RMP.4 - Preserve a Criticism-Ready Research Trace

Type: Research Method Practice practitioner method pattern Status: Development preview

RMP.4:1 - Problem Frame

Use this pattern when research material must remain inspectable, or when a colleague receives an apparent result but cannot recover how the work produced it. Begin by naming one examination the material should support and locating the records on which that examination depends.

For example, residents recorded downstream salinity during a thaw, but the reference reach could not be visited. Keeping the chart and a tidy folder is useful, yet it cannot restore the missing paired observation. A useful trace identifies the missing site and period, the reason, what the remaining observations can support and which comparison is no longer available.

This pattern governs the account of an inquiry’s actual course and retained material for a named examination. It connects plans, conduct, changes and the material used in a result. Its first useful result is a ResearchTracePackage: an inspectable account with the necessary records or reliable references to them, including limitations that change the receiving use. It need not be one archive or a copy of all the data.

“Criticism-ready” means that the material needed for the declared examination can actually be inspected under the stated conditions. A partial package can honestly explain why that examination is impossible. It cannot claim readiness for the examination its own limits prevent.

Enter while work is being planned or performed, before important records are lost. Alternatively, enter directly with inherited work and surviving material. There is no requirement to invent a new prospective protocol for completed work or to repeat RMP.1–RMP.3.

If an existing competent notebook, laboratory record or repository already supports the intended examination, use it. RMP.4 is needed when the conduct-to-material connection, change history or receiving limitation needs work. Evaluating the substantive explanation, reproducing a computation, issuing an access decision and performing new research remain separate contributions.

RMP.4:2 - Problem

Research accounts are often written from the final result backwards. Earlier plans disappear, a missed observation becomes a blank cell, failed computations are omitted, and consequential human or AI choices become an undifferentiated “processing” step. A reader sees the conclusion but cannot examine how the available material came to bear on it.

Preserving everything is not a sufficient answer. Large collections can omit the decisive exclusion or timing information, be inaccessible in practice, or disclose material that its holders have not authorized for that use. A public summary may be readable while lacking the detail needed for a particular criticism.

How can a practitioner preserve enough of actual inquiry to make its important connections inspectable, recover inherited work honestly, and expose the exact limits of loss, reconstruction and restricted access?

RMP.4:3 - Forces

RMP.4:4 - Solution

Name the receiving examination, preserve its consequential material before loss, connect the records to actual choices and results, and state what the receiving reader can and cannot inspect. Keep original material, supported reconstruction and irrecoverable loss distinct throughout.

RMP.4:4.1 - Name the examination and choose the entry

State what someone is trying to examine. They may need to understand an exclusion, test whether a comparison was actually made, inspect an interpretation’s grounding, or recover the inputs behind a reported computation. Identify the bounded question, design and material at issue.

A final claim is not a prerequisite. During preparation, the examination can concern whether the planned comparison was carried out or whether the forthcoming analysis has the material it needs. Preserve that intended use without inventing a result or ClaimCandidate. When a result is later supplied, connect its actual uses to the retained material.

Choose the entry that fits the work:

• Planned or ongoing work: recover the question/design and operational choices; identify what would be lost unless captured now; arrange and maintain the actual records.
• Existing or inherited work: begin with the receiving question, claimed result if any, and surviving original records. Recover the design, protocol and conduct only as far as those records support them.

RMP.1–RMP.3 can supply missing question, design or operational decisions. Direct entry is appropriate when compatible inputs already exist. In a legacy inquiry, uncertainty about an earlier design is part of the account, not permission to attribute a newly written protocol to earlier researchers.

Bound the package by the examination. Name which study portion, source set, encounters, specimens or computation it covers and the relevant period or version. If the receiving use changes, reconsider whether the preserved material is sufficient for that new use.

RMP.4:4.2 - Preserve the smallest material whose loss would matter

Ask what the critic would be unable to distinguish if one record disappeared. Start with those distinctions. Examples include which specimen was measured, which document was inspected, why a source was excluded, or whether an outcome was already visible when a choice was made.

Arrange capture at the point where the relevant event or choice occurs. An existing qualified laboratory or data practice may already do this. Reference its records and qualifications rather than reproduce them in an RMP form.

Preserve the applicable plans with their versions and timing. A registration, protocol and analysis intention may have different contents and dates. Keep enough to compare intention with conduct, including amendments. A record’s title cannot establish its prospective status.

Distinguish these evidential situations in ordinary language:

Retain the distinction even if all four occur in the same notebook or account. Date a reconstruction as a reconstruction. A polished description of earlier intentions does not establish that those intentions preceded data access.

