Four Status Values, Six Transition Modes: A Survival Taxonomy for the Battery Registry

Count the canceled US battery plants and you are counting a schema.

Start with what the instruments can represent. Clean Economy Tracker offers Planned, Under Construction, Operational, Canceled; finalized bankruptcies and anything described as indefinite route into Canceled, and canceled projects then drop off the map entirely. Clean Investment Monitor works from announced, under-construction, completed, and canceled, and handles pauses, scale-backs, restructurings, and pivots in narrative text rather than as states. NLR's lithium-ion battery supply chain database runs Planned, Under Construction, Pre-commercial/Startup, Commercial, with no field for pause, transfer, conversion, or liquidation at all.

Four buckets. Feed them a liquidation with a sixty-percent shell, an indefinite construction pause on a completed building, a change of control on a line that never stopped shipping, and a chemistry swap on an operating plant, and the four events come out the other side either identical or invisible. That result characterizes the schema. It carries no information about the industry.

The events separate cleanly once you stop asking a four-state field to hold them. What follows is six transition types visible in the registry, each anchored to a named facility, each scored for what carried and what reset. All statuses are as of July 31, 2026. The question at every row is the same one a supply model has to answer: for each asset class the model depends on, does the transition preserve it, destroy it, or leave it in a state that no public source resolves?

The third outcome is recorded here as a value in its own right, not as a blank.

The survival matrix

Read the columns. The rows are the labels the trade press argues about; the columns are where a supply model actually lives.

Legend.

• Carried: public evidence establishes continuity.
• Partial: continuity for part of the asset class, discontinuity or silence for the rest.
• Incomplete: the asset class was never fully established, work continues, so neither continuity nor reset applies.
• Reset: public evidence establishes discontinuity.
• Renewed: re-established with the successor on new terms rather than carried on old ones.
• Undetermined: public sources exist and do not resolve it.
• Not addressed: no reviewed public document raises the question.
• Not examined: outside this review's scope, an evidence gap on my side rather than on the record's.
• N/A: the asset class never existed here, so survival does not apply.

Seven rows, six modes. 5a and 5b are one mode observed at two distances along the same reset: both hold site, structures, and ownership constant while resetting product-specific tooling, process validation, and the whole acceptance chain. Kokomo is simply earlier on that path.

The modes do not partition the space, and they are not stable in time. Mode 2 is the precondition for Mode 1 rather than an alternative to it, and none of the trackers in this set can record the transit between them; a paused project becomes a liquidated one the moment a filing lands, with nothing moving in the status field beforehand. Mode 6 is a compound of transfer and application change by construction, which is the tell that these behave as dimensions that co-occur rather than as mutually exclusive categories. A facility can occupy two at once and migrate between them with no public event to mark the migration.

Mode 1. Liquidation with a partially built asset

Scale. Announced nameplate is not restated in the reviewed bankruptcy record. The usable scale proxy is a 96 MW power allocation. Demonstrated output: none, ever.

Apex 1, the Ascend Elements cathode precursor plant in Hopkinsville, is the cleanest liquidation on the registry. Construction stopped near sixty percent. The bankruptcy auction moved the property to a Turner-Kokosing joint venture, and local reporting since has described the new owner looking for another user or purchaser. The evidentiary walkthrough belongs to this issue's Apex 1 audit. What the taxonomy needs from Apex 1 is the permit column.

Kentucky's environmental database still carries air, wastewater, stormwater, and hazardous-waste authorizations as active, under Ascend. The DOE environmental record puts air permit expiry at January 30, 2028 and the discharge permit at March 31, 2028. Neither is shown as transferred. Kentucky requires notice and an administrative permit revision after a change of ownership, and a discharge-permit transfer takes effect only when the Division of Water acknowledges it. So the honest entry has three parts: the authorizations exist, they attach to an entity that no longer owns the site, and the transfer mechanism is defined but not documented as having started.

Power is thinner. The 96 MW allocation from the local cooperative and TVA surfaced in the bankruptcy financing as a best-efforts covenant, not a confirmed allocation. Interconnection survival is not publicly determinable. Without a confirmed interconnection, a sixty-percent shell is a warehouse with unusually generous ceiling height.

Mode 2. Pause without a restart date

Scale. 27 GWh nameplate with a 2027 production start, per the partner's still-posted project page. Demonstrated output: zero.

Synergy Cells, the GM and Samsung SDI joint venture in New Carlisle, paused in May 2026 with the exterior substantially complete. No ownership change has been reported. Nothing in the partner, state, utility-regulator, or county records reviewed establishes a restart date, a revised production start, a capacity revision, or an equipment cancellation. Building-variance filings that surface in search carry October 2025 application dates, ahead of the pause, and say nothing about activity since. The project website and corporate page still display the pre-pause plan, which is a fact about website maintenance rather than about schedule.

