Sport and racing
Today's board
not read
The sport pack answered 404.
Autonomous decision system · Sylphx
Evidence in.
Mandate-safe act or abstain out.
Kalkas turns sealed point-in-time evidence into one calibrated belief, checks it against an explicit mandate and risk policy, and returns a single reproducible receipt — an admitted act, or a reasoned abstention that still settles and learns.
live:falsesimulation-onlyno custodyabstain is a result
- 6
- stages, one receipt per run
- 3
- domain modules, one kernel
- 0
- live effects in the shipped adapter
sample data
Sport and racing
Ascot 15:40 — 8-runner handicap
decision clock 14 Mar 2026, 15:32 UTC
Full-field belief separated two runners above the confidence floor, stake stayed inside the ruin bound, and the simulated wager settled against the published result.
stage 1 of 6
Evidence
Every fact carries when it was observed, when it became available, and where it came from.
Field and draw
Declared racing corpus · sealed digest
observed 14 Mar 2026, 14:05 UTC · available 14 Mar 2026, 14:06 UTC
8 declared runners, 2 late non-runners excluded before inference.
Pre-off quotes
Quote snapshot · per-runner availability time
observed 14 Mar 2026, 15:20 UTC · available 14 Mar 2026, 15:20 UTC
Best reachable price per runner; stale quotes past the window refused.
Going and conditions
Course bulletin
observed 14 Mar 2026, 13:40 UTC · available 14 Mar 2026, 13:41 UTC
Going updated twice; only the reading available at decision time was used.
refusal boundary · Evidence published after the decision time is refused before inference.
Illustrative worked example. Field names match a real run receipt; the values are samples and no production run exists yet.
Today
Three packs running now, one record.
Read from each pack as this page loads: what the sport board decided and why, how far the valuation record has got, and whether any quant strategy passed its gate. No sample data in this section.
Financial analysis
Forward-score record
accumulating
An unpublished number of valuations issued; no horizon has closed yet.
Quant research
Promotion gate
not read
The quant pack did not answer in time.
Why this exists
Decision systems fail in three predictable places.
Each one is a missing binding, not a missing model. Kalkas exists to bind all three: evidence to time, belief to authority, and action to an outcome.
01
Analysis without authority
A probability is not permission. Dashboards hand you a belief and leave the mandate, bounds, and expiry to a human under time pressure — so the recorded decision is the one nobody can reproduce.
02
Stories told in sample
In-sample fit is cheap to find and easy to sell. Without a held-out window, a coverage floor, and a deflation step, promotion is a preference rather than a measurement.
03
Abstention nobody records
Systems that only log action hide the moment they should have declined. If not acting is invisible, you cannot tell discipline from absence.
The loop is the product
Six stages. One typed outcome. Nothing skipped.
Every run walks the same path, and every stage has a named way to refuse. A stage that cannot refuse is a stage that cannot be trusted.
- Sealed, point-in-time observations
Evidence
Every fact carries when it was observed, when it became available, and where it came from.
refuses: Evidence published after the decision time is refused before inference.
- A distribution, not a vibe
Calibrated belief
Versioned models turn evidence into one belief with explicit uncertainty and lineage.
refuses: A belief without usable uncertainty, or with mismatched lineage on replay, is refused.
- Explicit authority, explicit bounds
Mandate and risk
Nothing acts by default. A mandate states the objective, the bounds, and the expiry.
refuses: No mandate, an expired mandate, or a violated bound means no action is requested.
- One replayable decision receipt
Act or abstain
Each run ends in exactly one typed outcome: an admitted act, or a reasoned abstention.
refuses: Rejected candidates and final abstentions never invoke an effect adapter.
- Idempotent, receipt-bound, simulated today
Admitted effect
Only a replay-verified admitted act reaches an adapter, under the requested idempotency key.
refuses: A conflicting idempotency key fails closed and the original receipt stands.
- Outcomes close the loop
Settlement and learning
Acted and abstained runs settle against the real outcome and emit evaluation and learning receipts.
refuses: An unbound or replayed settlement fails closed instead of crediting a run twice.
Domain modules
Three domains today, one kernel underneath.
A domain module supplies typed evidence, belief, gates, and settlement. The kernel owns the lifecycle, so a new domain adds meaning — not a second decision engine.
Full-field, pre-off, source-bound
Sport and racing
Ranked beliefs across a whole field, priced against the quotes that were reachable before the off.
- · Stake bounds and maximum-loss policy per run
- · Ruin policy evaluated before any stake is requested
Out-of-sample, cost-aware, merge-gated
Quant research
Research decisions where promotion depends on held-out evidence, not on in-sample fit.
