Delay analysis methods

A method whose data requirements your project record cannot meet is not an option. It does not matter that time impact analysis fits the fact pattern if the monthly updates were never statused, or that a collapsed as-built would be devastating if the as-built dates exist only as percentage complete. What is left after the record set has done its filtering is usually two methods, sometimes one, and occasionally none that a tribunal will accept.

So the order of the questions is: what do we have, then what can be run on it, then which of those survives cross-examination. Not the reverse. Delay analysis proves element two of the four-element test every extension of time claim has to satisfy (causation) and a method that cannot be run on the available data proves nothing at all.

Which methods can your records actually support?

Work from the record set inward. A baseline and an as-built alone support only the as-planned versus as-built family. A reliable series of contemporaneous updates supports windows analysis, which is the best-evidenced position available. Missing or corrupt updates push the analyst into recreated schedules or an as-built-only method, both of which are expensive and both of which invite attack. The table below is the whole practical payload of this subject.

Two dependencies in that table are worth stating in their own right, because they are where most projects actually fail. First, no accepted baseline means no datum. The opponent attacks the baseline as unrealistic, and the employer's standing defence (the baseline was never accepted, so the contractor's own plan proves nothing) has no answer. Second, the loss of contemporaneous updates is the single largest evidential loss available on a construction project. It removes the entire dynamic family in one move and leaves only static methods that cannot show the critical path as it actually shifted.

SCL Delay and Disruption Protocol, 2nd Edition (February 2017), Core Principle 1 recommends that the parties agree the record regime in writing at or before the outset: types, who produces and checks them, frequency, format, ownership and storage. Almost nobody does it. It remains the cheapest risk mitigation available on any project, and it is the only point at which the method question is genuinely open. The full record-category analysis, and what each absence forecloses, sits in the records that decide whether you have a claim at all.

One contract override sits above all of this. If the contract mandates a method, it governs regardless of technical preference. NEC4 ECC clause 63.5 effectively mandates a prospective impact assessment against the Accepted Programme, and many US and Middle East EPC contracts mandate time impact analysis by name. A mandated method whose data requirements the project cannot meet is a contractual problem, not merely a forensic one.

How the methods sort: three axes, not a list of six

The six methods in the SCL Delay and Disruption Protocol, 2nd Edition (February 2017) are not a flat list. They sort on three axes that decide everything about how an analysis behaves: whether it is prospective or retrospective, whether the critical path is treated as static or dynamic, and whether it reasons cause to effect or effect to cause. Learn the axes and the six methods stop needing to be memorised.

Read from the SCL Protocol 2nd Edition (February 2017). The three-axis sort is inferential: the corpus's own, not a classification the Protocol itself states.

The static/dynamic axis is the one that decides credibility on a real project. "The" critical path is a false singular: multiple critical paths, near-critical paths compressed to within days of criticality, and paths that become critical mid-project are all normal. Any analysis premised on a single fixed critical path across a whole project is either describing a very simple project or producing a modelling artefact. That is precisely why the dynamic methods: SCL Time Slice Windows, AACE MIP 3.3 and 3.4, exist: they re-determine the critical path in each window instead of assuming it.

The prospective/retrospective axis is set by the Protocol's own hinge. Core Principle 4 tells the parties not to "wait and see" and to deal with Employer Risk Events as the work proceeds. Core Principle 5 states the prospective test, grant an extension to the extent the event is reasonably predicted to prevent completion by the then prevailing date. Core Principle 11 is the hinge: where assessment occurs after completion or well after the event, "the prospective analysis of delay referred to in the guidance to Core Principle 4 may no longer be appropriate."

