These drafts introduce the project without claiming a new verification primitive or presenting the default privacy policy as compliance advice. Replace bracketed details before sending.
Subject: A five-minute test for verifiable grant-report numbers
I built an open-source workflow for drafting funder outcome reports where every number is computed by deterministic SQL and carries a receipt. A fail-closed gate blocks any displayed number that cannot be traced back to one of those receipts.
The synthetic housing demo takes about five minutes and runs offline. I would value your reaction to two questions: could a grant or program staff member follow the trace, and where would mapping your actual metric definitions become difficult?
Demo: https://github.com/ChelseaKR/outcome-receipts/blob/main/docs/TRY_THE_DEMO.md
Please do not send real service rows or client identifiers. An anonymized column inventory and a written metric definition are enough for the mapping discussion.
Title: Testing an offline trust chain for grant-report numbers
I have been working on an open-source reporting workflow for a specific failure mode: a correct outcome figure gets copied through several tools, or a drafting model introduces a plausible number that has no source.
The workflow computes every figure with deterministic SQL, attaches a receipt, applies small-cell controls, and blocks export unless every displayed number binds to a publishable receipt. A named human still approves the result. The language-model drafting seam is optional and off by default.
There is a five-minute synthetic housing demo here: https://github.com/ChelseaKR/outcome-receipts/blob/main/docs/TRY_THE_DEMO.md
I am looking for practitioner feedback, especially on the difficult part: mapping a funder's metric definition onto one organization's schema. What would you need to see before trusting the trace, and which definitions are hardest to translate into a reproducible query?
Grant-report numbers should not become less trustworthy when prose gets easier to draft.
I built outcome-receipts, an open-source, offline-first workflow that computes
each figure with deterministic SQL and attaches a receipt. A fail-closed gate
blocks export if any displayed number cannot be traced to a publishable receipt.
The optional model writes prose only; it never supplies figures.
The five-minute synthetic housing demo shows the report, receipt manifest, privacy boundary, human approval, and re-derivation check: https://github.com/ChelseaKR/outcome-receipts/blob/main/docs/TRY_THE_DEMO.md
I would especially like feedback from nonprofit grant, program, evaluation, and data staff who translate funder definitions into reports.
Title: Every number is a receipt: a fail-closed trust chain for outcome reports
Outcome reporting often depends on metric definitions that must be translated into one provider's HMIS or service-data schema. This session demonstrates an open-source, offline-first workflow that maps explicit requirements to reviewed SQL, computes aggregate figures with receipts, suppresses small cells, and refuses export when a displayed number lacks evidence. Participants will inspect a synthetic housing example from source schema through a funder-readable trace and discuss where metric-mapping uncertainty must route back to a human. The session does not present the tool as a reporting standard or compliance system; it focuses on reproducibility, privacy boundaries, and reviewable decisions.
Thank you for trying the demo. I am tracking whether people can complete three steps without assistance: produce the report, locate the definition behind a figure, and run verification. If one of those steps was unclear, please tell me the command or page where you stopped. If you reached the mapping question, an anonymized column inventory and a plain-language metric definition are the most useful next inputs.