Workflow Refusal Handbook

A refusal record distinguishes four operational responses. Each response preserves the current evidence and names what can happen next.

response	meaning	next admissible action
stop	the current handoff must not proceed	preserve state and inspect the named blocking condition
rerun	the question remains valid and recoverable evidence is missing or invalid	correct the declared condition and create a new run record
narrow	a weaker scope remains supported	issue a revised bounded question or recommendation
refuse	no responsible action exists inside the current policy or authority	retain the refusal until new evidence or authority changes the precondition

dda

• current posture: recommend_with_downgrade

stop when

• comparator drift or missing external execution parity still materially limits this public workflow claim
• Do not treat a single repeat as decision-grade if calibration drift reappears.
• Do not proceed if digest reproducibility control material is unavailable.

rerun when

• external comparator claim support is still advisory
• claim support is not yet strong enough for an unqualified recommendation
• comparator drift or missing external execution parity still materially limits this public workflow claim
• shared-peptide pressure changes protein-level conclusions even when peptide counts look stable
• contaminant promotion inflates confidence when blank carryover is not inspected

narrow when

• Cross-engine DDA drift still makes protein-facing promotion unsafe beyond the bounded benchmark story.
• Promote the paired DDA transfer loss into a dedicated literature-backed claim row and keep protein-facing trust downgraded until live replay or stronger cross-engine proof lands.
• external comparator claim support is still advisory
• claim support is not yet strong enough for an unqualified recommendation

refuse when

• comparator drift or missing external execution parity still materially limits this public workflow claim
• Do not treat a single repeat as decision-grade if calibration drift reappears.
• decision-grade condition is not satisfied: Target-decoy evidence remains visible and calibration stress does not collapse confidence framing.
• decision-grade condition is not satisfied: Comparator evidence or known-loss dossiers keep search-adapter behavior honest against external engines.

evidence paths

• artifacts/intelligence/recommendation-packets/dda.json
• contradiction_triage:dda:1
• blank
• pooled_reference
• digest_reproducibility_reference
• carryover_blank
• artifacts/lab/flagship-follow-up-outcomes/dda.json

dia

• current posture: recommend_with_downgrade

stop when

• comparator drift or missing external execution parity still materially limits this public workflow claim
• Do not use the run for a decision-grade claim if the library reference is missing.
• Do not interpret successful extraction as biological closure when expected peptides still disappear without explanation.

rerun when

• external comparator claim support is still advisory
• claim support is not yet strong enough for an unqualified recommendation
• vendor and library comparison gaps remain open
• comparator drift or missing external execution parity still materially limits this public workflow claim
• library incompleteness hides true peptide absence behind extraction failure
• ion-mobility or vendor-conditioned assumptions make the output look richer than the evidence posture warrants

narrow when

• DIA transition-grade confidence still conflicts with wider vendor- and library-parity expectations outsiders may assume.
• Keep DIA bounded to library-conditioned review until broader vendor-conditioned confrontation and rerun proof exist.
• external comparator claim support is still advisory
• claim support is not yet strong enough for an unqualified recommendation

refuse when

• comparator drift or missing external execution parity still materially limits this public workflow claim
• Do not use the run for a decision-grade claim if the library reference is missing.
• decision-grade condition is not satisfied: Import, transition, protein, and biological interpretation tiers all remain above bounded scientific thresholds.
• decision-grade condition is not satisfied: Library coverage, ion-mobility coverage, and absent-expected-peptide pressure all remain inside documented interpretation limits.
• library dependence was already the dominant public cap on DIA authority
• the thinner package family still showed unstable transfer under matrix shift
• The matrix-shift repeat exposed library-conditioned fragility, so the follow-up consumed queue and still forced the recommendation back to refusal.

evidence paths

• artifacts/intelligence/recommendation-packets/dia.json
• contradiction_triage:dia:1
• blank
• pooled_reference
• library_reference
• bridge_sample
• artifacts/lab/flagship-follow-up-outcomes/dia.json

lfq

• current posture: recommend_with_downgrade

stop when

• comparator drift or missing external execution parity still materially limits this public workflow claim
• Do not spend the assay if only one extra sample can be added to a fragile contrast.
• Do not proceed when the run order and bridge design cannot be controlled tightly enough to learn anything new.

rerun when

• external comparator claim support is still advisory
• claim support is not yet strong enough for an unqualified recommendation
• comparator drift or missing external execution parity still materially limits this public workflow claim
• MNAR missingness makes the apparent rescue of a contrast look stronger than it is
• batch drift dominates the signal when bridge material is absent or underused

narrow when

• LFQ review-grade abundance confidence still conflicts with the harder cohort and missingness behavior named by the literature.
• Keep LFQ decision-grade wording blocked until harsher cohort pressure and observed outcome closure both exist.
• external comparator claim support is still advisory
• claim support is not yet strong enough for an unqualified recommendation

refuse when

• comparator drift or missing external execution parity still materially limits this public workflow claim
• Do not spend the assay if only one extra sample can be added to a fragile contrast.
• decision-grade condition is not satisfied: Missingness mechanism and replicate structure stay explicit enough that abundance summaries do not overclaim robustness.
• decision-grade condition is not satisfied: Batch posture and normalization drift remain inside bounded interpretation limits.
• repeatability was never the missing piece; cohort design weakness was
• the public package already showed missingness pressure large enough to block escalation
• The extra LFQ repeat only confirmed that missingness and cohort-shape weakness still dominate, so the loop consumed time without improving biological clarity.

