OPEN SOURCE · AGENT READY
Sherlock Reasoning Layer (SRL)
Solve what the internet can’t agree on.
- Moon landing faked?
- Market manipulated?
- Hidden agenda—or just noise?
Drop the claim. SRL investigates, attacks every angle, and returns what actually holds up—with sources.
Not opinions. Not narratives.
Just what survives.
Open source · agent-ready · built for real answers · Real installs: python -m srl.cli.run · see README
Every claim becomes a case.
Most people don’t want the truth—they want a story that feels right. SRL doesn’t care what feels right. It cares what holds up under pressure.
Search gives you more noise. SRL gives you a conclusion. It doesn’t just find information. It builds competing explanations, tries to break them, and keeps going until only one stands—or tells you none do.
Five moves, in order—no shortcuts, no assumptions
1 · GATHER
2 · ATTACK
3 · PROVE IT
This is what thinking like a detective actually looks like—six roles on three beats: gather, attack, prove.
The Case Isn't Closed.
It's Computed.
Every claim is tested against alternatives. Every conclusion is backed by sources. Every gap is called out, not ignored.
- If the story falls apart, you’ll see where.
- If it holds up, you’ll see why.
- It doesn’t tell you what people think—it tells you what survives.
Ask anything. Get a case file.
Your agent runs the full investigation, then returns report.md (clear answer), report.json (structured data), and worldstate.json (full reasoning trace). Message from Telegram or any hook you wire—it drops the files in thread so you can open, diff, or forward without digging through logs.
You don’t scroll. You conclude.
Some questions don’t have answers.
Most just haven’t been tested properly.
SRL exists for both—messy claims your agents can actually run through a real loop.
- “Was it faked?” → tested
- “Was it coordinated?” → tested
- “Is there real evidence?” → tested
Not dismissed. Not assumed. Tested.
Under the hood the trail never goes cold: 404s and paywalls become evidence, archives plug in when you allow it, and each run can inherit the last—no re-starting from zero.
Show the work. Or it didn’t happen.
Why SRL is different: most tools repeat what’s popular, hide uncertainty, and stop at surface-level answers. SRL tests competing realities, measures confidence, and shows its work.
Every step leaves a trace—sources, attacks, dead ends, final verdict. No black box. No blind trust.
Claim
Sources
Attacks
Verdict
It doesn’t tell you what people think. It tells you what survives when you turn the pressure up.
Detective Board — live cases on the site
The board runs a beefier build of SRL than the default open-source stack—wider budgets, longer loops, and publish-ready dossiers the team curates here. Follow us on X (link on the board) to pitch what we should stress-test next; we pull suggestions from the community when we schedule the queue.
All activity updates on this site—timers, live runs, and queued slots refresh as the mesh moves. Every six hours we queue a new case so there is always forward motion on what is live and what is next.