The choice of stance is itself a contested issue, and it moves the day count. Fluor Ltd v Shanghai Zhenhua Heavy Industry Co Ltd [2018] EWHC 1 (TCC) held that "a prospective analysis … does not necessarily produce the same answer as an analysis carried out retrospectively", expressly qualifying the suggestion in Walter Lilly & Co Ltd v Mackay [2012] EWHC 1773 (TCC) that a properly performed analysis of either kind should reach the same result. Where the contract picks a stance, it governs: NEC4 ECC clause 63.5 mandates a prospective assessment, and JCT's "fair and reasonable" assessment is prospective. An adjudicator revising a JCT extension of time must perform the contract's prospective assessment rather than a purely retrospective one (Mace Construct Ltd v Baltic Investment Holdings Ltd [2026] EWHC 976 (TCC)).

As to which method persuades: SCL method 4, as-planned versus as-built windows, is the one courts most often find persuasive, because it is fact-rooted rather than model-rooted. Methods 1 and 6 (impacted as-planned and collapsed as-built) are the most attacked. Method 5 is rarely used.

How the SCL methods map onto AACE 29R-03, and where they do not

They map, but they are not translations of each other, and the gaps run both ways. AACE International RP 29R-03, Forensic Schedule Analysis (rev. 25 April 2011) has no dedicated method implementation protocol for the SCL Protocol's Retrospective Longest Path. The SCL Delay and Disruption Protocol, 2nd Edition (February 2017) has no analogue for MIP 3.4, the contemporaneous split, or for MIP 3.5, recreated updates. Anyone writing as though the two taxonomies are a one-to-one dictionary is wrong in both directions.

Inferential where the taxonomies do not meet: the pairings follow Popescu's harmonisation, and the "no analogue" rows are the corpus's reading of the gaps, not a published concordance.

The mapping is drawn from Anamaria Popescu (Ankura), Harmonizing SCL D&D2 and AACE 29R-03 (5 November 2025, 11th International SCL Conference, Seoul), whose thesis is worth carrying whole: SCL supplies the strategic and principled frame, 29R-03 supplies the technical protocol, and a competent expert uses both. (The first row is worth a flag on the fact-versus-inference line: the Protocol's six-method table names the windowed variant. Treating the single gross comparison as the same family run once is an inference from the mapping, not a statement in the Protocol.)

Cite the AACE layers, not just the number. The useful part of 29R-03 is the taxonomy, which lets an expert say exactly what was done. The basic method is Observational (examine the schedules as they are) or Modelled, insert or extract delays and re-run CPM. Observational sub-layers are Static (single fixed critical path) versus Dynamic (critical path tracked as it shifts), and Gross (one analysis) versus Periodic (segmented). Modelled sub-layers are Additive (insert delay fragnets) versus Subtractive (extract them), and Single Base / Multiple Base / Single Simulation. Stating "MIP 3.3, observational / dynamic / contemporaneous as-is" tells a tribunal precisely what was done. Stating "a windows analysis" does not.

MIP 3.4, the gap on the SCL side, is not an academic curiosity. It isolates non-progress revisions (logic changes, calendar changes, re-sequencing) from actual progress variance, which is exactly the separation a concurrency argument needs to be tested properly. 29R-03's own commentary treats MIPs 3.1 to 3.5 as generally inadequate for rigorous concurrency and pacing analysis. That usually requires a modelled or split approach plus factual records. And 29R-03 is not the whole of AACE's claims guidance: AACE International RP 130R-23, Demonstrating Entitlement to Cumulative Impact Claims in Construction (15 May 2023) covers the cumulative-impact side and is frequently missed.

Time impact analysis or impacted as-planned?

Both are prospective, both reason from cause to effect, and both work by inserting delay fragnets into a schedule and re-running CPM. The difference is the base. Impacted as-planned inserts the fragnets into the original as-planned baseline: one base for every event, which is why AACE International RP 29R-03 (rev. 25 April 2011) classifies it as MIP 3.6, Modelled / Additive / Single Base. Time impact analysis inserts each fragnet into the baseline as updated to that event's point of impact: a different base per event, hence MIP 3.7, Modelled / Additive / Multiple Base.

That distinction has a hard practical consequence, and it is the records point again in a different costume: a project with no reliable contemporaneous updates cannot run time impact analysis at all. It will be pushed down to impacted as-planned, the method with the lowest data requirement and the worst reputation. The choice between MIP 3.6 and MIP 3.7 is very rarely a free one.