evidence paths

• artifacts/intelligence/recommendation-packets/lfq.json
• contradiction_triage:lfq:1
• blank
• pooled_reference
• batch_bridge
• replicate_balance_audit
• artifacts/lab/flagship-follow-up-outcomes/lfq.json

multiplex

• current posture: do_not_recommend

stop when

• no additional family-specific condition is published

rerun when

• public comparator-backed claim support is refused
• biological grounding remains thin
• vendor and library comparison gaps remain open
• operational burden remains too high for a justified recommendation

narrow when

• Multiplex has a paired public benchmark surface, but the companion stress result defeats outsider authority and no dedicated lab consequence packet closes the downstream gap.
• Keep multiplex internal support only until companion pressure passes and dedicated outsider review and lab consequence packets exist.

refuse when

• keep multiplex at internal support until the family earns its own outsider review and lab consequence closure
• Multiplex has a paired public benchmark surface, but the companion stress result defeats outsider authority and no dedicated lab consequence packet closes the downstream gap.

evidence paths

• artifacts/intelligence/recommendation-packets/multiplex.json
• contradiction_triage:multiplex:1

ptm

• current posture: recommend_with_downgrade

stop when

• operational burden remains too high for a justified recommendation
• comparator drift or missing external execution parity still materially limits this public workflow claim
• Do not proceed when site ambiguity is still the main story.

rerun when

• operational burden remains too high for a justified recommendation
• external comparator claim support is still advisory
• claim support is not yet strong enough for an unqualified recommendation
• comparator drift or missing external execution parity still materially limits this public workflow claim
• site ambiguity survives the rerun and leaves the lab with a more expensive version of the same story
• enrichment or motif pressure creates a convincing signal without a targetable site-specific conclusion

narrow when

• PTM localization confidence still conflicts with the temptation to read occupancy or regulation into a narrower phospho-oriented evidence surface.
• Keep PTM release language bounded to localization and ambiguity until broader family-specific comparator pressure exists.
• external comparator claim support is still advisory
• claim support is not yet strong enough for an unqualified recommendation

refuse when

• operational burden remains too high for a justified recommendation
• comparator drift or missing external execution parity still materially limits this public workflow claim
• decision-grade condition is not satisfied: Localization confidence remains high enough that ambiguous site groups do not dominate the claimed biology.
• decision-grade condition is not satisfied: Only PTM families with explicit credibility tracks and scope limits can support decision-facing interpretation.

evidence paths

• artifacts/intelligence/recommendation-packets/ptm.json
• contradiction_triage:ptm:1
• enrichment_blank
• site_localization_reference
• unmodified_counterpart_control
• artifacts/lab/flagship-follow-up-outcomes/ptm.json

targeted

• current posture: recommend_with_downgrade

stop when

• operational burden remains too high for a justified recommendation
• comparator drift or missing external execution parity still materially limits this public workflow claim
• Do not proceed when heavy references or calibration standards are missing.

rerun when

• operational burden remains too high for a justified recommendation
• external comparator claim support is still advisory
• claim support is not yet strong enough for an unqualified recommendation
• comparator drift or missing external execution parity still materially limits this public workflow claim
• coeluting interference produces clean-looking transitions that still misstate the biology
• heavy-light mismatch or calibration drift turns the panel into an operationally neat but scientifically weak artifact

narrow when

• Targeted operator-facing confidence still conflicts with missing Skyline-class comparator and calibration realism.
• Keep targeted language out of calibration-clean and vendor-parity authority until the missing confrontation lands.
• external comparator claim support is still advisory
• claim support is not yet strong enough for an unqualified recommendation

refuse when

• operational burden remains too high for a justified recommendation
• comparator drift or missing external execution parity still materially limits this public workflow claim
• decision-grade condition is not satisfied: Chromatogram QC, calibration standards, and interference behavior all remain explicit before transition handoff.
• decision-grade condition is not satisfied: Blank, heavy-reference, and calibration-standard controls all stay visible before claims leave advisory status.

evidence paths

• artifacts/intelligence/recommendation-packets/targeted.json
• contradiction_triage:targeted:1
• blank
• heavy_reference
• calibration_standard
• interference_scout_injection
• artifacts/lab/flagship-follow-up-outcomes/targeted.json

Rule

If the best downstream action is stop, rerun, narrow, or refuse, the public recommendation remains weaker than a full recommendation. A retry creates a new run or decision record; it does not erase the refusal that justified it.