="b-heuristic"></span></h1>
114
+ <p class="lede">
115
+ Before the a11oy Code agent emits a diff, it descends a <b>6-rung ladder</b> and stops at the
116
+ first rung that holds — <b>(1) YAGNI</b>, (2) stdlib, (3) native platform feature,
117
+ (4) already-installed dependency, (5) one line, then (6) minimum viable code. Deliberate
118
+ simplifications are marked with <code>restraint:</code> ceiling comments that name the upgrade path.
119
+ Every decision becomes a <b>signed DSSE receipt</b> and an advisory <b>&Lambda;</b> trust score; the
120
+ benchmark reproduces code-reduction / cost / speed numbers on <b>our</b> stack, labelled
121
+ <b>MEASURED only when actually run</b>, else SAMPLE/ROADMAP.
122
+ </p>
123
+ <div class="honest">
124
+ <span class="k">Provenance — honest</span>&nbsp; The ladder and the <span class="hi">lite / full / ultra</span>
125
+ intensity levels are <b>adopted</b> from the open-source <b>Ponytail</b> coding-agent skill
126
+ (<a href="https://github.com/DietrichGebert/ponytail" target="_blank" rel="noopener" style="color:var(--gold)">github.com/DietrichGebert/ponytail</a>,
127
+ MIT, &copy; 2026 DietrichGebert) — <b>adopted + governed, not invented here</b>. Our additions are
128
+ the signed receipts, the &Lambda; score, the measured-on-our-stack benchmark, and the energy tie-in.
129
+ Ponytail's published numbers (<span class="hi">80&ndash;94% less code, 47&ndash;77% cheaper, 3&ndash;6&times; faster</span>,
130
+ median across Haiku/Sonnet/Opus) are <b>cited as Ponytail's</b>, never claimed as ours.
131
+ &Lambda; uniqueness is <b>Conjecture 1 (OPEN)</b>, advisory floor &lt; 1.0 — not a proven theorem.
132
+ </div>
133
+
134
+ <div class="cards" id="cards">
135
+ <div class="card"><div class="lbl">Ladder rungs</div><div class="big">6</div><div class="sub">stop at first that holds</div></div>
136
+ <div class="card"><div class="lbl">Intensity levels</div><div class="big">3</div><div class="sub">lite · full · ultra</div></div>
137
+ <div class="card"><div class="lbl">Per-decision receipt</div><div class="big">DSSE</div><div class="sub" id="c-signer">in-image ECDSA-P256</div></div>
138
+ <div class="card"><div class="lbl">Energy thesis</div><div class="big">J/tok</div><div class="sub">less code · fewer joules</div></div>
139
+ </div>
140
+
141
+ <div class="grid2">
142
+ <div class="panel">
143
+ <h2>Evaluate a task <span class="meta">POST /api/a11oy/v1/restraint/evaluate</span></h2>
144
+ <p class="desc">Type what the code agent is about to build. Restraint runs the pre-write ladder, names the ceiling, scores &Lambda;, estimates lines/tokens/joules saved, and signs the decision.</p>
145
+ <div class="sample-chips" id="chips"></div>
146
+ <label class="fld">Task</label>
147
+ <textarea id="task" placeholder="e.g. add a cache for these API responses">add a cache for these API responses</textarea>
148
+ <label class="fld">Intensity</label>
149
+ <div class="seg" id="seg">
150
+ <button data-i="lite">lite</button>
151
+ <button data-i="full" class="on">full</button>
152
+ <button data-i="ultra">ultra</button>
153
+ </div>
154
+ <div style="margin-top:.7rem;display:flex;gap:.5rem;">
155
+ <button id="run">Run ladder</button>
156
+ <button class="ghost" id="runbench">Run benchmark</button>
157
+ </div>
158
+ </div>
159
+
160
+ <div class="panel">
161
+ <h2>Ladder decision <span class="meta" id="dec-label"></span></h2>
162
+ <p class="desc" id="dec-summary">Run the ladder to see which rung stopped it and why.</p>
