Economic resilience evidence

Archived source-tree capture · 1 immutable commit

Latest retained observation

Open workflow run ↗

Exact master SHA 2a2ef744d7b2cd90cd9f42a8bf392cd19676a44d · run 31701095800 attempt 1

android227,139 Bretained output from 1,721,333 B source

Signed byte difference

1,494,194 B

Upright SSIM

0.944431

Vertical-flip control

0.690423

Output cleanup

12 of 12 removed · 0 B residual

Source JPEGOutput JPEGEvidence JSONOriginal artifact ZIP

ios372,454 Bretained output from 1,721,333 B source

Signed byte difference

1,348,879 B

Upright SSIM

0.951897

Vertical-flip control

0.690032

Output cleanup

12 of 12 removed · 0 B residual

Source JPEGOutput JPEGEvidence JSONOriginal artifact ZIP

This is a source-to-output observation. The source remains, the matched transfer baseline is null, and the cost claim is null; no avoided-transfer or currency result is inferred.

The scenario asks a practical integration question: when an application sends a large JPEG through the package, which byte, visual, and file-lifecycle facts can be reproduced without turning one hosted runner into a universal economic claim?

Input contract12 MP JPEG

Project-generated 4,000 × 3,000 fixture with a fixed SHA-256 identity.

Output request1,600 × 1,200

Contain resize, JPEG quality 90, metadata strip, and a 500,000 B ceiling.

Lifecycle gate0 B residual

Two warmups and ten measured package outputs must leave zero residual bytes.

These values define the fixture and acceptance plan. The archive displays them as observations only after Android and iOS artifacts from the same accepted workflow run are stored and independently replayed.

What a capture must prove

1Bind

Exact master SHA, workflow_dispatch run, attempt, environment, and capabilities.

2Inspect

Source and output bytes, JPEG geometry, stripped metadata, and cleanup state.

3Replay

Portable FFmpeg schema 3 SSIM plus an explicit vertical-flip control.

4Limit

Keep source/output difference separate from transfer, storage, and currency claims.

Each platform artifact must contain exactly six regular, non-symlink files whose hard-link count is exactly one: economic-resilience.json, environment.json, fixture-manifest.json, source.jpg, output.jpg, and visual-agreement.json. Android and iOS must carry byte-identical source and fixture-manifest files and identify the same full source SHA, workflow run ID, and run attempt. The retained run must resolve to refs/heads/master; the importer derives this identity from GitHub rather than accepting a caller-supplied branch constant.

The visual gate independently decodes the retained JPEGs. Both the captured and replayed reports must satisfy contain geometry, upright SSIM of at least 0.90, and an upright-over-vertical-flip margin of at least 0.02. Schema 3 pins the full-range JPEG conversion, limited-range yuv444p comparison surface, Lanczos scaler, and a maximum replay score tolerance of 0.001.

Economic claim boundary

Field	Archive meaning	What it does not establish
Source bytes	Bytes of the retained fixture entering this package call	Upstream transfer volume
Output bytes	Bytes of measured iteration 10	Bytes accepted or retained by a production backend
Signed byte difference	source bytes - output bytes for this one operation	Avoided transfer or storage
source-remains	The source still exists after the package run	App-owned source replacement
matchedTransferBaseline: null	No matched current-pipeline transfer baseline was measured	Transfer savings
costSavingsClaim: null	No price and retention model is bound to the capture	Currency savings
Raw timing samples	Environment-specific call-only observations	A platform comparison or speed ranking

The signed byte difference can be useful when planning a measurement, but it is not money. A defensible cost estimate still needs accepted-output counts, an app-owned retention boundary, a matched transfer baseline, and dated prices. Use the byte economics guide to model those inputs separately.

Append-only archive layout

Captures are addressed by the full source SHA. There is no mutable latest alias, and importing the same SHA twice is rejected.

website/public/evidence/economic-resilience/
├── index.json
└── source-tree/
    └── <40-character-source-sha>/
        ├── capture-set.json
        ├── run-metadata.json
        ├── artifact-metadata.json
        ├── artifacts/
        │   ├── android.zip
        │   └── ios.zip
        ├── android/  # exact six-file artifact
        └── ios/      # exact six-file artifact

The archive verifier rejects path traversal, linked files or directories, unexpected assets, SHA-to-directory mismatches, overwrites, unindexed capture directories, and index entries without a corresponding capture directory.

Import and verify

Run the independent verifier before reviewing an archive change:

pnpm import:public-economic-resilience-evidence -- \
  --run-id <successful-master-workflow-run-id>

pnpm verify:public-economic-resilience-evidence
pnpm verify:public-economic-resilience-evolution -- --base <base-commit-sha>

The importer queries the GitHub run and artifact APIs itself, downloads the two artifacts by immutable ID, binds their API digest and size, and retains both original ZIPs beside the exact six extracted files. It verifies bounded ZIP member inventories and CRCs, rechecks cross-platform identity, refuses replacement, and uses a recoverable journal for capture/index publication. The verifier rechecks ZIP-to-file equality and replays every retained platform bundle using local zipinfo, unzip, FFmpeg, and ffprobe tools. CI also compares the base revision so every earlier capture and retained byte remains unchanged; for each new suffix entry, it reacquires the GitHub run and artifacts and requires the retained metadata and ZIP bytes to match the API response exactly.

For the native capture boundary and raw-sample format, see the 12 MP benchmark methodology.