| OsProject | |
|---|---|
| id | reel-driven-development |
| state | active |
| owner | WolfgangFahl |
| title | Reel Driven Development |
| url | https://github.com/WolfgangFahl/reel-driven-development |
| version | 0.0.2 |
| description | Reel Driven Development - turn recorded user walks into domain stories and outcome objects |
| date | 2026-08-14 |
| since | 2026-07-30 |
| until | |
pip install reel-driven-development
# alternatively if your pip is not a python3 pip
pip3 install reel-driven-development
# local install from source directory of reel-driven-development
pip install .
pip install reel-driven-development -U
# alternatively if your pip is not a python3 pip
pip3 install reel-driven-development -U
Nobody reads requirements documents. But everybody can press record.
For decades we asked users to translate their frustration into tickets, specs and user stories — and lost most of the truth in translation. Meanwhile the most honest requirements document already exists: a user walking through your software, thinking aloud.
"I like this ... the feature that is missing is ... unfortunately here I can not ..."
That soundtrack is pure gold. Until now it evaporated the moment the meeting ended.
Reel Driven Development (RDD) turns that recording — the reel — into the driver of software development:
This is Domain Driven Design and Specification by Example — grounded in real behavior instead of workshop guesswork.
On 2026-07-18 we recorded the first public RDD reel for the nicescholia migration: a 3:54 min walk through the author aspect of Tim Berners-Lee (Q80) — legacy Scholia, snapquery named parameterized queries, two live failures, and the new nicescholia dashboards.
The mishap that proves the method: the microphone was forgotten — the reel has no sound. It worked anyway. Reverse RDD: the narration was inferred by AI from a dense frame sweep (1 frame / 10 s, every frame analyzed) and is marked as hypothesis throughout; the 17-hop graph walk itself is fully frame-grounded evidence. The silent reel still yielded 4 findings routed to 3 issue trackers.
Full reel with all 17 hops, video, inferred transcript and findings: nicescholia discussion #12
|
|
| Hop 3 — the walk starts: Q80 author profile on legacy Scholia | Hop 11 — finding F1: "Unrecognized error format" — the error itself is unusable |
|
|
| Hop 14 — finding F2: wdt prefix not registered although the prefix merger is active | Hop 15 — nicescholia Endpoint Monitor: where the walk should eventually run natively |
Findings F3 (empty Use aspect) and F4 (legacy under load / 503) went to WDscholia/scholia; F1 and F2 to snapquery — every frustration became a tracked, actionable item. That is the loop: walk, record, semantify, fix.
Next reel: with audio — what the user actually said, KANO-linked, next to what the AI guessed they would say.
Every reel becomes a queryable page in the knowledge graph — name, date, participants, transcript, state. The growing List of Recordings is the public window into this: first proof-of-concept recordings are up, and more are published as they are cleared.
And you can walk one yourself: the GenWiki demo reel on rdd.bitplan.com — no account, no password, your verdicts stay on your device.
State-of-the-art Agentic AI makes audio transcription and screenshot analysis of videos cheap and reliable. The narrative of a video arrives in a format software development can use directly. The full loop — from a user's sigh to shipped software — is closing fast.
We believe Reel Driven Development will catch on very quickly.
RDD is one piece of our Semantification approach: applying Agentic AI to turn use cases into knowledge graphs — for the systematic reduction of user frustration. Software that fits its use cases is notoriously hard to create. Letting people simply show their experience changes the economics of getting it right.
The science behind it: Reel Driven Development - RDD on our research wiki.
What was the last thing a user showed you that no ticket ever captured?