Actual retention, access and authorized destruction requirements come from their owners. Preserve permitted evidence of loss and its consequences where possible; do not retain prohibited material merely to improve a trace.

RMP.4:4.3 - Connect actual conduct, transformations and uses

Trace one consequential result or planned comparison back to its material. Identify the actual observation, source, encounter or computation and the selection or transformation between it and the receiving use.

For observations, retain the specimen, case, participant, site or sample identity at the permitted precision, together with the conditions needed to interpret it. Keep raw indications, qualified measurement results and derived values distinguishable. FPF C.16 supplies measurement discipline; a spreadsheet cell alone does not establish its unit, uncertainty or qualification.

For sources, distinguish a catalogue description from a document actually inspected. Preserve source and edition, locator, dating and provenance, the selection reason and any relevant transcription or translation. Keep contrary material and unexamined source families visible where they affect the question.

For transformations, retain the input–operation–output connection at the granularity the examination needs. Identify consequential filtering, recoding, deduplication, aggregation, translation, manual correction and model or computational steps. Preserve the actual version and parameters when they can change the result. FPF A.10 supplies the general claim-bound evidence/provenance discipline; this pattern makes the inquiry’s actual research-use connections recoverable.

For computation, a useful trace can connect input versions, code and environment dependencies, parameters or seeds where relevant, the particular run and its outputs. Include relevant failures and manual interventions. An executable script plus a data folder does not by itself identify the run behind a displayed result.

For qualitative work, retain the grounding material and the researcher’s consequential interpretations and changes under the selected approach. Researcher positioning or a change in questioning may be essential evidence; a universal coding-agreement score may not be appropriate.

For evidence synthesis, preserve searches as performed: sources or platforms, dates, query versions and resulting material, followed by screening, duplicate handling, exclusions and extraction changes. This supports examination of how the corpus was obtained. It does not perform appraisal or synthesis merely by recording their intended Methods.

In a mixed inquiry, preserve each component’s trace and the actual combining or interpretive choices. Expose shared data and assumptions. Two streams of records do not establish independent corroboration if both depend on the same selected participants or retrospective source.

RMP.4:4.4 - Explain consequential changes and contributions

Preserve the earlier intention and actual change wherever their difference matters. State what changed, when, why, who made the consequential choice, which material it affects and what relevant data or results were available by then. Record uncertainty instead of filling an unknown reason with a plausible one.

Distinguish events with different consequences. A missed observation, excluded case, failed run, inaccessible source and revised interpretation are not interchangeable “cleaning.” A record can be intact yet excluded from a particular analysis; an observation can be absent because the event was missed rather than because its value was extreme.

If a choice was made after relevant outcomes were seen, retain that timing. Such a choice may still be justified for the actual Method and use, but cannot supply an advance commitment. A later sensitivity analysis or independent criticism is separately performed work, not a retrospective repair of the original timing.

Identify substantial human and tool contributions. For AI-assisted work, preserve what is needed to inspect the research-shaping contribution:

• the model or service identity available to the user, version or configuration when known, and timing;
• the relevant input context and output actually used, subject to the applicable protection and access conditions; and
• the responsible person’s checks, acceptance, rejection or revision and the effect on the research work.

Exact prompt content belongs in the retained trace when it can change the examined research choice. Routine spelling or editorial assistance need not generate a transcript archive. An unavailable internal model state remains unavailable; do not invent it to make a service appear replayable.

If an AI output proposes a source that does not exist, preserve its rejection where that matters to the account and use the actual checked source as evidence. The rejected output is not a source in the scientific chain.

RMP.4:4.5 - Determine actual access and the consequences of loss

Confirm what material survives and how the receiving reader can inspect it. A repository link, an access policy and a custodian’s assurance answer different questions. A public summary may omit the exact detail required for a criticism even though an authorized reader could inspect the retained original.

Apply the following distinctions to the named examination, not as a universal grading scale.

A qualified custodian’s inspection result, when available, retains its own scope and limits. It is neither equivalent to public access nor a compulsory substitute in every study. An access request cannot itself establish that access has been obtained.

RMP.4:4.6 - Return the trace with a truthful receiving scope

Present the shortest account that lets the receiving reader locate the needed material and understand the action-changing connections and limits. It should recover:

• the question, design, intended examination and study portion covered;
• the surviving plans and the basis for claims about actual conduct;
• the material used, consequential selection and transformations;
• changes, their timing and human or tool contributions; and
• actual access, reconstruction, loss and the examinations each permits or prevents.

