Draft signed path proof receipt

docs/path-proof-receipt.md

@@ -0,0 +1,98 @@
+# Path Proof Receipt v0
+
+## Decision
+
+The smallest useful receipt is a **runtime-signed commitment to an action and its observed execution path**. It must be emitted by a trusted recorder at the model/tool boundary, not composed by the agent whose claim is being checked.
+
+A self-reported model name or an unsigned transcript is provenance metadata, not proof.
+
+## Receipt
+
+The wire form is RFC 8785 canonical JSON. Hashes are lowercase SHA-256 hex. The signature covers the canonical JSON object with `attestation.sig` omitted.
+
+```json
+{
+  "v": 0,
+  "action": {
+    "kind": "gitea.issue.labels.replace",
+    "target": "stackchain/stackchain-lab-loop#3",
+    "request_sha256": "<sha256-of-exact-action-request>",
+    "result_sha256": "<sha256-of-exact-observed-result>"
+  },
+  "path": {
+    "model": {
+      "id": "<provider/model-version>",
+      "invocation": "<provider-or-gateway-invocation-id>"
+    },
+    "tools": {
+      "count": 1,
+      "trace_sha256": "<sha256-of-canonical-ordered-tool-events>"
+    },
+    "policy": {
+      "bundle_sha256": "<sha256-of-effective-policy-bundle>"
+    },
+    "code": {
+      "repo": "<immutable-repository-identity>",
+      "commit": "<full-commit-id>",
+      "entrypoint": "<path-or-image-digest>"
+    }
+  },
+  "evidence": {
+    "uri": "sha256:<content-addressed-evidence-bundle-digest>"
+  },
+  "attestation": {
+    "issuer": "<runtime-recorder-identity>",
+    "key_id": "<verification-key-id>",
+    "alg": "Ed25519",
+    "sig": "<base64url-signature>"
+  }
+}
+```
+
+`tools.trace_sha256` commits to a canonical ordered array of `{seq, tool, request_sha256, response_sha256}` records. The evidence bundle contains the exact request/result bytes, ordered tool records, effective policy bundle, model gateway record, and code checkout/image metadata. Secrets may be replaced by typed redaction commitments before hashing; the redaction manifest is part of the bundle.
+
+## Why each field survives minimization
+
+- `action` binds the receipt to the consequential effect rather than merely to an agent run.
+- `model.id` plus `model.invocation` lets the recorder correlate the claim with gateway/provider evidence. A model string alone proves nothing.
+- `tools.trace_sha256` commits to the complete ordered tool path without inflating the receipt.
+- `policy.bundle_sha256` identifies the policy actually loaded, including system/developer/project rules after assembly—not just a policy version label.
+- `code.commit` and `entrypoint` identify both source and the invoked path. For non-Git execution, `entrypoint` must be an immutable image/module digest.
+- `evidence.uri` makes all commitments independently retrievable and rejects a receipt whose evidence was discarded.
+- `attestation` establishes who observed these facts and detects mutation.
+
+Removing any one of these leaves one requested claim—actual model, tool, policy, code path, or action—unbound or unverifiable.
+
+## Recorder requirements
+
+1. The recorder is outside the agent process and intercepts model calls, tool dispatch/results, policy assembly, and the final action result.
+2. It derives model identity from the authenticated gateway/provider response, never from assistant text.
+3. It hashes exact bytes with media type and canonicalization rules recorded in the evidence bundle.
+4. It emits the receipt only after observing the consequential result and stores the evidence bundle under its digest.
+5. Its signing key is hardware- or service-isolated from agent-controlled tools. Key metadata states issuer, validity interval, and revocation status.
+
+## Verification algorithm
+
+A verifier MUST:
+
+1. Resolve a trusted key for `attestation.issuer` and `key_id`; reject revoked or out-of-validity keys.
+2. Recreate RFC 8785 canonical bytes without `attestation.sig` and verify Ed25519.
+3. Fetch `evidence.uri`, hash it, and require equality with the URI digest.
+4. Recompute the action request/result, ordered tool trace, and effective policy hashes from evidence.
+5. Verify the model invocation against authenticated gateway/provider evidence.
+6. Resolve the immutable code identity and require that `entrypoint` belongs to that checkout/image.
