1
Replay a policy change
Watch the same posts become a different feed.
Follow a post from candidate set to community ballot, approved policy, Bluesky feed, and inspectable Corgi receipt.
Product walkthrough · 4:14
See the complete governance loop.
A concise system overview followed by the reviewer-safe product demo, shown end to end.
01
Candidate posts
Corgi ingests public Bluesky posts and computes reusable post signals.
02
Ballot
Approved participants can vote on signals, topic priorities, and content rules.
03
Aggregation
Corgi aggregates complete ballots after the configurable voting window closes.
04
Review and approval
Results are reviewed before an operator approves or rejects the proposed policy.
05
Render in Bluesky
After approval and rescoring, Corgi serves the ordered feed for Bluesky clients to render.
06
Inspect in Corgi
Corgi shows the policy, score components, publication adjustments, and receipts.
DiscoverFollowingCorgi CommonsTools
Corgi rankEpoch 07
2
3
4
5
6
7
Corgi receipt
Why ranked #1
Every open-source project has one file everyone is scared to touch.
Recency
How recently the post appeared.
0.700
x 5%
= 0.035
Engagement
Replies, reposts, likes, and other public attention.
0.950
x 65%
= 0.617
Bridging
How well the post connects subgroups inside the community.
0.300
x 5%
= 0.015
Source diversity
Whether the feed is hearing from a wider set of sources.
0.350
x 5%
= 0.017
Relevance
How well the post matches the community's topic priorities.
0.360
x 20%
= 0.072
This post ranks here because the active policy is Conversation-heavy. Corgi stores the score components so people can inspect the mechanism on Corgi, while Bluesky shows the ordered posts.
Select a post to inspect its ranking. The rank numbers and receipt panel are Corgi-site annotations; standard Bluesky clients render the ordered posts, not Corgi score panels.
Counterfactual
Same posts. Different policy. Different feed.
Corgi separates post signals from community values. Changing the policy changes the order without pretending the posts themselves changed.
Conversation-heavy ranking
#1
Every open-source project has one file everyone is scared to touch.
score 0.757
#2
Published a dataset comparing alt-text quality across image models, plus the scoring rubric.
score 0.711
#3
The best research software is half methods section, half apology to future maintainers.
score 0.680
#4
Mapped accessibility gaps in open transit data, then shared the notebook and cleaned CSV.
score 0.569
Corgi Commons policy
#1
Published a dataset comparing alt-text quality across image models, plus the scoring rubric.
score 0.833
#2
Mapped accessibility gaps in open transit data, then shared the notebook and cleaned CSV.
score 0.799
#3
The best research software is half methods section, half apology to future maintainers.
score 0.692
#4
Field notes from a community mesh-network test, including the messy CSV and failed runs.
score 0.679
Calculation
Inspect the scoring math.
This illustrative replay keeps raw post scores fixed so you can isolate the policy change. Production rescoring recomputes candidate scores before publication.
Raw signal scores
These are the same post scores used by every policy in the walkthrough.
Swipe sideways to see all five signals.
| Post | Fresh | Likes | Bridge | Diverse | Match |
|---|---|---|---|---|---|
| Mapped accessibility gaps in open transit data, then shared the notebook and cleaned CSV. | 0.720 | 0.420 | 0.920 | 0.680 | 0.900 |
| A maintainer note on making privacy-preserving defaults easier to find in project docs. | 0.960 | 0.350 | 0.280 | 0.550 | 0.820 |
| Found a timezone bug in our benchmark harness. Patch and before-and-after runs are in the thread. | 0.640 | 0.580 | 0.220 | 0.420 | 0.480 |
| Every open-source project has one file everyone is scared to touch. | 0.700 | 0.950 | 0.300 | 0.350 | 0.360 |
| Published a dataset comparing alt-text quality across image models, plus the scoring rubric. | 0.680 | 0.620 | 0.880 | 0.840 | 0.940 |
| Field notes from a community mesh-network test, including the messy CSV and failed runs. | 0.460 | 0.280 | 0.740 | 0.720 | 0.860 |
| The best research software is half methods section, half apology to future maintainers. | 0.750 | 0.660 | 0.700 | 0.520 | 0.760 |
Formula
total = sum(raw signal score x active policy weight)
For the bridge-building policy, the weights sum to 100%: recency 15%, engagement 10%, bridging 35%, source diversity 15%, relevance 25%.
Topics shape relevance
Topic preferences are a separate policy map. They change the relevance signal only; they are not five more global ranking signals.
Rules shape eligibility
Adopted include rules act as an allowlist, exclude rules take precedence, and production adoption requires at least 30% support among ballots that submit content rules.
What an epoch means
An epoch is a stored feed policy. In production, a closed round is aggregated and reviewed before operator approval applies the complete policy and triggers a rescore.
Component scores explain the weighted sum. Publication-time adjustments, such as duplicate-link handling, can still affect final order and are called out in live receipts.
Where ranking lives
The feed changes in Bluesky. The receipt lives on Corgi.
This separation is intentional: people get a normal Bluesky feed, and the explanation layer stays inspectable on Corgi.
Bluesky
Shows the ordered posts.
Standard Bluesky clients render a normal custom feed in the order Corgi returns.
Corgi
Shows why that order happened.
Corgi hosts the score, receipt, community weights, and why-ranked explanation.
Future clients
Could bring more context inline.
A custom client could show explanations beside posts, but the current public path keeps them on Corgi.
Rank badges and receipt panels in this page are Corgi annotations. They explain the order, but they are not native Bluesky UI.
Any community
The mechanism is general. The policy is local.
- Neighborhood mutual aid can prioritize urgent, well-sourced local updates.
- Open-source maintainers can lift patches, docs, and release notes over generic project chatter.
- Tabletop creators can favor indie releases, safety tools, and useful play resources.
Inspect the live receipt path.
The walkthrough above is illustrative. The public demo starts from a frozen comparison corpus sourced from Corgi Commons and keeps every shadow vote isolated from production.
Demo posts on this page are illustrative. The reviewer demo identifies its frozen live-public snapshot and links back to source posts.