SolutionsScience
Science. From literature to hypothesis to experiment.
Literature graphs, experiment pipelines, and hypothesis tracking, run as enforced processes with signed, replayable memory.
Runs as a service today.
01Capabilities
The whole inquiry, on one graph.
A research group works across sources, hypotheses, and experiments. Science keeps them on one graph, with provenance on every finding and disagreements kept in view.
- Literature into a typed graph
- Papers, claims, and citations become a typed property graph. What a source says, and what cites it, are edges you can traverse.
- Hypothesis tracking with provenance
- Each hypothesis carries its state and its evidence. Every finding traces back to the run, the source, and the reasoning that produced it.
- Experiment pipelines with gates
- Experiments run as enforced processes. A stage advances only when its criteria are met, and the run stops for a human where it should.
- Contradictions surfaced
- When sources disagree, the disagreement surfaces as a typed conflict cell and stays in the record until new evidence settles it. No fact wins by being written last.
- Replay and asOf
- Any conclusion replays from the journal, and asOf answers what was believed when. A finding can be read in the light of what was known at the time.
02Walkthrough
One contradiction, surfaced and resolved.
Two sources disagree. Most memory systems keep whichever was written last. Here is what happens instead.
fact.recorded
Source A reports that compound X inhibits pathway Y. Signed, timestamped, appended.
fact.recorded
Source B reports no measurable effect of compound X on pathway Y. Also signed and appended.
conflict.surfaced
The projection types the disagreement as a conflict cell on the X-to-Y claim. Neither fact wins by being written last.
pipeline.dispatched
An experiment pipeline targets the disagreement, with its own gates and journal.
fact.recorded
The new result lands above the old facts, not over them. The conflict resolves forward.
memory.asof
A query as of last month still shows the conflict as it stood. The record admits what was unknown, and when.
A conclusion that has lost its provenance has to be re-established from scratch. A signed graph keeps it standing.
03Calibration
Calibrated to research fidelity.
The harness (agent, skills, tools, process, memory) is tuned against evals on research-task fidelity and process adherence. The claim is fidelity to the process you defined; the console never presents a discovery.
- Evals on research fidelity
- Evals are defined on research-task fidelity and on adherence to the process a task requires. Both are scored.
- Converged per task
- Process, skills, and memory are tuned until runs hit the target score for the research task at hand, and hold it when the evals re-run.
- No discovery claims
- Calibration aligns a run to the research process you defined. It does not claim a discovery, and the console never presents one.
04The console
What the console shows.
- A literature and citation graph.
- A hypothesis tracker with states: open, supported, contradicted.
- An experiment-pipeline run.
- An evidence view, with a contradiction surfaced as a typed conflict.
The graph, the gates, and the journal are open source. Calibration to your field, and day-to-day operation with your group, are what we bring.
What the group believed, when, and why: one graph keeps it, disagreements included.
Bring a small literature set; watch the first conflict surface.
A Science demo ingests a small literature set you bring and surfaces its first typed conflict.
Or emailhello@a5c.ai
The intro call is 30 minutes and free.