Candidate construction guide · native Resolve not tested

Forecast driver branch · public R03 repair kit

This packages the saved R03 implementation of forecast-driver-branch. Seedling plan and Studio capacity are the two existing contents, not new examples. Code, SVGs, content, controls, timing and Resolve guidance are unchanged. This is an original offline specimen, not a product capture.

The actual R03 movies already exist. Their hashes and 20 project-file bindings per movie are in actual-artifacts.json. No movie is bundled or rendered by this kit. Both frozen current verdicts are bounded-none-supported within the inspected STATIC regions and criteria. That is not quality approval, equivalence, a winner, moving review or Resolve execution approval.

Start here

• SOURCE_OBSERVATIONS.md, REVIEW.md and CURRENT_GAPS.md are the approved public reconciliation texts, copied without modification.
• CURRENT_STATUS.md separates current evidence from historical build notes, including the earlier README-context exception.
• RESOLVE.md gives this example's editable Fusion node graph, coordinates, state keys and numeric construction checks. It is not an executed project.
• design/default.json and design/adaptation.json are the bounded editing interface. recipe.json, content/*.json and assets/*.svg remain editable.
• VERIFICATION.md explains artifact origins, clocks and verification limits.

Exact offline checks

Extract the downloaded ZIP into a new empty directory and open a shell there. Use an existing Python 3 interpreter, a POSIX shell and cmp. Set FORECAST_PYTHON externally if the interpreter is not named python3. The sibling forecast-check directory must not exist. Run this block literally:

set -eu
python="${FORECAST_PYTHON:-python3}"
work="$(cd .. && pwd)/forecast-check"
mkdir "$work"
mkdir "$work/scratch"
"$python" -B verify_public.py --root .
"$python" -B reconstruct.py --destination "$work/reconstruction"
env TMPDIR="$work/scratch" "$python" -B test_kit.py --project "$work/reconstruction/project"
"$python" -B test_guide.py
"$python" -B test_density.py
"$python" -B test_comparison.py
"$python" -B handoff.py verify --guide guide.json --project "$work/reconstruction/project"
"$python" -B handoff.py frames --guide guide.json --project "$work/reconstruction/project" > "$work/controls.csv"
cmp controls.csv "$work/controls.csv"
"$python" -B -c 'from pathlib import Path; import sys, resolve_controls; sys.stdout.buffer.write(resolve_controls.generated_files(Path("."))["resolve-controls.csv"])' > "$work/resolve-controls.csv"
cmp resolve-controls.csv "$work/resolve-controls.csv"
"$python" -B verify_public.py --root . --project "$work/reconstruction/project"

These commands do not install dependencies, download, launch a browser, render or execute Resolve. They reconstruct the 430-file saved Remotion core plus the three-file selected overlay, run the 27 inherited tests and compare both exported control files byte-for-byte, including their CRLF line endings. The read-only public verifier checks the exact payload file set and all 433 reconstructed files. It also checks each movie's 20 file bindings against those reconstructed bytes; it does not claim a new movie render or viewing.

The ZIP SHA-256 comes from the external delivery receipt, not a self-hash inside the ZIP. To verify that pin, also pass --zip and --expected-sha256 to verify_public.py. Internal manifests alone are not an external trust root.

Preserved R03 scope

The saved R03 correction added three signed working-minus-comparison totals and each comparison's last-period value to the existing legend. Those values come from the unchanged authored arrays; they are not copied financial data. Legend placement is authored latitude. The source places values at line ends; this kit makes no exact-pixel matching claim.

Events remain at local frames 1, 7, 10, 16, 25, 29 and 33. Frame 33 reuses the state from frame 25 and holds through frame 47. The output is 48 frames at 24000/1001 fps, 1920 × 1080, silent, with time base 1/24000. Local PTS is frame * 1001; the exclusive end is 48048 ticks. The 15-frame final numeric hold lasts 0.625625 seconds; it is not a pixel-freeze claim.

For deliberate edits, copy the kit, change its design files within the existing validate_design limits, and reconstruct into a different fresh destination with --design-root. Modified designs are not the sealed R03 defaults and cannot inherit the existing movies' bindings or review. The secondary setting has no supplied number. Do not invent one, add table-body data or imply working podcast, transcript, scrolling or second-input behavior.

R02_SCOPE.md, R03_SCOPE.md, source-excerpt.json and the preservation proofs retain their historical wording and bytes. Their then-pending render/review statements do not describe the current pairing. The original guide, repair, proofs and raw failure records remain untouched. No registered-example, export, source-quality, moving, Resolve or publication count is added.