The SCL Protocol's 2nd Edition changed its own position here, and the change is under-reported. The 1st edition (October 2002) expressed a preference for time impact analysis. The 2nd Edition (February 2017) removed that preference for retrospective work. Time impact analysis is now positioned as the contemporaneous, prospective tool under Core Principle 4, and is expressly not endorsed for time-distant analysis under Core Principle 11. Citing the Protocol as authority for running a TIA years after completion is citing a superseded edition.

One further trap for prospectively-assessed forms. Under NEC4 ECC, the assessment stands or falls on the Accepted Programme, and nobody can invent one: in Premier Modular Ltd v Maidstone and Tunbridge Wells NHS Trust [2026] EWHC 1404 (TCC) an NEC4 adjudication decision was refused enforcement in part because the adjudicator relied on an "updated" Accepted Programme when both parties agreed the programme had not been updated.

Is a named method a rule of law?

No. Three common-law jurisdictions have said so within the last seven years, in three different ways, and none of them treats either the SCL Delay and Disruption Protocol, 2nd Edition (February 2017) or AACE International RP 29R-03 (rev. 25 April 2011) as binding. What each requires instead is that the method be driven by the factual evidence, that the analyst say what was done, and that every departure from the named method be explained.

Jurisdictional. White Constructions (NSWSC), Thomas Barnes (TCC) and Walsh (ONSC) are all first instance. No appellate authority on method choice was located in any of them.

Two negative findings belong here, stated plainly rather than hedged, because knowing what cannot be argued is worth more than a paragraph of speculation.

There is no US decision between 2023 and 2026 condemning a named forensic delay methodology. Not time impact analysis, not collapsed as-built, not windows. A cross-examination premised on the proposition that a named method is discredited as such has no authority behind it in US practice.

There is no reported judgment deciding a case on "float suppression" as a named doctrine. Float suppression: artificial constraints, excessive lags, preferential logic dressed as physical logic, duration padding, open ends, resource-levelling artefacts, quiet baseline re-sequencing: is real, and it is run against baselines and updates by both sides. But it operates as an evidential and credibility attack, not a cause of action. Pleading it as though it were a freestanding wrong invites a strike-out of the label and loses the substance with it.

The convergent rule across every forum: pick the method the records support, state the taxonomy label, state the assumptions, and explain every departure from the named method.

What tribunals actually punish

Data quality, not method choice. The recurring findings across US, English and Canadian decisions are about missing baselines, absent updates, progress recorded as percentages rather than actual start and finish dates, and schedule logic that the contemporaneous documents do not support. A modest windows analysis run on clean statused monthly updates beats a methodologically orthodox analysis run on reconstructed or re-logicked schedules, every time.

Alares Construction, Inc. v Department of Veterans Affairs, CBCA 6149, 7071, 7597 (21 March 2025) is the sharpest recent demonstration of what that costs. The contractor claimed the Department of Veterans Affairs had caused all 765 days of delay. The Board found 218 days of VA-responsible compensable delay, adopting the owner's expert's critical-path analysis and drawing adverse inferences from unexplained schedule logic changes in the contractor's monthly updates. The contract required monthly updates reflecting activity and event changes and accurate actual-versus-planned progress. The failure to maintain them cost the contractor the benefit of the doubt on every contested period.

The older US authority is to the same effect. In George Sollitt Construction Co. v United States, 64 Fed. Cl. 229 (2005) the expert's analysis was discounted for missing baseline schedules, absent monthly updates during startup and completion, progress reported as percentages rather than actual start and finish dates, and inaccurate critical-path activity dates. None of those findings is about the technique the expert selected. They are all about the contemporaneous record the technique was run on.

Which edition are you citing?

Cite both frameworks with their dates, every time, because readers assume there is a newer one and because one of them has not been revised in fifteen years. As at August 2026 the SCL Delay and Disruption Protocol is in its 2nd Edition, February 2017, superseding the 1st edition (October 2002) and Rider 1 (July 2015). AACE International RP 29R-03, Forensic Schedule Analysis, stands at rev. 25 April 2011.

