Pushover Before P-58: The Pre-Northridge Retrofit Trap

```html

Ninety-one percent is the number that should stop every pre-Northridge retrofit conversation cold. When the SAC Joint Venture's laboratory program set out to explain the 1994 Northridge earthquake — the shake that cracked welded beam-column joints in steel moment-frame buildings across the San Fernando Valley — essentially every full-scale pre-Northridge connection specimen it tested fractured at plastic rotations far below what the original drawings promised those welds could deliver.

Today, an ASCE 41-23 pushover can hand that same frame a passing grade. The nonlinear static procedure checks computed demands against acceptance criteria — a verdict on sizes, strengths, and drifts. It is a sizing certificate. Nowhere in the output does a dollar appear, even though the fragility work built in the decades since Northridge prices precisely what the criteria skip: brittle weld fracture, splice failure, and repair bills that run to eight figures.

For pre-Northridge steel moment frames, that gap is the binding constraint of 2026. FEMA P-58 converts component fragilities into expected annual loss, repair cost, and downtime — the questions an owner actually asks. An engineer who stops at acceptance criteria delivers a compliant building whose own fragilities imply losses nobody measured. The passing grade and the true liability turn out to be two different documents.

Two Calculators, One Frame

A single analytical model can feed both calculators, and that is exactly the trap. The same frame can clear the ASCE 41-23 nonlinear static procedure at both BSE-E hazard levels yet carry a FEMA P-58 loss distribution whose far tail sits above the lender's line, because the two methods do not share an outcome space: one counts rotations and demand ratios, the other counts dollars. Where the arithmetic splits explains why retrofit scope is sized by the first calculator but frozen only by the second.

Start with the calculator every engineer runs first. The nonlinear static procedure pushes the frame laterally under a fixed load pattern to a target displacement δt = C0·C1·C2·C3·Sa(Te)·Te²/g — the modified coefficient method, where Te is the effective fundamental period and the four coefficients convert elastic spectral demand into an inelastic displacement estimate. Each component is then checked against the Chapter 11 steel acceptance criteria: deformation-controlled actions, chiefly beam-hinge regions, by inelastic rotation; force-controlled actions by strength.

The pre-Northridge connection barely registers in that loop. ASCE 41 models the original welded joint as a force-controlled brittle action judged against lower-bound strength multiplied by the knowledge factor κ — 1.0 with comprehensive documentation, 1.25 usual, 1.33 limited. The calculation consumes one strength number and emits one demand ratio; it never produces a fracture probability. If the monotonic push keeps connection demand below that adjusted bound, the weld exits the report as compliant, with nothing in the output hinting how the same joint behaves under the cyclic records that fractured so many of them in 1994.

The Basic Performance Objective for Existing Buildings evaluates two deterministic points and nothing else: BSE-1E, the more frequent of the two earthquake hazard levels, demanding Life Safety, and BSE-2E, the rarer earthquake hazard level, demanding Collapse Prevention. One elastic spectrum per level, one verdict per component, and no distribution of outcomes anywhere in the objective.

FEMA P-58 runs a different machine. It starts from the site hazard curve, drives nonlinear response-history analyses across suites of ground motions to extract engineering demand parameters — peak and residual interstory drift, floor acceleration — then feeds those demands into assembly-level fragilities built from a median capacity and a dispersion β of roughly 0.4 to 0.7 for drift-sensitive components. Consequence functions convert each damage state into dollars per unit, and PACT samples the entire chain in Monte Carlo fashion across thousands of realizations.

At the output stage the units change completely. The static procedure returns a binary pass/fail per component at two discrete hazard levels; P-58 returns continuous distributions — expected annualized loss as a percentage of replacement value, probable maximum loss at defined return-period events, percentile repair-time and casualty curves. A demand ratio cannot be arithmetically reconciled with a PML percentage, which is why on a 2026 submittal the ASTM E2026 reviewer and the CMBS lender read the P-58 appendix first, and why a BPOE pass is a code-compliance statement, not a statement of financial survivability.

On pre-Northridge frames the divergence has a precise mechanical cause. The pushover's monotonic load pattern may never drive a brittle connection past its fracture point, so the post-fracture strength drop never registers and the frame reports green with the weld intact on paper. P-58's record-to-record sampling routinely pushes local interstory drift into the brittle-initiation zone of those same welds, prices the fracture through the fragility, and stacks the worst realizations into the PML tail. For sizing members the static procedure remains the right tool; for freezing scope, P-58 wins outright, because it alone prices the failure mode the pushover cannot see.

Run them in that order on every 2026 pre-Northridge scope: the pushover to size, P-58 to freeze.

The Record

After the January 17 Mw 6.7 Northridge earthquake, engineers who opened up welded steel moment-frame buildings documented cracked or fractured welded moment connections in numerous buildings — and the post-earthquake evaluation record shows nearly every one had been designed and inspected as code-compliant. Hold that against the reflex to certify a retrofit on a single calculator. The frames that failed brittlely had cleared the design review and the field inspection of their era, frequently on the exact joints that broke, which is why a BPOE pass reads in this record as a milestone, not a completion certificate.