+7. Apply action-specific checks (for example, read Gitea state back and compare it with `result_sha256`).
+
+Any missing evidence or unsupported identity yields **unverified**, not partial success.
+
+## Threat boundary
+
+This receipt proves only what the trusted recorder observed. It does not prove the provider ran the advertised weights unless the provider supplies a separately trusted model attestation, and it cannot survive compromise of both recorder and signing key. Those limitations must be reported explicitly rather than papered over with transcript hashes.
+
+## Adversarial-review questions
+
+1. Can the `model.invocation` be replayed or bound to a different request?
+2. Is one `policy.bundle_sha256` sufficient to represent dynamic policy/tool authorization changes during the run?
+3. What canonical evidence-bundle format permits selective disclosure without making redactions ambiguous?
+4. Should code provenance require a signed build/SLSA subject instead of a Git commit for production actions?
+5. Which action-specific freshness field (trusted timestamp, nonce, or transparency-log index) is the minimum needed to prevent receipt replay?

media/x-teaser/README.md

@@ -1,58 +0,0 @@
-# Season Zero — work-derived X recut
-
-A deterministic 30-second vertical teaser replacing the rejected abstract visualizer with a legible evidence chain:
-
-1. the live Stackchain Lab issue queue rendered as a work board, plus Season Zero epic #7;
-2. Vincent's issue #1 contribution receipt and independent `8 TESTS PASS` verification;
-3. the verified 2026-08-07 Daily Drop artwork and Slop Cannon scene 04 as transformation inputs;
-4. the four Episode 01 mutation cards extracted from the verified pilot render;
-5. the audience gate: **WHICH TIMMY GETS TO REMEMBER THIS?**
-
-The persistent pipeline rail, moving scan, board drift, source pan/zoom, transfer beam, card reveals, and orbiting final forms make one continuous transformation rather than a static screenshot slideshow.
-
-## Artifacts
-
-- [`stackchain-season-zero-work-recut.mp4`](stackchain-season-zero-work-recut.mp4) — delivery master
-- [`contact-sheet.jpg`](contact-sheet.jpg) — six-frame review strip
-- [`stills/`](stills/) — five representative full-size review frames
-- [`build_work_recut.py`](build_work_recut.py) — deterministic renderer and original synthesized soundtrack
-- [`source/`](source/) — checksummed work-derived image inputs
-
-## Revised X copy — 216 characters
-
-> Real work enters the Slop Cannon: a live Timmy/Vincent queue, contribution receipts, 8/8 verification, Daily Drop art, and four mutations.
->
-> Wizard. Machine. Creature. Talking Turd.
->
-> Which Timmy gets to remember this?
-
-**Do not publish without approval in human gate #15.**
-
-## Verification
-
-```text
-container: MP4
-video: H.264, 720x1280 (9:16), yuv420p, 30 fps
-audio: AAC, 48 kHz, mono
-duration: 30.000000 seconds
-size: 5,325,465 bytes
-full decode: pass
-encoded audio: -20.1 dB mean, -2.2 dB max
-sha256: 2436d193e5d7ee0660318f585d5b84356cbf623a034d63355ed9232258c1f08d
-```
-
-Re-run:
-
-```bash
-python3 media/x-teaser/build_work_recut.py
-ffprobe -v error -show_entries format=duration,size:stream=codec_name,codec_type,width,height,pix_fmt,r_frame_rate,sample_rate,channels -of json media/x-teaser/stackchain-season-zero-work-recut.mp4
-ffmpeg -v error -i media/x-teaser/stackchain-season-zero-work-recut.mp4 -f null -
-sha256sum media/x-teaser/stackchain-season-zero-work-recut.mp4
-```
-
-## Input provenance
-
-- `daily-drop-door-phone.png` is the verified Daily Drop preview at `/root/daily-drop/drops/2026-08-07/the-door-that-refused-a-master-key-phone.png`; SHA-256 `9a72358821414328de129a96461b1f4265be356987f4009037e5ae2f0cab138a`.
-- `cannon-scene-04.png` is scene 04 from `/root/tiktok-slop-cannon/drops/2026-08-07-1786114450/`; SHA-256 `3080c4135c1595251097387b1e9b8a6022688c3f3ed3db4a919ed9e347d51a15`, matching its source manifest.