The same records can carry several of these meanings. A package with linked governed records need not duplicate them. Use stable identities or versions where a later change would otherwise make the account ambiguous.

Demonstrate one actual path through the account. Start from a displayed value, interpretive excerpt or planned comparison and locate its sources and choices. If the path ends at a missing record or unavailable access, state that endpoint and its consequence rather than filling it with a general assertion of transparency.

Return a limited package when it is the truthful useful result. For example: “The downstream measurements and handling decisions can be inspected; the intended paired thaw-event comparison cannot.” The package is useful because it makes the boundary actionable, not because it turns every receiving question into an answerable one.

RMP.4:5 - Archetypal Grounding

These constructed cases illustrate trace content and receiving limits. They report no actual environmental, laboratory, historical or computational result.

RMP.4:5.1 - ResidentStream-S17: preserve a useful trace after losing the comparison

The selected design would compare salinity-related observations at a downstream reach and a reference reach during an informative thaw event. Assume the measurement method, calibration basis, safe access and intended site-period relation were qualified when the design was chosen.

During the event, downstream observations are made, but the reference reach becomes inaccessible. The team retains the following linked account:

Two examinations now differ. A reader can inspect the downstream measurement and handling history. The same reader cannot perform the original paired thaw-event comparison because the reference observation does not exist. Keeping all the surviving files does not remove that difference.

A reference sample taken later in dry weather is new material under different hydrological conditions. Its use requires a justified comparability premise, not a change of file name or a routine replacement of the blank cell.

Suppose volunteers began sampling before any protocol was recorded. The team can reconstruct part of their procedure from contemporaneous sample labels, instrument logs and later reports. The reconstruction names those bases and their uncertainty. It is dated when reconstructed and does not claim that all choices were prospectively fixed.

Private land-access details can remain controlled under the actual permissions. The public account identifies the omitted material’s research role and the examinations possible without it. It promises neither the lost event nor unrestricted independent inspection.

RMP.4:5.2 - Pump-P4: distinguish absence, exclusion and interpretation

On the valid-marker branch of the seal study, specimen P4-07 is damaged during handling before one planned observation. The trace retains its allocation and material identity, the handling record, the time and basis of the reported damage, the absent observation, the applicable prior exclusion intention and the decision actually taken.

The comparison dataset links the remaining observations to their specimen records and preserves why P4-07 contributes no value at that point. Three readings from another seal remain three readings of that seal; trace organization cannot create independent specimens.

A critic can now examine whether the missingness or exclusion threatens the comparison. That judgement belongs to the applicable analytical or criticism Method. The trace neither repairs the lost observation nor grants authority to change the pump’s maintenance interval.

RMP.4:5.3 - A computation with a research-shaping AI suggestion

A researcher uses an AI service to suggest a preprocessing change for a model comparison. In this constructed case, the retained account identifies the supplied design, original data and code version, the relevant proposal and input context, and the researcher’s decision to revise and use the change. The actual run connects the changed code, environment and parameters to its output.

The service also suggests a nonexistent supporting article. The researcher rejects it after source checking and retains the actual source used. The false citation remains a rejected suggestion; it does not become evidence merely because it is preserved in the trace.

Later, the original service version is unavailable. A reader can inspect the retained suggestion, the human decision and the implemented change. They cannot infer that the same service response can be regenerated. A new computational run may test the retained implementation, but it is new work and does not establish replay of the original AI interaction.

RMP.4:5.4 - Historical and interpretive records support different criticisms

For an inquiry into whether a directive preceded a factory’s practice change, the historian retains inspected document identifiers and editions, dates and provenance, the distinction between catalogue entries and inspected contents, contrary evidence and transcription decisions. An unavailable earlier archive series leaves the rival earlier chronology unresolved; its absence is not evidence that the earlier decision never occurred.

An inquiry into how former workers understood the closure uses a different relation. The trace can preserve situated encounters, permitted excerpts, the researcher’s position and reasons for changing prompts or interpretation. A new prompt can be consistent with that inquiry while a move to a population-wide prevalence claim changes its question/design. Criticism follows the declared approach and grounding; the trace does not impose coder agreement as the universal test of interpretation.

RMP.4:6 - Bias-Annotation

A trace can faithfully preserve a biased selection. Ask whose cases, sources, encounters and failed work were never captured, and whether the account identifies the effect of that absence. A folder containing only successful runs or convenient sources invites a different reading from a record of the actual selection.