The laboratory campaign explains why compliance failed to protect those frames. Across the full-scale connection tests compiled in the post-earthquake state-of-the-art report on connection performance, pre-Northridge details — complete-joint-penetration flange welds with backing bars left in place — typically fractured at plastic rotations below about 0.01 rad. The post-earthquake detailing provisions targeted 0.02–0.03 rad of capability for the ductile details written afterward. The installed base carries a systematic ductility deficit, not a scattering of bad welds, and the ledger below sets the numbers side by side.

SAC traced the mechanism to metallurgy, not modeling. According to the post-earthquake metallurgical findings, the low-toughness E70T-4 self-shielded flux-cored weld metal specified through the pre-Northridge era, acting with the notch effect of the left-in-place backing bar at the flange-weld root, created brittle initiation sites that activated at modest demand. Scope the implication precisely: no refinement of the analytical model — recalibrated hinges, a tighter loading protocol, a fresher hazard curve — changes weld-metal toughness or removes a physical notch. The deficit lives in the hardware, so the remedy menu is hardware too.

Put annual dollars on the deficit and the two-calculator problem stops being abstract. According to the FEMA P-58 background studies and the ATC-58 pilot applications, code-conforming modern frames produce expected annualized losses near or below roughly 1% of replacement value, while non-ductile legacy frames run at multiples of that. A pushover cannot surface the spread: ASCE 41 acceptance criteria return pass or fail at a demand parameter, never an annualized cost of ownership, so a frame can sit inside every acceptance limit while its loss engine runs well past the modern benchmark.

Then the transaction layer binds, and it binds harder than the code check. Under ASTM E2026-style seismic due diligence, CMBS lenders commonly screen deals at a PML expressed as a percentage of replacement value — the line this guide treats as binding. In the current CMBS cycle the PML estimate travels into the loan file, where it governs refinancing and sale: a building can satisfy every ASCE 41 acceptance criterion and still trip the screen, drawing lender-required scope, escrows, or repricing. The gate that decides whether the asset trades is not the one the engineer signs.

None of this disciplines scope unless loss reduction is physically attainable, and the record closes that question. The post-earthquake evaluation framework organizes the upgrade decisions, and the connection-modification schemes documented in AISC Design Guide 12 — haunches, cover plates, side plates — were shown in testing to restore connection ductility toward the target levels set for post-Northridge details. That tested recovery is the empirical basis for every P-58 loss improvement claimed later in this guide. Practical close: when the paired run at the governing rare event misses the lender gate, escalate to supplemental damping or DG12 connection replacement, and import the tested scheme rotations into the fragility inputs rather than accepting a generic modeled gain.

Gate Order

Order is the entire design decision here. The ASCE 41-23 nonlinear static procedure is a component-level pass/fail screen; FEMA P-58 is a financial model wearing structural clothing. Size with the first, freeze with the second — and never let the cheaper calculator close the job.

The row-two inversion is the mechanism worth internalizing. A pushover hands the pre-Northridge connection a single strength and a deformation capacity and climbs one curve; whether the weld actually fractures at the low or high end of its real distribution never enters the math. P-58 fragilities are built on that distribution, so a damper-only scheme that suppresses drift but leaves brittle splices in place can post a worse repair bill than a scheme that also replaces connections — even though the pushover scored it better. According to a paper titled "Advancing fracture fragility assessment of pre-Northridge welded column splices" hosted on ResearchGate, characterizing these splice fragilities remains an open research problem, which means the analyst — not the software vendor — owns the dispersion assumption, and it belongs in the submittal.

The role split in row three is not stylistic. Because ASCE 41-23 is the referenced compliance path under the 2024 IBC/IEBC, only the NSP result can be enforced at a permit counter; P-58 carries no such standing anywhere. So the framework cannot crown P-58 the sole gatekeeper, and it cannot let the NSP arbitrate money. Each calculator does the one job the other structurally cannot.

Row four explains most bad outcomes honestly. The NSP is days of work on a model that already exists; a defensible P-58 run is weeks of ground-motion curation, fragility curation, and specialist review. When fee proposals come back trimmed, Gate 2 is what got cut. Make scope freeze contractually contingent on the paired run, and require the fragility assumptions for every pre-Northridge component class as a signed deliverable. If your engineer proposes closing on the pushover alone, you have hired someone answering the wrong question.

What the Data Doesn't Tell You

Every number in a FEMA P-58 report for a pre-Northridge frame is a transplant — and transplants carry rejection risk the report never prints. The PML that decides whether retrofit scope freezes or escalates is not a measurement of your building. It is a chain of three transfers (repair costs, structural fragilities, ground-motion response) plus a recovery model that stops at the property line, and each link was forged on different steel, different crews, and different earthquakes than yours. None of this reverses the gate order described above; it changes how much margin you should demand before accepting a pass.

