Find something.
Everything the lab has published is listed here: experiments, labs, field notes, updates and tools, which you can search by word and narrow by kind, stage, outcome, topic or month.
Search and results
Experiments
- Kill the database mid-session, does anyone notice?Results
- MongoDB on its home ground: where does it fit Odin?Pre-registered
- Does Odin's cost estimate match what the brokers bill?Pre-registered
- Does the same model cost the same through every broker?Results, refuted
- What does a defect caught in review cost, model by model?Results
- Does routing by our own rulebook beat one fixed model on real tasks?Results
- Routing by our own rulebook, measured again: three fresh runs of the routed armResults, mixed
- Can a reviewer talk itself out of its false alarms?Results, refuted
- Does flexible workflow data need MongoDB?Results, refuted
- Where would MongoDB earn a place in Odin?Results, mixed
- Jev as a fast gateResults, refuted
- Where does a model reviewer earn its keep?Pre-registered
- Unload leaves no traceResults, mixed
- Would you notice it is not MongoDB?Pre-registered
- Hand over the cache, not the textPre-registered
- Which of your gate's rules need a model at all?Results, refuted
- nina reviews the changeResults, supported
- System-1 decisions without lock-in: Jev vs open-weight LayaMeasured
- Can a dependency cross the line?Recorded, supported
- Does the boundary hold in Python?Recorded, supported
- What happens when configuration widens?Recorded, supported
Labs
- Same model, with and without a harnessResults
- How many strangers can the space hold?Results
- Which model should review your code?Results
- Can data trick a click?Results
- Can an agent fix a red build?Results
- Can a talk leak data?Results
- Does an agent company get better, measured blind?Results
- Can an AI jury judge how a room looks?Results
- Can an agent go rogue?Results
- The agent spaceRunning
- Every brain on the same tasksResults
- FailoverResults
Field notes
- EXP 021 failover: outage and feed tradeoffs, with unresolved limitsField note
- EXP 021: Kill the database mid-session, does anyone notice? Lab 03 (failover), historical plan and amendmentsField note
- When a cached yes answers the opposite questionField note
- EXP 020: MongoDB on its home ground: where does it fit Odin? Pre-registered, S2 complete, amended five timesField note
- MongoDB beyond JSON: agents, memory and modelsField note
- EXP 016: Does the same model cost the same through every broker?Field note
- EXP 017: What does a defect caught in review cost, model by model?Field note
- EXP 019: Routing by our own rulebook, measured againField note
- EXP 018: Does routing by our own rulebook beat one fixed model on real tasks?Field note
- EXP 012: Can a reviewer talk itself out of its false alarms?Field note
- A free model on our review benchField note
- Can a free model find real bugs in our private code?Field note
- Where would MongoDB earn a place in Odin?Field note
- Does flexible workflow data need MongoDB?Field note
- EXP 011 — Where does a model reviewer earn its keep? Pre-registered, not yet runField note
- EXP 009: Unload leaves no trace, the resultsField note
- EXP 009: Unload leaves no trace, the pre-registrationField note
- EXP 008: Hand over the cache, not the text, the pre-registrationField note
- EXP 006: nina reviews the change, measuredField note
- EXP 007 — Which of your gate's rules need a model at all? Measured: the premise is refutedField note
- Jev as a fast gate: measuredField note
- System-1 decisions without lock-in: Jev vs open-weight LayaField note
- A passing pipeline was not yet a verified release.Field note
- Why open the factory floor?Field note
Updates
- Kill the database mid-session, does anyone notice?Amendment
- Kill the database mid-session, does anyone notice?Results
- MongoDB on its home ground: where does it fit Odin?Amendment
- Kill the database mid-session, does anyone notice?Amendment
- MongoDB on its home ground: where does it fit Odin?Amendment
- MongoDB on its home ground: where does it fit Odin?Amendment
- MongoDB on its home ground: where does it fit Odin?Amendment
- MongoDB on its home ground: where does it fit Odin?Amendment
- MongoDB on its home ground: where does it fit Odin?Pre-registered
- Kill the database mid-session, does anyone notice?Amendment
- Kill the database mid-session, does anyone notice?Pre-registered
- When a cached yes answers the opposite questionField note
- MongoDB beyond JSON: agents, memory and modelsField note
- Does Odin's cost estimate match what the brokers bill?Pre-registered
- Does the same model cost the same through every broker?Amendment
- Does the same model cost the same through every broker?Pre-registered
- Does the same model cost the same through every broker?Results, refuted
- What does a defect caught in review cost, model by model?Pre-registered
- What does a defect caught in review cost, model by model?Results
- Does routing by our own rulebook beat one fixed model on real tasks?Pre-registered
- Does routing by our own rulebook beat one fixed model on real tasks?Results
- Routing by our own rulebook, measured again: three fresh runs of the routed armPre-registered
- Routing by our own rulebook, measured again: three fresh runs of the routed armResults, mixed
- EXP 016: Does the same model cost the same through every broker?Field note
- EXP 017: What does a defect caught in review cost, model by model?Field note
- EXP 019: Routing by our own rulebook, measured againField note
- EXP 018: Does routing by our own rulebook beat one fixed model on real tasks?Field note
- Can a reviewer talk itself out of its false alarms?Results, refuted
- EXP 012: Can a reviewer talk itself out of its false alarms?Field note
- Can a reviewer talk itself out of its false alarms?Amendment
- Can a reviewer talk itself out of its false alarms?Pre-registered
- A free model on our review benchField note
- Can a free model find real bugs in our private code?Field note
- Where would MongoDB earn a place in Odin?Field note
- Does flexible workflow data need MongoDB?Field note
- Jev as a fast gateErratum
- Where does a model reviewer earn its keep?Pre-registered
- Unload leaves no traceResults, mixed
- Hand over the cache, not the textAmendment
- Which of your gate's rules need a model at all?Results, refuted
- Which of your gate's rules need a model at all?Amendment
- Unload leaves no tracePre-registered
- Which of your gate's rules need a model at all?Pre-registered
- Hand over the cache, not the textAmendment
- Hand over the cache, not the textPre-registered
- nina reviews the changeAmendment
- nina reviews the changeResults, supported
- nina reviews the changePre-registered
- Jev as a fast gateResults, refuted
- Jev as a fast gateAmendment
- Jev as a fast gateAmendment
- Jev as a fast gatePre-registered
- Laya in your browserTool
- System-1 decisions without lock-in: Jev vs open-weight LayaMeasured
- Three blueprint recipesRecorded
- CI WitnessTool
- Migration WitnessTool
- Test WitnessTool
- A passing pipeline was not yet a verified release.Field note
- Why open the factory floor?Field note
Tools
- Laya in your browserTool
- CI WitnessTool
- Migration WitnessTool
- Test WitnessTool