- · Out-of-sample coverage floor with abstentions counted
- · Deflated performance floor after selection cost
Knowledge-dated, restatement-aware
Financial analysis
Analytical stances that respect knowledge dates, restatements, and survivorship.
- · Mandate-safe stance under declared analysis bounds
- · Settlement and evaluation bound to the same run
Also carried by the same kernel: Threshold · Weather · Sports event · Analysis — forecast packs for graded and labelled questions where the outcome is a probability rather than a price. Domain reference
Operator console
The console shows receipts, not screenshots.
Verified outcomes, the run ledger, evidence admitted to each run, evaluation against the holdout, and the service posture — all read-only, all labelled with what is live and what is not.
Sport and racingAdmitted act
Ascot 15:40 — 8-runner handicap14 Mar 2026
Observed outcome: 1 winner; simulated stake settled at the quoted price.
Quant researchAdmitted act
Cross-sectional equity signal — 400 names, 30-day horizon09 Feb 2026
Horizon closed: realised return recorded against the simulated intent.
Financial analysisReasoned abstention
Restated margin trajectory — listed industrial22 Jan 2026
Outcome recorded against the abstention; no stance was published.
weekly verified outcomes
3
Sample ledger: 3 of 4 records end in a settled act or abstention. The console computes the same figure from the same records; neither is a production number.
belief against mandate floor
floor
Proof, not promises
A verified outcome is seven conditions on one run.
Not a screenshot, not a backtest slide. The count the product reports is the number of runs that satisfy every condition below.
- 01Evidence carries observation time, availability time, and provenance.
- 02Replay over the same inputs and model version is deterministic.
- 03The belief exposes uncertainty and source lineage.
- 04The mandate admits or refuses without an override.
- 05The run ends in a typed act or abstention receipt.
- 06The outcome resolves and binds to that exact run.
- 07Settlement emits evaluation and learning receipts.
LimitWhat fails closed
- · A fact published after the decision time never reaches a model.
- · In-sample fit alone never promotes a candidate.
- · A retry with a conflicting effect key is refused, not replayed.
- · A rejected candidate never plans and never reaches an adapter.
PostureWhat is live today
The sport, financial, and quant packs run hosted and publish their records read-only: no bets, no orders, paper trading only. Today's record shows them. The operator entry point and the console run locally against sealed corpora and checked-in fixtures. There is no external effect and no customer data. Trust and posture.
Lines we do not cross
What Kalkas will never do.
These are product constraints, not disclaimers bolted onto a release.
- No live decision, order, wager, or external effect exists in the shipped adapter.
- No custody of funds, accounts, or customer assets, and no counterparty reach-through.
- No guaranteed profit, no promised strike rate, and no performance claim sold as fact.
- No action when evidence, authority, replay, settlement, or effect identity is incomplete.
Commercial shape
Priced for decision intelligence and proof.
A subscription against the decision system and its evidence trail — never a share of winnings, a performance fee, or a cut of anyone's position.
Availability
Published plans are not open yet: there is no released hosted service to sell. The commercial model is documented now so the shape is not improvised later — no fee tied to outcomes, no custody, no performance guarantee.
Questions
Straight answers.
Is this a betting or trading system?
No. Kalkas is the decision layer: evidence, belief, mandate, act or abstain, settlement, and learning. The shipped effect adapter is simulation-only, there is no custody, and nothing here is investment or betting advice.
What makes a decision “verified”?
Seven conditions have to hold for the same run: sealed point-in-time evidence, deterministic replay, explicit uncertainty and lineage, an admitting mandate, a typed act or abstention, an outcome that resolves and binds to that run, and settlement that emits evaluation and learning receipts.
Why does abstaining count as a result?
Because acting without authority, confidence, or coverage is the defect. Abstention carries a reason, stays in coverage, and settles — so a system that abstains correctly is measurable rather than merely quiet.
How is this different from an analysis dashboard?
Analysis stops at a belief. Kalkas binds the belief to authority and risk, produces one reproducible outcome per run, closes the loop with settlement, and refuses anything it cannot evidence.
What can I run today?
The operator entry point and the read-only console run locally against sealed corpora and checked-in fixtures. The hosted service is not live; every surface says so.
Where does my data go?
Nowhere yet: the shipped paths are local and simulation-only. The hosted surface is not released, and the site itself collects nothing beyond what serving a page requires.
Bring a decision you already make.
Start with the local loop: sealed evidence, a candidate and baseline, a mandate, and a scenario that shows exactly how the run refuses. Then read the receipt it hands back.