Two columns need isolating.

Permits: the public air record is a draft permit from September 2025, with the notice stating that a later decision would establish issuance or denial. No final decision was located. No facility-specific discharge, stormwater, or pretreatment authorization was located either, and an April 2026 utility-regulator order describes a proposed water-service arrangement rather than an executed plant agreement. Permit survival is undeterminable here for a specific reason: permit establishment was never publicly completed.

Workforce: the general contractor acknowledged laying off some project workers in fall 2025 without stating a number, and Indiana's WARN registry contains no notice for the project. That absence carries far less information than it looks like it does. Federal notice obligations attach at a covered-event threshold in the range of fifty affected workers, and reductions below that line leave no registry trace of any kind. The instrument has a detection floor, and both reductions documented in this set sit under it: nineteen at Chattanooga, an unstated count at New Carlisle. Reading an empty WARN registry as evidence of a stable workforce is reading a null result as a measurement.

Mode 3. Milestone delay with the balance sheet intact

Scale. The reviewed filings attach no full-plant nameplate. The only capacity figure is a first 11,000 tonnes-per-annum tranche, which the certified $103 million manufacturing tax-credit allocation requires to be placed in service by April 7, 2028. Demonstrated output to date is sample material only.

NOVONIX moved mass production at its Chattanooga synthetic graphite plant into the second half of 2027. Facility, Panasonic supply relationship, financing, and tax-credit allocation all remain in place. Equipment installation and commissioning remain incomplete. A June 2026 C-sample met the company's internal specifications; Panasonic's validation is still pending.

This mode gets misread in both directions. The commercial relationship genuinely survived, and the delay does not imply a lost customer. What has not been established is deliverable output under that relationship. The qualification state is neither carried nor reset. It is unfinished, and unfinished is a category, not a softer grade of carried.

The workforce signal moved independently of the disclosed status. Nineteen positions were cut in January 2025, taking reported local employment from 122 to 103, and a June 2026 state grant-clawback demand credited the company with 85 jobs against a 290-job commitment. Tennessee's WARN archive contains no entry, for the structural reason already given: a nineteen-person reduction is invisible to a fifty-person instrument.

Mode 4. Control transfer of a line that never stopped

Scale. Nameplate is not restated in the reviewed transaction materials. Demonstrated output is continuous production at undisclosed volume.

Fixx took control of the Smyrna cell plant on March 31, 2026. The plant had been producing since 2025 and, per the buyer's only publicly identified offtaker on its Q2 FY2026 call, continued producing through the transaction. Physically this is the highest-continuity transition in the set. It is also the one where the physical column does the most work covering for silence elsewhere. Reviewed disclosures do not inventory retained equipment, manufacturing execution software, warranties, or licensed process rights, and plant headcount and workforce-transfer terms were not disclosed. Facility permit transfers were not publicly verified in the transaction materials.

The column where this mode actually resolves is qualification, handled below.

Modes 5a and 5b. Conversion, executed and contracted

Scale. Spring Hill's original announced plant investment was $2.3 billion; the converted share of nameplate is not disclosed, and demonstrated ESS-LFP output is a July production start at undisclosed volume. Kokomo has roughly 19 GWh under conversion by trade account, with zero demonstrated converted output.

Spring Hill converted part of its operating capacity from EV cells to lithium iron phosphate cells for stationary storage. Disclosed retooling ran $70 million, completed in under five months, with production beginning in July. Furloughed staff returned.

Kokomo is the contracted variant, and its physical detail is inferred entirely from trade reporting rather than joint-venture disclosure. The Electronic Times describes three of four Plant 1 lines under conversion: one roughly 7 GWh nickel-based line and two LFP lines totaling roughly 12 GWh. Samsung has not confirmed that allocation. What the JV partner has disclosed is both ends of the commercial chain, a four-year storage supply agreement downstream and an LFP cathode agreement upstream from 2027. Output remains prospective.

Conversion is the mode that most reliably escapes the cancellation count while pulling capacity out of the EV-addressable pool. This publication's earlier denominator analysis identified at least 20 GWh directly restated to storage, with as much as 184 GWh carrying a migration signal, none of it appearing in any cancellation total. Credit eligibility and compliance consequences belong to the companion analysis. The concern here is the physical and operational layer, and that layer is more disturbed than the retooling figure implies.

Why installed equipment is not one asset class

"The installed equipment survives" is the most abused survival claim in the taxonomy, because it treats a cell line as a single object. A cell line is a sequence of process stages, and fungibility falls as you move downstream. The best public process-level treatment, an RWTH Aachen/PEM flexibility study, splits the line into electrode production, cell assembly, and finalization and finds exactly that gradient.

