Systems atlas

Architecture is wherethe product promisebecomes enforceable.

A public map of how the showcase systems handle constraints, tradeoffs, validation, review states, and proof artifacts.

Trace explorer

Architecture shown asan inspectable path.

Choose a system and follow how input becomes a decision, how weak input is guarded, and where the public proof lives.

SquadBrain trace

Competitive practice flow with validation boundary.

Roster input becomes normalized practice state, adaptive priority, match result, recomputed score, and reviewable rank movement.

01InputRoster + session state

02RulesPriority + match constraints

03GuardrailRecompute results

04OutputProgress + proof

Risk handled: untrusted competitive submissions.Proof: result-validation.ts + ADR-01

Open proof file Open decision records

Architecture lenses

Same projects,different engineering questions.

Decision records

Tradeoffs made visible instead of hidden inside the build.

ADR-01

Keep competitive scoring recomputable

SquadBrain does not treat client-reported score as the source of truth for competitive movement.

Tradeoff: More validation logic, less trust in the UI layer.
Benefit: Rank movement can be defended and suspicious results can be held for review.

Inspect example

ADR-02

Practice priority beats random recall

The practice queue uses due state, misses, stability, exposure, and prompt mix.

Tradeoff: More state modeling, less simple randomization.
Benefit: The product can explain why a card appears next.

Inspect example

ADR-03

Evidence before packet language

Lasting Ground output language is downstream of source-lane validation and evidence scoring.

Tradeoff: More cautious output, fewer unsupported conclusions.
Benefit: The artifact stays reviewable and source-bounded.

Inspect example

ADR-04

Uncertainty is a product state

Missing or weak source lanes are shown as caution/missing states, not hidden.

Tradeoff: The output is less magically complete.
Benefit: Reviewers can see source coverage honestly.

Inspect example

Distinctive signals

Specific capability signals, not generic portfolio claims.

Validation boundaryDoes not trust the prettiest UI state

Competitive results are recomputed before rank movement.

Explainable operationsEvery priority has a reason

Practice queues and source lanes expose the factors behind decisions.

Honest outputUncertainty is shown directly

Missing or weak evidence becomes a visible product state.

Reviewable deliveryPublic proof paths are first-class

Demos, tests, CI, coverage, ADRs, and sample artifacts are connected.

Risk / guardrail map

Advanced work is often about knowing what not to automate blindly.

RiskGuardrailPublic proof

Untrusted competitive submissionsServer-style recomputation and risk flagsresult-validation.ts

Practice loop becomes random triviaPriority scoring and explainable reasonsadaptive-practice.ts

Source context becomes overconfident languageEvidence score and support-depth gateevidence_scoring.py

Incomplete lane looks completeVisible caution and missing statespacket_composer.py