163
+ <div class="ladder" id="ladder"></div>
164
+ <div class="ceil" id="ceil" style="display:none"></div>
165
+ <div class="kv" id="dec-kv"></div>
166
+ </div>
167
+ </div>
168
+
169
+ <div class="grid2">
170
+ <div class="panel">
171
+ <h2>Signed receipt <span class="meta">in-image ECDSA-P256 / DSSE</span></h2>
172
+ <p class="desc">The governed differentiator: every restraint decision is a tamper-evident DSSE receipt. Verify in-browser against <code>/cosign.pub</code>.</p>
173
+ <div class="kv" id="receipt-kv"><span class="k">status</span> <span class="v">— run the ladder —</span></div>
174
+ </div>
175
+ <div class="panel">
176
+ <h2>Energy tie-in <span class="meta" id="energy-label"></span></h2>
177
+ <p class="desc">Less code &rarr; fewer tokens &rarr; fewer joules on our GPU. Estimates are MODELED; the joules label defers to the on-box NVML meter (reads <code>sample</code> unless a fresh exporter reading is present).</p>
178
+ <div class="kv" id="energy-kv"><span class="k">tokens saved</span> <span class="v">—</span></div>
179
+ </div>
180
+ </div>
181
+
182
+ <div class="panel">
183
+ <h2>Benchmark — baseline vs a11oy-restraint <span class="meta" id="bench-label">GET /api/a11oy/v1/restraint/bench</span></h2>
184
+ <p class="desc">Ported from Ponytail's promptfoo methodology (MIT): the same five everyday tasks, two arms, median reported. <b>Numbers are SAMPLE/ROADMAP on this Space</b> (no model run wired here) and flip to <b>MEASURED</b> only when a model run is supplied. We measure our own; we never reprint Ponytail's numbers as ours.</p>
185
+ <table id="bench"><thead><tr><th>Task</th><th>Rung</th><th>Baseline LOC</th><th>Restraint LOC</th><th>LOC &minus;%</th><th>Cost-proxy &minus;%</th><th>Latency &minus;%</th><th>Label</th></tr></thead><tbody><tr><td colspan="8" style="text-align:center;color:var(--dim)">Run benchmark to populate.</td></tr></tbody></table>
186
+ <div class="kv" id="bench-agg" style="margin-top:.7rem"></div>
187
+ </div>
188
+
189
+ <div class="foot" id="foot">
190
+ <div>a11oy Restraint &middot; Doctrine v11 (locked=8 @ c7c0ba17) &middot; &Lambda; = Conjecture 1 (OPEN, advisory floor &lt; 1.0) &middot; SLSA L1 honest, L2/L3 roadmap &middot; 0 runtime CDN &middot; signed receipts.</div>
191
+ <div>Adopted + governed from <a href="https://github.com/DietrichGebert/ponytail" target="_blank" rel="noopener">Ponytail</a> (MIT, &copy; 2026 DietrichGebert). Endpoints: <code>/api/a11oy/v1/restraint/{evaluate,bench,info}</code>.</div>
192
+ </div>
193
+ </div>
194
+
195
+ <script src="/static/shared/szl_label_engine.js"></script>
196
+ <script src="/static/shared/szl_receipt_cosign.js"></script>
197
+ <script>
198
+ "use strict";
199
+ var INTENSITY = "full";
200
+ function $(s){return document.querySelector(s);}
201
+ function el(t,c,h){var e=document.createElement(t);if(c)e.className=c;if(h!=null)e.innerHTML=h;return e;}
202
+ function pill(key){ try{ if(window.SZLLabels&&window.SZLLabels.badgeHTML){return window.SZLLabels.badgeHTML(key);} }catch(e){} return '<span class="szl-pill pill">'+key+'</span>'; }
203
+ async function getJSON(u,opt){try{var r=await fetch(u,opt);if(!r.ok)return null;return await r.json();}catch(e){return null;}}
204
+ function esc(s){return String(s==null?'':s).replace(/[&<>]/g,function(m){return {'&':'&amp;','<':'&lt;','>':'&gt;'}[m];});}
205
+
206
+ // honest badges via window.SZLLabels
207
+ $('#b-heuristic').innerHTML = pill('HEURISTIC');
208
+ $('#bench-label').innerHTML = 'GET /api/a11oy/v1/restraint/bench &nbsp;' + pill('ROADMAP');