-- The four mutation images were extracted at 5s, 9s, 13s, and 17s from `episode01-four-timmy-mutations.mp4` on branch `timmy/8-four-timmy-mutations`, commit `afa304dc6b5d9d3cc0ede17d6f4f0dfb67fc6b70`; source MP4 SHA-256 `5ddc9a9da1b4b2456faf0b390e5c1be0d3cf48112b417fc075473c90e335d1e4`.
-- Board, epic, receipt, and test copy are drawn from Gitea issues #7 and #1 and their complete comments read before rendering. The work-board composition is an editorial rendering of that live queue state, not a claim that Gitea supplied a screenshot endpoint.

media/x-teaser/build_work_recut.py

@@ -1,286 +0,0 @@
-#!/usr/bin/env python3
-"""Deterministically build the work-derived Season Zero X teaser."""
-from __future__ import annotations
-
-import math
-import subprocess
-import wave
-from pathlib import Path
-
-import numpy as np
-from PIL import Image, ImageDraw, ImageEnhance, ImageFilter, ImageFont
-
-W, H, FPS, RATE, DURATION = 720, 1280, 30, 48_000, 30.0
-ROOT = Path(__file__).resolve().parent
-SOURCE = ROOT / "source"
-OUT = ROOT / "stackchain-season-zero-work-recut.mp4"
-TMP_VIDEO = ROOT / ".video.mp4"
-TMP_AUDIO = ROOT / ".audio.wav"
-BOLD = "/usr/share/fonts/truetype/dejavu/DejaVuSans-Bold.ttf"
-REG = "/usr/share/fonts/truetype/dejavu/DejaVuSans.ttf"
-COLORS = {"ink": "#070b14", "panel": "#111827", "line": "#334155", "white": "#f8fafc", "muted": "#a9b7cc", "gold": "#ffc857", "cyan": "#5ee7f2", "violet": "#c69cff", "green": "#8ff0a4", "orange": "#ff9c66"}
-ASSETS = {name: Image.open(SOURCE / filename).convert("RGB") for name, filename in {
-    "drop": "daily-drop-door-phone.png", "cannon": "cannon-scene-04.png",
-    "wizard": "wizard.png", "machine": "machine.png", "creature": "creature.png", "turd": "talking-turd.png",
-}.items()}
-
-
-def font(size: int, bold: bool = False):
-    return ImageFont.truetype(BOLD if bold else REG, size)
-
-
-def ease(x: float) -> float:
-    x = max(0.0, min(1.0, x))
-    return x * x * (3 - 2 * x)
-
-
-def fit_crop(image: Image.Image, box: tuple[int, int, int, int], zoom: float = 1.0, pan_y: float = 0.5) -> Image.Image:
-    x0, y0, x1, y1 = box
-    bw, bh = x1-x0, y1-y0
-    scale = max(bw/image.width, bh/image.height) * zoom
-    rw, rh = max(1, int(image.width*scale)), max(1, int(image.height*scale))
-    resized = image.resize((rw, rh), Image.Resampling.LANCZOS)
-    left = max(0, (rw-bw)//2)
-    top = max(0, min(rh-bh, int((rh-bh)*pan_y)))
-    return resized.crop((left, top, left+bw, top+bh))
-
-
-def rounded_image(canvas: Image.Image, image: Image.Image, box: tuple[int, int, int, int], radius: int = 24, zoom: float = 1.0, pan_y: float = 0.5, border: str = COLORS["line"]):
-    crop = fit_crop(image, box, zoom, pan_y)
-    mask = Image.new("L", crop.size, 0)
-    ImageDraw.Draw(mask).rounded_rectangle((0, 0, crop.width-1, crop.height-1), radius, fill=255)
-    canvas.paste(crop, (box[0], box[1]), mask)
-    d = ImageDraw.Draw(canvas)
-    d.rounded_rectangle(box, radius, outline=border, width=3)
-
-
-def text(d: ImageDraw.ImageDraw, xy, value: str, size: int, fill: str = COLORS["white"], bold: bool = False, anchor: str | None = None):
-    d.text(xy, value, font=font(size, bold), fill=fill, anchor=anchor)
-
-
-def centered(d, value: str, y: int, size: int, fill: str = COLORS["white"], bold: bool = False):
-    text(d, (W//2, y), value, size, fill, bold, "ma")
-
