Observed sign inversion vs expected monotonic rise. Kept as negative control.
MatSci starts with TST
Watch a system refuse fake certainty.
ParetoSol is a MatSci review surface that begins with transition state thinking, then refuses the pretty map when the committor, grain, or survey reliability does not support it. The default pattern is Saddlescope: every uncertain claim gets an owner, a boundary, and a receipt.
claim-first demo
Nylon bonds to ceramic
Status
UNKNOWN
- Searching observations
- Finding receipts
- Checking grain
- Looking for bridge evidence
default pattern
Saddlescope: every uncertain claim gets an owner, a boundary, and a receipt.
Most demos show what a model concluded. This pattern shows what the system refuses to conclude, where the claim sits on the landscape, and the exact work that would move it.
Claim under investigation
"Nylon bonds to ceramic."
UNKNOWN
no receipts examined yet
observations
receipts
declared grain
saddlescope - energy landscape
reject basin
accept basin
saddle unknown
Saddlepit
the panel nobody else ships
Why not?
A rejection is not a dead end. It is a work order. Perturb the case and watch the state move across the landscape: open, withheld, rejected, Saddlepit, or scout-admissible.
Stack enough contradicting observations at a fine grain and the claim falls into a Saddlepit: undecidable not from too little evidence, but too much that disagrees. Coarsening the grain past REV drains it.
Open gaps
estimated effort to close: 0h
Decision DNA
horseshoes margin
Proceed anyway, on the record.
Sometimes you erase a distinction you cannot reconstruct: pool the coupons, coarsen past REV, or ship on a sketch. That is allowed only when the margin spend is declared and logged.
Erasure margin burndown
greenwatchcorrectivedepleted
*GREEN - spend freely
You are inside reserved margin. Erasures here auto-log; no reason required.
Margin use log0% spent
- No erasures yet. The log is append-only.
the ontology
Four kinds of no answer.
Collapse these and you either hallucinate a verdict or hide a contradiction. Saddlescope returns the right hole by name.
VOID
Void
No value can exist. The question does not apply at this grain.
hole by subtractionGHOST
Ghost
Value unknown but shaped. It has a place waiting for a receipt.
epistemic gapOK
Verdict
Exactly one value clears the boundary. Decidable, owned, and receipted.
decidableBOTH
Saddlepit
Too many contradicting values. Maximally honest, minimally actionable.
hole by additionwhat do I do Monday morning?
Healthy rejections become evidence queues.
Every gap in a decision becomes work with an owner. Check the missing receipts off and the claim climbs the landscape in real time.
Missing second replicate. Blocks promotion above coupon grain.
Owner boundary unassigned. Blocks any allowable claim.
Missing evidence4 open
- Second microscopy receiptbench
- Assign owner boundarymaterials
- Repeat coupon - resolve disagreementbench
- Coordinate review - committor checkstatistics
honest rainbow saddle
You can like the rainbow. The chart still has to confess.
The saddle is a map, not the territory. ParetoSol colors by committor q, draws the energy saddle as a ghost, mutes unsurveyed regions with CATZOC-style confidence, and refuses the single-saddle picture when the spectral gap is not earned.
- q=0.5 ridge is the decision surface; energy is only a projection.
- Survey confidence suppresses color where observations are thin.
- Sampled trajectories sit on top of the surface, never underneath it.
- No spectral gap means no clean saddle; render the cluster instead.
client evidence packs
Claude's June 30 files are now part of the demo surface.
The client site carries the BioTwin and composites checklists as inspectable public artifacts. They are not hidden implementation notes: they define evidence types, posture caps, sketch limits, typed bridges, and the repeated rule that review surfaces do not authorize material or clinical action.
Accelerator checklist
Typed molecular, structural, kinetic, PK/PD, MSM, counterfactual, and patient-bridge evidence gates.
Open checklistSketching + evidence ontology
Sketch provenance, tail-aware error budgets, interphase evidence, A/B-basis gates, and aging axes.
Open checklistHorseshoe margin
Declared erasure-margin consumption: proceed only when the spend is bounded, logged, and not authority.
Inspect SaddleScopecTST
Transition-saddle review
See q, kappa, MEO, recrossing, and receipt states without turning scout bands into truth.
Launch the demoMCP
Public inspection endpoint
Use JSON-RPC tools and resources to inspect cTST posture, contracts, routes, and boundaries.
Open MCP manifestsigned in
Tons more to try
Open mobile review, materials refinement, polymer forensics, gradient discovery, music/TST samples, and scene-frame APIs.
Sign in if invitedPinto architecture
The demo is a receipt renderer, not a dashboard.
Claude's BioTwin accelerator note is now part of the posture: FORZA and SCD-II++ provenance stay re-evaluable, open predicates do not collapse into false claims, and contradictions are contained as Saddlepit/Belnap-style gluts until a reviewed microtheory resolves them.
Semiring provenance
Keep the derivation sketch; evaluate later as probability, reliability, NPV, or count.
BioTwin handoff
Change-of-support is explicit from molecular to cellular to patient-scale claims.
Local closure
LCWA names what is complete. Everything else remains open, not silently false.
Oracle drift
Minted history gets oracle stamps so later REV changes create named gluts, not stale certainty.
after sign in
The teaser opens into a playground.
The public site shows the discipline. A session opens the surfaces: review hub, phone lane, polymer forensics, gradient discovery, Masai crossing theatre, music/TST samples, and scene-frame JSON/SVG.
beta signup
Join the free beta
We are looking for people who will try the public cTST teaser, inspect the MCP surface, sign in to the larger review playground, and leave blunt feedback. The beta is free; the useful payment is telling us what failed.