The cost functions were built for someone else's building. According to the FEMA P-58 consequence database, unit repair costs were assembled largely from modern West Coast construction assemblies — post-Northridge detailing, contemporary labor and means-and-methods. Drop them onto a pre-Northridge welded moment frame and you import a bias nobody quantifies: repairing cracked pre-Northridge welds can plausibly price above or below the tabulated values depending on access, panel-zone condition, and how much of the original assembly survives. You get a point estimate with no confidence interval, because the underlying cost data carries none. Spot-check several component costs against a local fabricator's actual repair pricing before the total drives scope.

The fragilities circle back to the same specimens. P-58's structural fragilities were calibrated primarily on modern detailing, so a pre-Northridge frame forces custom medians — and the natural source is the SAC connection test program, the same small specimen set the retrofit logic already leans on. The loss estimate's structural core therefore double-counts one limited sample, and nothing in the PACT output flags the circularity. Ask directly how many unique specimens back each custom median; if the answer is a handful, treat the median as a hypothesis, not a finding.

The pushover has acknowledged blind spots. The nonlinear static procedure applies a fixed lateral-load pattern that omits higher-mode and torsional demands. On a plan-irregular tower, the pass/fail verdict can hinge on pattern choice alone — ASCE 41-23 itself restricts or conditions the NSP for such buildings and points toward dynamic procedures. Read that restriction as a scope instruction: where the code procedure is conditional, the paired P-58 run is not optional polish; it is the only check that sees the torsional response the pattern cannot.

Suite selection moves the answer more than the retrofit does. P-58 outputs respond materially to ground-motion suite selection and scaling. Two equally defensible suites — same hazard level, same site conditions — can shift PML by tens of percent in relative terms, enough to cross the lender's replacement-value gate in either direction without touching the building. Require disclosure of record count, scaling method, and pulse screening, then rerun with a second independently selected suite. If the two runs straddle the gate, the honest reading is not "pass"; it is unresolved — and unresolved scopes escalate.

Downtime percentiles stop at the property line. P-58 repair-time outputs assume crews mobilize and permits issue promptly, but post-event functional recovery is frequently governed by utility restoration and inspection backlogs outside the building boundary — a distinction the federal functional-recovery guidance develops at length. Because expected annual loss folds downtime into the financial result, systematically optimistic repair times flatter the very metric the gate watches. Overlay a utility-restoration scenario on any downtime-driven call.

Finally, the report is discoverable. No code requires a P-58 analysis; it is voluntary. A voluntary report showing high losses can become discoverable evidence against the owner or engineer — the counterweight that keeps P-58 from serving as an unquestionable final word. Route the engagement through counsel and frame it as a decision document, not a certification.

None of these caveats licenses closing a job on a BPOE pass alone — the paired run remains the freeze condition. What they set is margin policy for the current retrofit cycle: demand suite disclosure, specimen counts, and a second-suite rerun; if PML sits anywhere near the line once those checks land, treat the uncertainty itself as the escalation trigger and add damping capacity per ASCE 41-23 Chapter 14 rather than argue the error bars down.

A Pre-Northridge Seven-Story LA Office

Gate 1 runs in Perform-3D as a nonlinear static procedure with first-mode and uniform load patterns. It returns Te = 1.6 s against a BSE-2E demand of Sa = 0.42 g, giving a target displacement of δt = 0.85 × 1.05 × 1.0 × 1.0 × 0.42 g × (1.6 s)² ≈ 0.96 m. At that displacement the model peaks at 2.2% interstory drift at floor 4 — a mid-height concentration typical of pre-Northridge frames — and the as-built weld demand-capacity ratio reads 1.3. A haunch retrofit detailed per AISC Design Guide 12 pulls the DCR to 0.9. Every check ASCE 41-23 requires at BPOE now passes. On paper, the job is closed.

Close the file with a deliberate split. The ASCE 41-23 compliance sheets go to the jurisdiction as the legally required artifact; the lender receives a P-58 loss summary in standard PML-report format for the ASTM E2026 review. The transmittal letter states the uncomfortable part plainly: scope escalated because PML moved from above the lender's line to below it — not because any code criterion changed. Between the first PACT run and the second, the building became stronger than the code demanded, because the party funding the retrofit prices tail losses the code never counts.

A BPOE pass is a structural verdict, not a financial one, and 2026 retrofit contracts that treat it as both are the ones that reopen. The belief that "the pushover meets BPOE, so the retrofit is done" fails mechanically: the ASCE 41-23 nonlinear static procedure returns component acceptability at a single hazard pair, while the PML a CMBS lender or ASTM E2026 reviewer enforces is a loss distribution integrated over record-to-record variability. Code compliance and financial survivability are different gates, and the second one decides scope. Five rules turn that distinction into award conditions.

Rule 1 — never freeze scope on a passing NSP alone. Make the companion FEMA P-58 run a condition precedent to contract award, with two gates: PML at the governing rare event held within the lender's replacement-value screen, and expected annual loss at or below 1%. Treat any pushover-only closure as an incomplete deliverable, and pay