-
-def base(t: float) -> Image.Image:
-    y = np.linspace(0, 1, H)[:, None, None]
-    top = np.array((5, 10, 20))[None, None, :]
-    bottom = np.array((22, 15, 40))[None, None, :]
-    grad = top*(1-y) + bottom*y
-    xwave = 5*np.sin(np.linspace(0, 9, W)[None, :, None] + t*.6)
-    arr = np.broadcast_to(grad, (H, W, 3)) + xwave
-    im = Image.fromarray(np.clip(arr, 0, 255).astype(np.uint8), "RGB")
-    d = ImageDraw.Draw(im)
-    for i in range(7):
-        yy = int((t*52 + i*195) % (H+160))-80
-        d.line((0, yy, W, yy-120), fill=(25, 42, 65), width=2)
-    return im
-
-
-def chrome(d: ImageDraw.ImageDraw, t: float, active: int):
-    d.rectangle((0, 0, W, 76), fill="#080d18")
-    d.line((0, 76, W, 76), fill=COLORS["line"], width=2)
-    text(d, (32, 23), "STACKCHAIN LAB", 25, COLORS["white"], True)
-    text(d, (W-32, 25), "SEASON ZERO", 20, COLORS["gold"], True, "ra")
-    nodes = [(64, "WORK"), (215, "CHALLENGE"), (390, "MUTATE"), (563, "AUDIENCE")]
-    y = 1208
-    d.line((64, y, 656, y), fill=COLORS["line"], width=5)
-    progress = min(1, max(0, t/DURATION))
-    d.line((64, y, 64+592*progress, y), fill=COLORS["cyan"], width=5)
-    for i, (x, label) in enumerate(nodes):
-        c = COLORS["gold"] if i == active else (COLORS["cyan"] if i < active else COLORS["line"])
-        d.ellipse((x-9, y-9, x+9, y+9), fill=c)
-        text(d, (x, y+27), label, 15, c, True, "ma")
-    scan = int((t*155) % (H-160)) + 80
-    d.rectangle((0, scan, W, scan+2), fill="#5ee7f222")
-
-
-def issue_card(d, box, number, title, owner, state, accent):
-    d.rounded_rectangle(box, 14, fill=COLORS["panel"], outline=accent, width=2)
-    x0, y0, x1, _ = box
-    text(d, (x0+15, y0+13), f"#{number}", 19, accent, True)
-    text(d, (x1-15, y0+15), state.upper(), 16, COLORS["muted"], True, "ra")
-    words = title.split()
-    lines, line = [], ""
-    for word in words:
-        trial = (line+" "+word).strip()
-        if d.textlength(trial, font=font(19, True)) > x1-x0-30:
-            lines.append(line); line = word
-        else: line = trial
-    lines.append(line)
-    for i, ln in enumerate(lines[:2]): text(d, (x0+15, y0+46+i*25), ln, 19, COLORS["white"], True)
-    text(d, (x0+15, y0+104), f"agent:{owner}", 16, accent)
-
-
-def scene_board(im: Image.Image, t: float):
-    d = ImageDraw.Draw(im)
-    local = t
-    centered(d, "THE WORK IS PUBLIC", 112, 35, COLORS["white"], True)
-    centered(d, "A LIVE QUEUE. A REAL HANDOFF.", 158, 21, COLORS["cyan"], True)
-    sway = int(7*math.sin(local*1.8))
-    d.rounded_rectangle((28+sway, 205, 692+sway, 1015), 24, fill="#0b1220", outline=COLORS["line"], width=3)
-    text(d, (52+sway, 228), "stackchain / stackchain-lab-loop", 20, COLORS["muted"], True)
-    for x, label, c in [(50, "READY", COLORS["cyan"]), (274, "CLAIMED", COLORS["gold"]), (498, "REVIEW", COLORS["green"])]:
-        text(d, (x+sway, 281), label, 17, c, True)
-    issue_card(d, (48+sway, 320, 254+sway, 455), 7, "The Cannon Needs a Pilot", "vincent", "ready", COLORS["violet"])
-    issue_card(d, (48+sway, 475, 254+sway, 610), 13, "Attack the Cannon Chain", "vincent", "ready", COLORS["cyan"])
-    issue_card(d, (274+sway, 320, 480+sway, 455), 16, "Build the X teaser", "timmy", "claimed", COLORS["gold"])
-    issue_card(d, (498+sway, 320, 672+sway, 455), 1, "Prove the Lab Loop", "vincent", "review", COLORS["green"])