Both of those statements carry a qualification that ought to travel with them into an expert report.

AACE 29R-03. The public table of contents still shows rev. 25 April 2011. A draft revision circulated around 2023: regrouping the methods into four groups, with a public critique arguing that collapsed as-built should be moved to "available but not recommended", and no published revision has been located. AACE's member catalogue was not accessible for this note, so treat rev. April 2011 as current but check the AACE library before citing the revision date in an expert report. Citing 29R-03 undated reads as though the writer does not know it has not been revised in fifteen years.

SCL Protocol. No 3rd edition exists and no revision programme is announced. That finding rests on the absence of any 3rd edition on the SCL site and in 2025–26 commentary, and SCL International's protocols page was not retrievable. High confidence, not certainty.

One thing not to do: do not quote a paragraph number for the Protocol's six-method table. It is variously cited between 11.5 and 11.7 across published commentary, and the discrepancy is not resolved here. Cite the Protocol, the edition and the method by name. Leave the paragraph number out unless you have the PDF open in front of you. The same discipline applies to every clause citation in a delay report: form, edition and clause travel together or the citation is not a citation.

Choosing between the methods your records will support

The real choice is narrower than the literature suggests. You are not choosing among six SCL methods or nine AACE method implementation protocols. You are choosing among the two or three that your record set can carry, filtered further by whatever the contract mandates and by whether the assessment is contemporaneous or time-distant. Run those filters honestly and the decision usually makes itself.

The filters, in the order they actually bite:

1. What does the record set contain? An accepted baseline with logic, constraints and calendars? A complete statused update series with narratives? Native schedule files or PDFs? Daily reports, and RFI and submittal logs dated on both sides? Answer this before naming a method, not after.
2. Does the contract mandate a method or a stance? NEC4 ECC clause 63.5 effectively mandates a prospective impact assessment against the Accepted Programme. Many US and Middle East EPC contracts mandate time impact analysis by name. A mandated method governs regardless of technical preference.
3. Is the assessment contemporaneous or time-distant? SCL Delay and Disruption Protocol, 2nd Edition (February 2017), Core Principles 4 and 5 govern while the works are live. Core Principle 11 is the hinge to retrospective analysis once the assessment is time-distant.
4. Is concurrency in issue? If so, note that AACE International RP 29R-03 (rev. 25 April 2011) treats MIPs 3.1 to 3.5 as generally inadequate for rigorous concurrency and pacing analysis without a modelled or split approach plus factual records.
5. Have you stated the taxonomy label and explained every departure? Name the SCL method and the AACE MIP with its full taxonomy string, then justify each departure: reasoned and grounded in contemporaneous records under Thomas Barnes & Sons plc (in administration) v Blackburn with Darwen Borough Council [2022] EWHC 2598 (TCC) and Walsh Construction v Toronto Transit Commission, 2024 ONSC 2782. Fatal when silent.

If the filters leave nothing (no accepted baseline, no update series, no individuated causal links) what is left is a total time claim: the extension asserted as actual completion minus contractual completion, against a bundle of employer events with no critical-path attribution. It is the weakest form of extension of time claim there is, and SCL Delay and Disruption Protocol, 2nd Edition (February 2017), Core Principle 17 discourages exactly that practice. Advanced as a first resort it reads as a concession that the records were not kept, and tribunals read it that way.

Which returns the whole subject to where it started. The method is not chosen when the dispute starts. It is chosen when the record regime is set, or allowed to drift, at the outset of the works. By the time anyone is arguing about MIP 3.6 versus MIP 3.7, that decision was taken months or years earlier by people who were not thinking about a claim at all.

About this material Library content is general information about construction claim practice, not legal advice. Entitlement, deadlines and procedure are governed by your own contract and by the law of the place the work is performed. Reviewed August 7, 2026 · Report a correction