```

Frequently Asked Questions

What knowledge factor does ASCE 41 apply when checking the pre-Northridge welded joint as a force-controlled brittle action?

The lower-bound strength is multiplied by a knowledge factor κ of 1.0 with comprehensive documentation, 1.25 for usual knowledge, and 1.33 for limited knowledge.

Which two hazard levels and performance objectives does the Basic Performance Objective for Existing Buildings actually evaluate?

It evaluates just two deterministic points — BSE-1E, the more frequent earthquake hazard level demanding Life Safety, and BSE-2E, the rarer level demanding Collapse Prevention — with one elastic spectrum per level and no distribution of outcomes anywhere in the objective.

Why can a pre-Northridge frame report green under the pushover even though its welds are brittle?

The nonlinear static procedure's monotonic load pattern may never drive a brittle connection past its fracture point, so the post-fracture strength drop never registers and the frame reports green with the weld intact on paper.

What dispersion values does FEMA P-58 use for the assembly-level fragilities of drift-sensitive components?

P-58 builds drift-sensitive component fragilities from a median capacity and a dispersion β of roughly 0.4 to 0.7.

If my connections pass their acceptance criteria, is there any other pre-Northridge failure mode left unpriced?

Yes — papers such as 'Evaluation and Remediation of Pre-Northridge Steel Moment Frame Column Splices' treat splice fracture as its own fragility, meaning a frame can satisfy connection-level acceptance criteria while the fracture regime behind the 91% lab record persists unpriced at the splice.

On a 2026 submittal, who reads the FEMA P-58 appendix before anything else?

The ASTM E2026 reviewer and the CMBS lender read the P-58 appendix first, because a demand ratio cannot be arithmetically reconciled with a PML percentage and a BPOE pass is a code-compliance statement rather than a statement of financial survivability.

Quick answers

Also worth reading: ASCE's 2024 Infrastructure Report Card Data-Driven Analysis of America's Aging Bridge Network: ASCE's 2024 Infrastructure Report Card · ASCE Discontinues Iconic Bridge Calendar Series After 2023 A Look at Its Engineering Legacy: ASCE Discontinues Iconic Bridge Calendar · Changes in ASCE 7-16 Structural Design Load Standards Revisited: Changes in ASCE 7-16 Structural