-    d.rounded_rectangle((54+sway, 670, 666+sway, 962), 18, fill="#16112a", outline=COLORS["violet"], width=3)
-    text(d, (78+sway, 694), "EPIC #7  //  SEASON ZERO", 21, COLORS["violet"], True)
-    text(d, (78+sway, 744), "THE CANNON NEEDS A PILOT", 28, COLORS["white"], True)
-    text(d, (78+sway, 801), "6-episode character evolution chain", 19, COLORS["muted"])
-    for i, line in enumerate(("✓ persistent Timmy", "✓ power / scar / cost / receipt", "✓ audience choice advances canon")):
-        text(d, (90+sway, 846+i*33), line, 19, COLORS["green"] if i < 2 else COLORS["gold"], True)
-    text(d, (50, 1060), "TIMMY  →  VINCENT  →  HUMAN REVIEW", 24, COLORS["gold"], True)
-
-
-def scene_receipt(im: Image.Image, t: float):
-    d = ImageDraw.Draw(im)
-    local = t-5
-    centered(d, "CHALLENGED. THEN VERIFIED.", 116, 31, COLORS["white"], True)
-    x = 38 + int(8*math.sin(local*2))
-    d.rounded_rectangle((x, 205, W-38+x-38, 945), 22, fill="#0c1423", outline=COLORS["cyan"], width=3)
-    text(d, (x+24, 230), "ISSUE #1  /  CONTRIBUTION RECEIPT", 19, COLORS["cyan"], True)
-    d.rounded_rectangle((x+22, 278, W-62, 520), 18, fill=COLORS["panel"], outline=COLORS["line"], width=2)
-    text(d, (x+44, 302), "VINCENT", 23, COLORS["violet"], True)
-    text(d, (W-84, 307), "INDEPENDENT PASS", 16, COLORS["muted"], True, "ra")
-    text(d, (x+44, 355), "Fetched PR #4 from Timmy's branch", 20, COLORS["white"])
-    text(d, (x+44, 397), "8 TESTS PASS", 39, COLORS["green"], True)
-    text(d, (x+44, 452), "Label transition regression verified", 20, COLORS["muted"])
-    d.rounded_rectangle((x+22, 555, W-62, 804), 18, fill="#17122b", outline=COLORS["violet"], width=2)
-    text(d, (x+44, 580), "DEFINITION OF DONE", 19, COLORS["violet"], True)
-    checks = ("Both agents read assigned tickets", "Both leave verified evidence", "Timmy → Vincent → review", "Credentials remain host-local")
-    for i, line in enumerate(checks):
-        text(d, (x+48, 633+i*40), "✓", 22, COLORS["green"], True)
-        text(d, (x+82, 635+i*40), line, 19, COLORS["white"])
-    text(d, (x+24, 856), "RECEIPT: ISSUE #1  •  PR #4  •  8/8", 21, COLORS["gold"], True)
-    centered(d, "CRITIQUE BECOMES A BUILD INPUT.", 1012, 23, COLORS["gold"], True)
-    centered(d, "NOT A COMMENT. A CONSEQUENCE.", 1055, 19, COLORS["muted"], True)
-
-
-def scene_sources(im: Image.Image, t: float):
-    d = ImageDraw.Draw(im)
-    local = t-10
-    centered(d, "REAL ARTIFACTS ENTER THE CHAIN", 112, 29, COLORS["white"], True)
-    shift = int(10*math.sin(local*.9))
-    rounded_image(im, ASSETS["drop"], (35+shift, 205, 350+shift, 930), 22, 1.03+local*.008, .46, COLORS["gold"])
-    rounded_image(im, ASSETS["cannon"], (370-shift, 205, 685-shift, 930), 22, 1.05+local*.01, .5, COLORS["orange"])
-    d = ImageDraw.Draw(im)
-    d.rounded_rectangle((48+shift, 835, 337+shift, 913), 12, fill="#070b14dd")
-    text(d, (65+shift, 852), "DAILY DROP", 19, COLORS["gold"], True)
-    text(d, (65+shift, 881), "THE DOOR / VERIFIED", 16, COLORS["white"], True)
-    d.rounded_rectangle((383-shift, 835, 672-shift, 913), 12, fill="#070b14dd")
-    text(d, (400-shift, 852), "CANNON OUTPUT", 19, COLORS["orange"], True)
-    text(d, (400-shift, 881), "SCENE 04 / SOURCE", 16, COLORS["white"], True)
-    # Animated transfer beam unifies the two work artifacts.
