office-hours asks 'is demand real?' - an evidence layer that keeps the answer honest after the session

`/office-hours` asks "is demand real?" six great ways, then the answer becomes a paragraph in a design doc. Three weeks later gbrain will faithfully recall that paragraph - but recall is not recheck, and the founder's day-one belief about demand ages silently while everyone is busy shipping.

I built the missing half and I'd rather compose with gstack than duplicate any of it: an open-source evidence layer (MIT, stdlib Python + MCP) where office-hours' answers become falsifiable claims instead of prose:

- **"demand reality"** -> a customer-class claim on a typed evidence ladder (`buyer:signup 0.40 / reply 0.50 / call 0.65 / signature 0.85 / payment 0.95`) - the tool caps confidence at the evidence tier, so a hot waitlist can't masquerade as proven revenue, and the verdict line carries a derived `demand-UNVALIDATED` stamp that regenerates on every write until real buyer evidence exists.
- **"narrowest wedge, learn from real usage"** -> the wedge's smoke test (landing page + waitlist + ads, outreach, preorder) with its pass bar pre-registered into the thesis *before* the experiment runs.
- **design-doc assumptions** -> claims with pre-registered falsifiers, re-checked on later runs; `stale` nags when a shipped bet's kill-conditions have gone unexamined.

Receipts, since claims about rigor deserve them: we benchmarked it three times, pre-registered each round, and published all three headline nulls with raw outputs - what survived is durable structure (0 format failures vs 213 for prose), updates at 0.54x the cost of re-narration, and zero fabricated evidence in 24 adversarial opportunities. The bug log ships in the repo (7 defects, each mapped to the invariant it earned). 30-second demo: `python tools/ledger.py demo`.

Repo: https://github.com/umair-tareen/rnd-skill - handoff mapping written up in [docs/USING_WITH_GSTACK.md](https://github.com/umair-tareen/rnd-skill/blob/main/docs/USING_WITH_GSTACK.md).

Concrete proposal, smallest first:
1. A line in USING_GBRAIN_WITH_GSTACK.md (or wherever ecosystem pointers live) noting the office-hours -> evidence-ledger handoff for people who want their design doc's assumptions tracked;
2. If there's appetite: office-hours optionally emits its six answers in a claims-shaped block so any evidence tool (mine or otherwise) can ingest them without parsing prose.

If neither fits gstack's direction, no hard feelings - closing this with "not our lane" is a perfectly good outcome, and the compat doc on my side works regardless. Happy to PR either option.

Read docs/USING_WITH_GSTACK.md and inspect the repository's ecosystem-pointer documentation and /office-hours entry point. Decide whether the accepted scope is a pointer or claims-shaped output; done means the chosen integration path is documented or implemented with its expected handoff behavior clear and verified.

Written by the indexing model from the issue text.