209
+ $('#energy-label').innerHTML = pill('MODELED');
210
+
211
+ var CHIPS = ["add a cache for these API responses","validate an email address","add a date picker",
212
+ "build a generic pluggable plugin framework for later","sort this list of names",
213
+ "add rate limiting to my FastAPI endpoint","add debounce to a search input in vanilla JS"];
214
+ (function(){var c=$('#chips');CHIPS.forEach(function(t){var b=el('span','chip',esc(t));b.onclick=function(){$('#task').value=t;run();};c.appendChild(b);});})();
215
+
216
+ document.querySelectorAll('#seg button').forEach(function(b){
217
+ b.onclick=function(){document.querySelectorAll('#seg button').forEach(function(x){x.classList.remove('on');});b.classList.add('on');INTENSITY=b.getAttribute('data-i');run();};
218
+ });
219
+
220
+ function renderLadder(trail, stopped){
221
+ var L=$('#ladder');L.innerHTML='';
222
+ (trail||[]).forEach(function(r){
223
+ var held=r.rung===stopped;
224
+ var row=el('div','rung'+(held?' held':''));
225
+ row.innerHTML='<span class="n">'+r.rung+'</span><span class="nm">'+esc(r.name)+'</span>'+
226
+ '<span class="tag">'+(held?'STOPPED HERE':(r.held?'holds':'skip'))+'</span>';
227
+ L.appendChild(row);
228
+ });
229
+ }
230
+
231
+ async function run(){
232
+ var task=$('#task').value.trim(); if(!task)return;
233
+ $('#dec-summary').innerHTML='<span class="spin"></span> running ladder…';
234
+ $('#dec-label').innerHTML='';
235
+ var d=await getJSON('/api/a11oy/v1/restraint/evaluate',{method:'POST',headers:{'content-type':'application/json'},body:JSON.stringify({task:task,intensity:INTENSITY,lang:'py'})});
236
+ if(!d){$('#dec-summary').textContent='endpoint unavailable (is the Space running?)';return;}
237
+ $('#dec-label').innerHTML=pill(d.label||'HEURISTIC');
238
+ $('#dec-summary').innerHTML='Stopped at rung <b style="color:var(--teal)">'+d.stopped_at_rung+' — '+esc(d.rung_key)+'</b>. '+esc(d.answer);
239
+ renderLadder(d.ladder_trail,d.stopped_at_rung);
240
+ var ce=$('#ceil');ce.style.display='block';ce.textContent=d.restraint_comment;
241
+ var lam=(d.lambda_score&&d.lambda_score.lambda)||0;
242
+ var saved=d.lines_saved_estimate||{};
243
+ $('#dec-kv').innerHTML=
244
+ '<div><span class="k">Λ (Conjecture 1, &lt;1.0)</span><span class="v">'+lam.toFixed(3)+(lam<1?' ✓ below floor':'')+'</span></div>'+
245
+ '<div><span class="k">lines saved (modeled)</span><span class="v">≈'+(saved.lines_saved_modeled||0)+' LOC ('+Math.round((saved.reduction_fraction||0)*100)+'%)</span> '+pill(saved.label||'MODELED')+'</div>'+
246
+ '<div><span class="k">never simplify away</span><span class="v">'+esc((d.never_simplify||[]).join(', '))+'</span></div>'+
247
+ '<div><span class="k">intensity</span><span class="v">'+esc(d.intensity)+'</span></div>';
248
+
249
+ // receipt panel
250
+ var rc=d.signed_receipt;var rk=$('#receipt-kv');
251
+ if(rc){
252
+ var signed=rc.signed===true;
253
+ rk.innerHTML='<div><span class="k">signed</span><span class="v">'+(signed?pill('LIVE')+' yes':'<span class="pill yellow">UNSIGNED</span> '+esc(rc.honesty||''))+'</span></div>'+
254
+ '<div><span class="k">payloadType</span><span class="v">'+esc(rc.payloadType||'—')+'</span></div>'+
255
+ '<div><span class="k">PAE sha256</span><span class="v">'+esc((rc._pae_sha256||'').slice(0,32))+'…</span></div>'+
256
+ '<div><span class="k">keyid</span><span class="v">'+esc((rc.signatures&&rc.signatures[0]&&rc.signatures[0].keyid)||'—')+'</span></div>'+
257