-    beam = int(95 + (local % 2.0)/2.0*520)
-    d.line((beam, 965, beam+70, 965), fill=COLORS["cyan"], width=7)
-    d.polygon([(beam+70, 953), (beam+94, 965), (beam+70, 977)], fill=COLORS["cyan"])
-    centered(d, "SOURCE  →  PRESSURE  →  MUTATION", 1025, 23, COLORS["cyan"], True)
-    centered(d, "THE ART CHANGES JOBS. IT ISN'T FILLER.", 1070, 18, COLORS["muted"], True)
-
-
-def scene_mutations(im: Image.Image, t: float):
-    d = ImageDraw.Draw(im)
-    local = t-16
-    centered(d, "ONE PILOT. FOUR CONSEQUENCES.", 108, 29, COLORS["white"], True)
-    specs = [
-        ("wizard", "WIZARD", "AMBIGUITY → SPELL", COLORS["violet"]),
-        ("machine", "MACHINE", "REPEATABILITY → EXACT", COLORS["cyan"]),
-        ("creature", "CREATURE", "HUNGER → SIGNAL", COLORS["green"]),
-        ("turd", "TALKING TURD", "RIDICULE → PUNCHLINE", "#ffd18a"),
-    ]
-    boxes = [(35, 190, 350, 610), (370, 190, 685, 610), (35, 650, 350, 1070), (370, 650, 685, 1070)]
-    for i, ((key, name, pressure, color), box) in enumerate(zip(specs, boxes)):
-        reveal = ease((local-i*.55)/1.3)
-        cx = (box[0]+box[2])//2
-        cy = (box[1]+box[3])//2
-        bw, bh = box[2]-box[0], box[3]-box[1]
-        sw, sh = max(8, int(bw*reveal)), max(8, int(bh*reveal))
-        live = (cx-sw//2, cy-sh//2, cx+sw//2, cy+sh//2)
-        rounded_image(im, ASSETS[key], live, 20, 1.0+.025*math.sin(local*1.7+i), .45, color)
-        d = ImageDraw.Draw(im)
-        if reveal > .78:
-            alpha_box = (box[0]+8, box[3]-94, box[2]-8, box[3]-10)
-            d.rounded_rectangle(alpha_box, 12, fill="#070b14e8")
-            text(d, (box[0]+20, box[3]-82), name, 20 if name != "TALKING TURD" else 17, color, True)
-            text(d, (box[0]+20, box[3]-48), pressure, 16, COLORS["white"], True)
-    centered(d, "PLANNED CARDS  •  EPISODE 01", 1120, 18, COLORS["gold"], True)
-
-
-def scene_gate(im: Image.Image, t: float):
-    d = ImageDraw.Draw(im)
-    local = t-25
-    for i, color in enumerate((COLORS["violet"], COLORS["cyan"], COLORS["green"], "#ffd18a")):
-        a = local*.8+i*math.pi/2
-        x, y = W/2+math.cos(a)*245, 390+math.sin(a)*130
-        r = 50+8*math.sin(local*3+i)
-        d.ellipse((x-r, y-r, x+r, y+r), fill=color, outline=COLORS["white"], width=3)
-    centered(d, "THE AUDIENCE ALTERS CANON", 125, 25, COLORS["gold"], True)
-    centered(d, "WHICH TIMMY", 590, 57, COLORS["white"], True)
-    centered(d, "GETS TO REMEMBER", 675, 49, COLORS["white"], True)
-    centered(d, "THIS?", 765, 76, COLORS["gold"], True)
-    d.rounded_rectangle((55, 892, W-55, 1000), 18, outline=COLORS["cyan"], width=3)
-    centered(d, "WIZARD  /  MACHINE", 917, 24, COLORS["white"], True)
-    centered(d, "CREATURE  /  TALKING TURD", 959, 21, COLORS["white"], True)
-    centered(d, "NO POST UNTIL HUMAN APPROVAL", 1072, 19, COLORS["muted"], True)
-
-
-def frame(t: float) -> Image.Image:
-    im = base(t)
-    if t < 5: scene_board(im, t); active = 0
-    elif t < 10: scene_receipt(im, t); active = 1
-    elif t < 16: scene_sources(im, t); active = 1
-    elif t < 25: scene_mutations(im, t); active = 2
-    else: scene_gate(im, t); active = 3
-    chrome(ImageDraw.Draw(im), t, active)
-    # Soft entrance/exit only; scene motion remains continuous within each stage.