• Mixing, coating, calendering. Highest fungibility. Nickel-based and LFP chemistries generally share the process sequence and roll-to-roll hardware inside a line's operating envelope. What changes: slurry formulation and rheology, coating weight, line speed, roll pressure, density targets. Slot dies may need replacement for different rheology or layer thickness.
• Electrode and vacuum drying. Ovens and vacuum systems may remain usable where web width, temperature range, and throughput are compatible. Drying curves, moisture specs, residence time, and cleaning validation reset.
• Slitting, cutting, notching (sizing the electrode and forming its tab geometry). Base machines may survive; tooling, cutting geometry, edge-quality settings, and inspection recipes change with cell design.
• Stacking or winding, tab welding, sealing. The study identifies assembly as the least fungible stage. Even a tab-position change inside the same pouch format can force new cutting, welding, sealing, alignment, and handling tools.
• Filling, degassing, resealing. Pouch finishing architecture may carry between pouch products. Fill quantity, wetting time, vacuum profile, seal geometry, and inspection recipes all require renewed process validation.
• Formation and aging. Chambers and trays can be modular. Protocols, current-voltage profiles, aging duration, grading criteria, tray geometry, and electrical contacting are normally matched closely to the specific product.
• End-of-line inspection and intralogistics. Conveyance and automated guided vehicles are among the most portable assets on the floor. Vision, dimensional, electrical, leak-test, and lot-release criteria have to be rewritten and revalidated.

No facility disclosure in this set resolves to that granularity. Spring Hill's public record says process retooling and attaches a number. It does not enumerate reused mixers, coaters, ovens, calenders, assembly stations, or formation channels. The number also invites an arithmetic error: $70 million against a $2.3 billion originally announced investment is not a reuse ratio. The denominator covers the whole plant, and neither disclosure states how much capacity the $70 million converts or which original cost categories it displaces.

The qualification column, where the count is zero

Qualification means one thing here. A specific customer has approved a specific product from a specific line, on the basis of that customer's own testing against its own acceptance criteria. It is not an attribute of a building. This publication's proof ladder treated qualification as customer- and product-specific, and the transition record vindicates that treatment.

Three states are available in the public record: requalification stated, prior approval expressly carried over, or the matter not addressed. Across every transition reviewed, the count in the first two states is zero.

Take Smyrna. The buyer's only publicly identified offtaker signed a new multiyear supply agreement after the change of control. Neither that disclosure, nor the pre-transfer supply agreement materials, nor the offtaker's most recent 10-Q or quarterly call states that a prior approval was reissued or carried across the transaction. Contractual continuity is real and documented; it is standing in for qualification continuity, which is stated nowhere. At Spring Hill the record establishes production and an intended intercompany route to the affiliated integrator, independently confirmed as production and not as acceptance. No delivery, repeat-run output, or external end-customer allocation was located, consistent with the conversion registry's finding that no reviewed conversion carried public shipment or acceptance evidence.

None of this establishes that qualification failed to survive. It establishes that qualification survival is unobservable in the public record. For a supply model the two converge, because an unobservable cannot be underwritten.

Two facilities that fit no mode

Ford's Marshall, Michigan plant has produced full production-intent cells through final inspection, with the company placing those cells into quality testing and commercial shipments later in 2026. That is a position on the equipment-to-validation-to-shipment sequence. It is not a cancellation, a transfer, or a conversion. Syrah's Vidalia operation reported 150 tonnes of qualification material year-to-date and an $8 million facility-linked production credit refund, while recurring commercial sales and final customer qualification remained prospective; the evidence supports a creditable production transaction and not yet recurring accepted supply.

Both sit between supply states rather than between transition outcomes. Forcing either into a mode would destroy the taxonomy's only useful property, which is that its categories track distinct resets.

Ranking the modes by what actually resets

Rank these seven facilities by capital destroyed and you get one sequence. Rank them by supply-relevant gates reset and the order changes.

Physical assets carry well. Site control survived all six modes, liquidation included, and structures survived everywhere. Organizational capability carries poorly and asymmetrically, and the public instruments that would detect its loss have detection floors well above the reductions actually occurring. Customer qualification carries worst, in the narrow and literal sense that not one transition in this set produced a public statement that it carried.

Which inverts the intuitive cost ordering. A two-year construction pause on an uncontested site with a draft permit destroys schedule and construction workforce and resets no qualification, because none existed to reset. Change of control on a running line preserves every physical asset, keeps the plant shipping through the closing date, and leaves the single column a buyer's underwriting depends on in a state no public document addresses. A chemistry swap on an operating line costs a disclosed retooling spend at roughly three percent of the plant's original announced investment, and resets formation protocols, assembly tooling, inspection recipes, and the entire customer-acceptance chain for a product with no acceptance history at that site.

Do not model from the label. The label is one of four values, and a facility can occupy two modes at once. Model column by column, and carry undetermined through the model as its own value. A four-state status field will otherwise resolve it toward continuity on your behalf, and nothing in the output will tell you that it did.