Retrospective knowledge can reshape memories of earlier reasons. Preserve the evidential basis of a reconstruction and the distinction between contemporaneous records and later accounts. More confident narration is not stronger evidence of timing.

Openness can distribute benefits and risks unevenly. Data holders, communities and participants may have legitimate authority over access and use. Record the resulting research limitation without either disclosing protected information or pretending that a public summary supplies hidden detail.

RMP.4:7 - Conformance Checklist

Use this checklist for the package’s declared receiving examination.

• The covered inquiry, study portion and examination are identifiable; an absent final claim is not replaced with an invented one.
• Plans, actual conduct, later reports, reconstructions and unknowns remain distinguishable.
• Consequential material can be followed through selection and transformations to its receiving use, at the granularity that examination requires.
• Changes preserve their earlier intention, actual effect, timing, reason and relevant prior information where supported; unknowns remain explicit.
• Missing observations, exclusions, failed runs, inaccessible sources and changed interpretations have their own supported consequences.
• Research-shaping human and AI contributions are inspectable to the needed and permitted extent; routine assistance does not trigger indiscriminate transcript capture.
• The account distinguishes surviving material, actual permitted access, public views, reconstruction and loss.
• Every claimed criticism-ready use has the necessary inspectable material. Prevented examinations are stated with their next source, access, narrower-use or design/question return.
• A new run or new data collection is represented as new work rather than as restoration of unsupported original conduct.

For a consequential reliance, follow one important use backwards to its material and compare a relevant prior commitment with actual conduct. Inspect the weakest required link and the access actually available to the intended reader. A reachable directory or an openness label is not the completion test.

RMP.4:8 - Common Anti-Patterns and How to Avoid Them

RMP.4:9 - Consequences

A receiving reader can distinguish what was planned, what is supported about actual conduct, how material reached a result, and where examination stops. Early capture reduces preventable loss; legacy entry can recover useful bounded accounts without pretending the past was better specified than it was.

The burden is selective recording, connection and access management. Some criticisms remain impossible, and some claims must be narrowed or withheld. This is a useful result when the alternative is an apparently complete account that silently substitutes confidence for missing evidence.

RMP.4:10 - Architectural Rationale

The trace is separate from the protocol because actual conduct can differ from intention, and inherited work may survive without a reliable prospective plan. Both results can be carried in the same notebook or governed collection; the distinction concerns their evidential uses, not document count.

Readiness is tied to a receiving examination because the same material can support one criticism and fail another. That relation gives a stopping rule for capture and a truthful account of loss. Universal capture has no comparable boundary and can create additional disclosure risk.

Existing domain records and generic provenance are the low-cost alternative. RMP.4 adds value where the research question, actual choices, transformations and receiving limitations need to be connected. It does not replace competent recordkeeping with a universal trace schema or transform a record into a warranted conclusion.

RMP.4:11 - SoTA-Echoing

Practice question. What should be retained and exposed so that a later reader can criticize actual research without confusing records, replay and scientific warrant?

Selected line. Adapt examination-relative provenance, timed change accounts and actually available access. Against either a final retrospective narrative or a publish-everything default, this preserves consequential differences with proportionate effort. The accepted trade-off is that some access-limited or irrecoverable material leaves narrower examinations possible rather than a universal openness or reproducibility claim.

This is a reasoned combination of methodological guidance, bounded failure evidence and practical record arrangements, not a universal empirical guarantee. Reopen the affected trace requirement when a new receiving use needs different material, a source or service change defeats a relied-on connection, or a lower-burden qualified practice preserves the same examination without the added apparatus.

RMP.4:12 - Relations

• RMP.3 supplies operational intentions and research-relative record needs. Capture can begin while those intentions develop.

• RMP.2 receives a lost comparison or other defeated design premise; RMP.1 receives a changed question or receiving use.

• FPF A.10 supplies claim-bound evidence and provenance discipline; FPF C.16 supplies the measurement account needed to distinguish indications, qualified results and derived values.

• Domain recordkeeping, source criticism, statistical and interpretive Methods supply their own conduct and results. Access, privacy, rights, community and resource owners supply the decisions that govern actual retention and inspection.

• RMP.5 uses the actual material and its limits to analyze and criticize the proposed claim. RMP.6 uses the recoverable basis to choose and interpret a relevant credibility examination. Their analytical and test results require their own Methods and actual work; a trace is not a performed reproduction test.

• RMP.7 can use the actual source material, relationships and receiving limits for a compatible corpus-synthesis question. The trace supplies that material; appraisal and integration require their own Method and result.

RMP.4:End