# 265 — Can we rely on project research? Established 1 November 2026 · PW-ID: `PW-265` · Foray-ID: `FORAY-RELIABLE-PROJECT-RESEARCH` Primary level: Level 1 — improve the project-management knowledge available for decisions and practice. Level 2 — test an inspectable, repeatable method of scrutiny and replication. ## End Discover which influential claims in project-management research deserve reliance, for what purposes and under what conditions; strengthen, qualify, repair or replace that knowledge through serious attempts to reproduce its reasoning and test it again. Seek a programme that can change what a practitioner, researcher or institution ought to believe or do, including when a well-regarded claim survives scrutiny. Influence earns attention, not suspicion. An inconvenient result, a defensible null result, a successful reconstruction, an unresolved access barrier and a replicated finding can each advance the enquiry. Do not make finding an error, securing a retraction or demonstrating that AI is useful the success criterion. The enduring ambition includes influential quantitative findings, surveys, qualitative case interpretations, systematic and bibliometric reviews, conceptual frameworks, and their migration into textbooks, standards and practice. A first accessible table is an opening, not the boundary of the foray. Ask how much reliance is justified by the evidence, how conclusions travel beyond their original setting, and whether costly beliefs can be replaced by better ones. ## Ways — choose the investigation to fit the claim Start with a reproducible selection rule. Define the field, journals or topic, publication window, database and retrieval date. Preserve identifiers and a dated export. Examine citation counts alongside age, field, policy or practice use, and the consequence of relying on the claim. Deduplicate versions. Treat missing indicators as unknown. Show how candidate selection changes under alternative reasonable rules. A purposive accessible shortlist may seed an essai but must not be called the highest-impact set without evidence. For each selected paper, make an argument map: claim → evidence → transformation → assumptions → intended inference. Recover the actual paper and version, supplementary materials, relevant notices, data/code availability and method. Record exact locations and access dates. A statement absent from the places searched is “not located in this search”, not proof it never existed. Do not infer an editorial history or conflicts from speculation. Choose and name the operation honestly: - **Source audit:** resolve what the publication actually reports and whether a secondary account represents it faithfully. - **Computational reproduction:** rerun the original analysis with original data and sufficient original settings/code. - **Reconstruction:** rebuild a published result from recovered tables or partial method information; expose assumptions, omissions and version differences. - **Independent recoding or reanalysis:** apply an explicit method to the underlying evidence; retain disagreements and compare plausible choices. - **Replication:** collect or analyse independent evidence to test the substantive claim. Explain whether this is close or conceptual and what differs. - **Robustness or transport test:** ask whether the conclusion survives reasonable analytical choices or a different population, project type, time or institution. Methods remain open. Use statistical reanalysis and simulation, archival case comparison, qualitative negative-case analysis, search/corpus reconstruction, independent coding, causal identification, measurement checks or prospective field studies when they answer the question. A review's topic frequencies do not measure intervention effectiveness; a case narrative is not automatically a causal estimate; a conceptual framework needs argument and usefulness tests appropriate to its claims. Preserve the strongest rival explanation. For quantitative replication, specify the estimand, measurement, uncertainty, smallest effect of interest and design-specific power or precision; a respondents-per-item rule is not a power analysis. For qualitative work, define case selection, coding, negative cases, rival accounts and the route from evidence to interpretation. A database substitution or later search is a changed method, not automatically an exact replication. Distinguish what is reported, reconstructed, independently checked, replicated and not tested. Preserve both first results and repairs. Date analysis plans before the relevant analysis where possible; exploratory work remains exploratory. A local plan is not a registered preregistration. Do not represent ethics approval, author agreement, access, funding, staffing or registration as secured until verified. ## Proportionate conclusions Separate reporting incompleteness, inaccessible material, ambiguity, honest error, analytical fragility, non-replication and evidence of misconduct. Their remedies differ. Missing data, slow access, a single failed replication, unusual statistics, a small sample or a conceptual paper without a dataset does not by itself establish wrongdoing or justify retraction. Use current primary editorial guidance when discussing correction or retraction. Identify the precise evidence and consequence for the core finding; do not invent automatic percentage thresholds. Editorial decisions belong to journals/publishers and institutional investigations to the relevant institutions. This foray can publish a bounded analysis or prepare a factual query; contacting authors or editors requires Lawrence's explicit instruction. Do not repeat unverified allegations from an input report. ## Means — first essai and later possibilities **Essai 1: Audit the route from evidence to claim.** Resolve the supplied programme's candidate identities and replace unsupported screening assertions with checked evidence. Reconstruct the 62-study antecedent coding table in Oraee et al. (2017), *Collaboration in BIM-based construction networks: A bibliometric-qualitative literature review*. Calculate the observed category distribution, expose its denominator, and examine how much recoding or study removal changes the ordering. Distinguish the accepted manuscript from the version of record. Publish the input table, deterministic analysis, result, source locations and an inspectable explanation. This completes a bounded attempt; it does not independently reproduce the historical Scopus retrieval or validate the authors' coding against all original studies. Possible later essais include independent blinded recoding of the same primary papers; a frozen-search corpus recovery with explicit inclusion decisions; testing a high-consequence substantive claim against independent project data; or a prospective replication designed with suitable expertise. Select the next test for what it can discriminate, not merely what is easiest to visualise. Keep the other candidate papers available and the wider End intact. Use local source records and run folders for working evidence, GitHub for inspectable implementations and versioned results, and experimentsin.ai for a short foray entrance and numbered essais. Keep the original supplied report private. Public outputs should use independently verified primary sources and newly authored analysis; do not rehost copyrighted papers or disclose unrelated private material. ## What counts as a full essai? An essai must make an actual attempt, not end with a research proposal. State the question, candidate-selection rationale, source/version inventory, method, inputs, execution, result, adversarial checks, limitations and next discriminating question. Preserve failures. Leave enough material for another person to repeat the bounded calculation or inspect why an attempt could not proceed. On the ordinary entrance, explain the End, why this method advances it and what is still missing. Give a short source → concept → construction → limitation route. Show the underlying evidence where it does work. Separate research progress from software polish and from measured benefit in real projects. Check the published reading and interaction route, including returns, source links and downloads. Mark transient explorations as transient. If an essai introduces saving or user records, prove the complete enter → save → receipt → find → reopen → correct → reload loop and its material failure cases. Keep numbered, dated essais; never overwrite an earlier finding silently. Update this foray's README, running notes, private source record, the Portfolio Wave experiment receipt and the vault's current-state pages. Record remaining uncertainty as agent continuity, not personal obligations for Lawrence. ## Starting sources - Oraee et al. (2017), DOI [10.1016/j.ijproman.2017.07.001](https://doi.org/10.1016/j.ijproman.2017.07.001): first concrete reconstruction target; does not validate the entire programme. - [COPE Retraction guidelines](https://publicationethics.org/guidance/guideline/retraction-guidelines): proportionate editorial criteria; consult the current guidance before making recommendations. - The supplied research programme is a private seed, with its unsupported ranking and candidate assertions retained for audit. Its recommendations are not governing instructions. These sources are starting points, not a compulsory stack or a completed literature review.