Skip to content

Current Capability Limits

Current capability is bounded by the weakest scientific, runtime, grounding, recommendation, or laboratory-consequence result in each workflow family. Code breadth and packet completeness are reported separately from authority.

Current Family-Level Ceiling

family strongest current honest sentence main limiting layer
dda review-grade, bounded the black-box dashboard defends review_grade_bounded; search remains import-backed and no repository-owned raw search lane closes that gap
dia outsider-auditable, bounded benchmark and runtime strength still narrow under library incompleteness and consequence pressure
lfq outsider-auditable, bounded the checked cohort and sparse-contrast lanes pass the current audit boundary, while normalization and transfer still block wider language
multiplex internal support only outsider trust still collapses under the current stress packet
ptm outsider-auditable, bounded localization evidence remains stronger than consequence confidence
targeted outsider-auditable, bounded calibration, interference, and downstream burden still narrow the sentence

Repository-Wide Limits

  • several families are now publicly challengeable, but there is still no repository-wide proof set that supports broad proteomics authority under the same standard
  • stronger core biology and chemistry coverage does not by itself erase transfer, runtime, or consequence limits
  • one strong family packet does not widen the sentence for weaker families
  • recommendation posture still narrows when grounded contradiction or downstream assay burden becomes explicit

These sentences reflect the current black-box result, not the strongest status requested by a release policy or an individual evidence root. For example, DDA benchmark freshness can support an outsider-auditable request while the end-to-end dashboard still lowers the permitted sentence to review-grade.

Consequence boundary

LFQ, PTM, and targeted remain bounded even though each has an outsider-readable packet. Their shipped consequence records end at exploratory-only follow-up; they do not establish broad clinical, production, or cross-laboratory utility.

The consequence maps expose conditions that can demote an otherwise strong analytical packet. A doubled assay burden, unresolved interference, weak requested-versus-observed agreement, or loss of a comparator or literature anchor can narrow the recommendation immediately.

Which Layer Currently Carries The Brake

layer what is already strong why it still does not widen the sentence alone
core broader biology, chemistry, spectra, mzML, PTM, DIA, and quantification depth stronger scientific machinery still needs transfer, runtime, and downstream proof
runtime replay, rerun kits, artifact stability, refusal, comparability reproducible runs do not settle scientific truth or assay worth
knowledge grounding, contradiction, citation pressure, biological context grounded sentences can still be analytically overconfident or too costly to act on
intelligence challenge, downgrade, regret, recommendation posture stronger analytical policy cannot outrun weaker benchmark, grounding, or consequence layers
lab control demand, readiness, refusal, outcome reconciliation practical follow-up can still be too narrow to justify broader public release language

Why The Limits Sit In Different Layers

The current ceiling is not caused by one kind of weakness only:

  • some limits are benchmark-transfer limits
  • some are runtime-lane and execution-realism limits
  • some are grounded-claim and contradiction limits
  • some are recommendation-pressure and regret limits
  • some are lab-readiness and consequence-burden limits

That is why no single package can honestly declare the whole repository ready on its own.

Capability That Does Not Remove The Limits

  • core is materially deeper across sequence, chemistry, spectra, mzML, DIA, PTM review, and quantification
  • runtime now has real replay, rerun, refusal, and artifact-stability routes
  • knowledge and intelligence now expose grounded contradiction, downgrade, and recommendation-pressure surfaces directly
  • lab now makes control demand, refusal, and requested-versus-observed follow-up visible enough to limit public language honestly

These capabilities are useful within their declared boundaries. None changes a family verdict until the governed evidence matrix records the stronger result.

What Stronger Code Still Does Not Automatically Earn

  • broad cross-family release language
  • decision-grade recommendation authority
  • vendor-parity execution claims
  • universal transfer across study design, instrument class, or sample complexity
  • stronger lab-confidence language than the current follow-up burden supports

Evidence Required Before Language Widens

  • stronger family-specific transfer packets where benchmark depth is currently narrower than the release sentence
  • runtime proof that survives more realism pressure where import_only or bounded raw_executable lanes still cap trust
  • grounding routes that keep contradiction and citation pressure visible when workflow packets look operationally strong
  • recommendation artifacts that survive challenge without leaning on weak lab consequence
  • downstream evidence that follow-up burden and requested-versus-observed outcomes justify stronger public claims

Closure requires a new checked record at the limiting layer, a rerun of the family matrix, and agreement between the generated verdict and the public sentence. New code or a new package directory is not closure by itself.

Inspect the limiting layer

The public sentence follows the weakest live layer. A wider sentence requires new evidence at that layer and a passing governed matrix; it cannot be inferred from implementation depth elsewhere in the system.