Skip to content

docs: add external-research validation to PRIOR_ART.md #4

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
wikithoughts merged 1 commit into from
Aug 30, 2026
Merged

docs: add external-research validation to PRIOR_ART.md #4

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
22 changes: 22 additions & 0 deletions docs/PRIOR_ART.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -26,4 +26,26 @@ we own end to end. (See `docs/DECISIONS.md` D18 for the reuse policy.)
fully-local research with adaptive engine selection and SearXNG; their private-documents RAG is on
our roadmap.

## What the wider ecosystem confirms

- **The "floor, not verified-true" framing is externally validated.** ["Cited but Not
Verified"](https://arxiv.org/abs/2605.06635) (arXiv:2605.06635, May 2026) benchmarked frontier
deep-research agents and found only 39–77% factual/citation accuracy — and accuracy *drops* as
tool-call count scales up. `council/fidelity.py` (D27) already refuses to overclaim: it's an
explicit lexical **grounding floor** (content-token overlap plus fabricated-number detection), and
a GROUNDED verdict is documented as "not obviously fabricated," never "verified true." This paper
is independent confirmation that the framing is the honest stance, not an undersell — if
well-resourced frontier systems can't clear the bar of "verified," a tool that states plainly it
measures a floor is more trustworthy than one that implies otherwise.
- **No comparable non-token peer network exists.** Among local/decentralized-compute-sharing
projects we surveyed — [GaiaNet](https://github.com/GaiaNet-AI) (added staking/validator
incentives in 2026), [Petals](https://github.com/bigscience-workshop/petals) (BitTorrent-style LLM
sharding with no incentive layer at all, flagged in research literature as vulnerable to
free-riding without one), and SwarmHarness ([arXiv:2605.28764](https://arxiv.org/abs/2605.28764), a
2026 academic proposal for token/incentive-aligned agent networks) — none implement a
non-transferable, no-token, earned-only credit model. D1 in `docs/DECISIONS.md` commits us to
exactly that: internal, non-transferable credit, no tradeable token, ever. Worth stating explicitly
rather than leaving implicit: as far as this survey found, that combination is a real positioning
differentiator, not just a self-imposed constraint.

Thank you to these authors and communities.
Loading