-    if t < .35:
-        im = ImageEnhance.Brightness(im).enhance(ease(t/.35))
-    if t > DURATION-.5:
-        im = ImageEnhance.Brightness(im).enhance(max(0, (DURATION-t)/.5))
-    return im
-
-
-def build_audio():
-    n = int(DURATION*RATE)
-    audio = np.zeros(n, dtype=np.float64)
-    rng = np.random.default_rng(1608072026)
-    for beat in np.arange(0, DURATION, .5):
-        i = int(beat*RATE); length = min(int(.16*RATE), n-i); x = np.arange(length)/RATE
-        audio[i:i+length] += .42*np.sin(2*np.pi*(74*x-28*x*x))*np.exp(-x*25)
-    for beat in np.arange(.5, DURATION, 1):
-        i = int(beat*RATE); length = min(int(.12*RATE), n-i); x = np.arange(length)/RATE
-        audio[i:i+length] += .12*rng.normal(0, 1, length)*np.exp(-x*34)
-    for beat in np.arange(.25, DURATION, .5):
-        i = int(beat*RATE); length = min(int(.035*RATE), n-i); x = np.arange(length)/RATE
-        audio[i:i+length] += .04*rng.normal(0, 1, length)*np.exp(-x*80)
-    stages = [(0, 5, (110, 165, 220)), (5, 10, (123, 185, 247)), (10, 16, (98, 147, 196)), (16, 25, (131, 196, 262)), (25, 30, (110, 220, 330))]
-    for start, end, chord in stages:
-        i0, i1 = int(start*RATE), int(end*RATE); x = np.arange(i1-i0)/RATE
-        env = np.minimum(1, x*2)*np.minimum(1, (end-start-x)*2)
-        for f in chord: audio[i0:i1] += .035*np.sin(2*np.pi*f*x)*env
-    audio = np.tanh(audio*1.4)
-    pcm = (audio*32767).astype("<i2")
-    with wave.open(str(TMP_AUDIO), "wb") as w:
-        w.setnchannels(1); w.setsampwidth(2); w.setframerate(RATE); w.writeframes(pcm.tobytes())
-
-
-def main():
-    cmd = ["ffmpeg", "-y", "-loglevel", "error", "-f", "rawvideo", "-pix_fmt", "rgb24", "-s", f"{W}x{H}", "-r", str(FPS), "-i", "-", "-an", "-c:v", "libx264", "-preset", "medium", "-crf", "20", "-pix_fmt", "yuv420p", "-movflags", "+faststart", str(TMP_VIDEO)]
-    proc = subprocess.Popen(cmd, stdin=subprocess.PIPE)
-    assert proc.stdin
-    try:
-        for i in range(int(DURATION*FPS)): proc.stdin.write(frame(i/FPS).tobytes())
-    finally: proc.stdin.close()
-    if proc.wait(): raise SystemExit("video encode failed")
-    build_audio()
-    subprocess.run(["ffmpeg", "-y", "-loglevel", "error", "-i", str(TMP_VIDEO), "-i", str(TMP_AUDIO), "-c:v", "copy", "-c:a", "aac", "-b:a", "160k", "-shortest", "-movflags", "+faststart", str(OUT)], check=True)
-    TMP_VIDEO.unlink(); TMP_AUDIO.unlink()
-    for sec, name in [(2, "01-board"), (7, "02-vincent-receipt"), (13, "03-source-artifacts"), (21, "04-mutations"), (27, "05-audience-gate")]:
-        frame(sec).save(ROOT / "stills" / f"{name}.jpg", quality=88)
-    print(OUT)
-
-
-if __name__ == "__main__": main()