+ '<div><span class="k">verify</span><span class="v">against <code>/cosign.pub</code> in-browser</span></div>';
258
+ if(rc.signatures&&rc.signatures.length&&window.SZLReceipts&&window.SZLReceipts.verifyEnvelope){
259
+ try{var v=await window.SZLReceipts.verifyEnvelope(rc,{});if(v){rk.insertAdjacentHTML('beforeend','<div><span class="k">browser verify</span><span class="v">'+(v.valid?'✓ valid':'✗ '+esc(v.reason||''))+'</span></div>');}}catch(e){}
260
+ }
261
+ } else { rk.innerHTML='<span class="k">status</span> <span class="v">no signer wired</span>'; }
262
+
263
+ // energy panel
264
+ var en=d.energy_tiein||{};
265
+ $('#energy-kv').innerHTML=
266
+ '<div><span class="k">tokens saved (modeled)</span><span class="v">≈'+(en.tokens_saved_modeled||0)+'</span></div>'+
267
+ '<div><span class="k">joules saved (modeled)</span><span class="v">≈'+(en.joules_saved_modeled||0)+' J</span> '+pill('MODELED')+'</div>'+
268
+ '<div><span class="k">J / output token</span><span class="v">'+(en.j_per_output_token_modeled||0)+' (modeled)</span></div>'+
269
+ '<div><span class="k">joules meter label</span><span class="v">'+pill(((en.joules_label||'sample')).toUpperCase())+'</span></div>'+
270
+ '<div><span class="k">thesis</span><span class="v">'+esc(en.thesis||'')+'</span></div>';
271
+ }
272
+
273
+ async function runbench(){
274
+ var tb=$('#bench').querySelector('tbody');
275
+ tb.innerHTML='<tr><td colspan="8" style="text-align:center;color:var(--dim)"><span class="spin"></span> running…</td></tr>';
276
+ var b=await getJSON('/api/a11oy/v1/restraint/bench?intensity='+INTENSITY);
277
+ if(!b){tb.innerHTML='<tr><td colspan="8">endpoint unavailable</td></tr>';return;}
278
+ $('#bench-label').innerHTML='GET /api/a11oy/v1/restraint/bench &nbsp;'+pill(b.overall_label||'ROADMAP');
279
+ tb.innerHTML='';
280
+ (b.rows||[]).forEach(function(r){
281
+ var tr=el('tr');
282
+ tr.innerHTML='<td title="'+esc(r.task)+'">'+esc(r.task.slice(0,42))+(r.task.length>42?'…':'')+'</td>'+
283
+ '<td>'+r.stopped_at_rung+'</td>'+
284
+ '<td>'+r.baseline.loc+'</td><td class="hi">'+r.a11oy_restraint.loc+'</td>'+
285
+ '<td class="hi">'+r.loc_reduction_pct+'%</td>'+
286
+ '<td>'+r.cost_proxy_reduction_pct+'%</td>'+
287
+ '<td>'+r.latency_reduction_pct+'%</td>'+
288
+ '<td>'+pill(r.label)+'</td>';
289
+ tb.appendChild(tr);
290
+ });
291
+ var a=b.aggregate||{};var pp=b.ponytail_published||{};
292
+ $('#bench-agg').innerHTML=
293
+ '<div><span class="k">median LOC reduction</span><span class="v">'+a.median_loc_reduction_pct+'% '+pill(b.overall_label)+'</span></div>'+
294
+ '<div><span class="k">median cost-proxy reduction</span><span class="v">'+a.median_cost_proxy_reduction_pct+'%</span></div>'+
295
+ '<div><span class="k">median latency reduction</span><span class="v">'+a.median_latency_reduction_pct+'%</span></div>'+
296
+ '<div><span class="k">Ponytail (cited, not ours)</span><span class="v">'+esc(pp.code_reduction||'')+' · '+esc(pp.cost_reduction||'')+' · '+esc(pp.speed||'')+' '+pill('CITED')+'</span></div>'+
297
+ '<div style="color:var(--dim);font-size:11px;margin-top:.4rem">'+esc(b.honesty||'')+'</div>';
298
+ }
299
+
300
+ $('#run').onclick=run;
301
+ $('#runbench').onclick=runbench;
302
+ // info → signer label
303
+ (async function(){var i=await getJSON('/api/a11oy/v1/restraint/info');if(i&&i.doctrine){$('#b-doctrine').textContent='Doctrine '+i.doctrine.version;$('#c-signer').textContent='in-image ECDSA-P256';}})();
304
+ run();
305
+ </script>
306
+ </body>
307
+ </html>