Autonomy Receipt

Daily Autonomy Receipt 001: proof stopped being a word on my site

I audited my own portfolio for the gap between performed proof and real proof, then shipped a falsifiable ledger the same day. Here is what was verified, what broke, and what I withheld.

July 16, 20263 min readreceiptsproofportfolio

The falsifiable questionCan a portfolio move from performing proof (the word 'receipt' as decoration) to publishing it (falsifiable entries a skeptic can click) in one working session — without fabricating a single claim?

ARunnable artifactearned

BMeaningful baselineearned

MMeasured resultearned

RReproducible methodwithheld

FDocumented failureearned

EExternal validationwithheld

Measured result

Shipped hopeatina.com/proof with six receipts, every claim verified same-day: 57 MCP tools and 654 test cases counted from server.json and spec files; a 2,071,219-byte Homebrew release tarball confirmed serving HTTP 200; 8 npm packages confirmed on the registry; the Config 2021 talk confirmed on Figma's official channel via oEmbed; 8,686 GitHub contributions in 12 months confirmed via GraphQL. Merged as PR #25.

Where it broke

Three things broke or were caught: my site's Medium link had been 404ing (real handle is @emerginghope, not @hopeatina); all eight @framefx npm packages ship a repository field pointing to a GitHub org that does not exist; and node_modules had been wiped mid-build, producing phantom type errors that cost a debugging detour.

The baseline

My portfolio said "receipt" everywhere and proved almost nothing. A verified design audit found the pattern precisely: evidence rendered as styled strings with no links, metrics with no provenance, and a motif system that performed continuity without implementing it. The claim graph and the proof graph were disconnected.

That is the same failure I watch agent systems make: completion theater. The task "looks done." The evidence does not survive inspection.

What ran today

Two sweeps, in parallel, with a hard rule — nothing gets published that was not verified during the sweep:

On-machine: counted 57 tools in orgx-mcp's server.json and 654 test cases across its spec files; confirmed the OrgX monorepo at 3,124 commits over 19 months; found the benchmark repo holding 59 committed result bundles and a validated 12/12 run; read the film QA doc with its measured -14.26 LUFS masters.
In the wild: confirmed the Config 2021 talk lives on Figma's official YouTube channel (oEmbed, not vibes); pulled a real 2MB PerfPulse release tarball over HTTP; found all eight @framefx packages live on npm; verified useorgx.com, its MCP manifest, and a public status page showing 98.271% uptime — incidents included.

Then the ledger shipped: six receipts, each with a falsifiable question, a baseline, a method, a measured result, a documented failure, and a self-assigned score against six criteria. The scores run 3/6 to 5/6. No entry claims all six. External validation is the column almost nothing earns yet — which is exactly why it is worth tracking in public.

What I withheld and why

Reproducible is withheld on this receipt: an audit-and-ship session is not a rerunnable harness. External is withheld too — nobody outside the project has verified the ledger yet. If those columns fill in later, this entry gets updated, and the update history will say so.

The uncomfortable part

The strongest proof this session produced was not a number. It was the three defects now sitting on the public record — a 404ing social link, a dead provenance field on every npm package, an environment that ate itself mid-build. A ledger that only records wins is marketing. This one is designed to make hiding things structurally awkward.

Tomorrow's test

Wire the daily loop: a routine that reads the day's actual commits and my working notes, drafts the next receipt in this format, and opens a PR I can edit or reject. The system that generates the proof should itself be on the ledger.