Revision log

Revision log

1. 2026-08-26Commercial

   The PECRON F3000LFP can now be ordered from its own exact page, so its exact-SKU link is switched on — and it is not called in stock

   The authenticated PECRON affiliate account for Before the Battery is affiliate ID 1260, and the dashboard states a contract rate of 5%. Earlier entries named the identifier 1275; that identifier is withdrawn and every PECRON destination now carries the parameter #ref=1260. The exact-SKU destination for the F3000LFP was generated by the official dashboard link generator and verified on 26 August 2026 to open the exact F3000LFP product and its exact variant, not another unit and not a family page. That page was observed at USD 899, the exact variant can be added to the cart and a checkout opens for it, and the page's own stock wording states that arrival at the warehouse is expected next week. The row is therefore recorded in a distinct, explicit state — orderable with stock inbound — and it is never described as in stock, shipping now, or immediate delivery. The exact-SKU button is offered only when the engine itself selects this exact unit, and it resolves through the same internal router, first-touch attribution and click reference used by every other merchant destination; the entry page for this unit still holds no merchant URL of its own. One reporting limitation is recorded rather than hidden: a fresh external click was counted in the affiliate dashboard as one visit for #ref=1260, but the dashboard's landing-page column omitted the product path and showed only the host with the variant and reference, even though direct navigation was verified to land on the exact product page. That is a merchant-side reporting truncation, not a failed link. The affiliate profile has no payment method selected and the displayed minimum payout is USD 200; payout readiness affects nothing technical, and no commission is claimed as proved. The 5% rate is recorded as evidence of the relationship only and is never an input to qualification, eligibility, ranking or order. Nothing technical moved: no threshold, rule, ruleset version, evidence grade, specification, reason code, fixture, architecture, ranking rule or other catalogue row was changed, and no other brand or cohort was touched.

2. 2026-08-25Method

   The PECRON range is recorded from its official manuals, and a 30-amp RV outlet is now credited only up to what that outlet can actually carry

   The PECRON F3000LFP row was re-closed from the official exact-SKU US manual rather than from a marketing page. Its wall recharge input is credited at 1,800 watts maximum on a 120-volt, 15-amp household circuit, corroborated by an independent test of the same model; the larger figure published elsewhere is a combined mains-plus-solar input claim and is no longer presented as an official contradiction. The manual-stated weight of 28.7 kilograms (63.3 pounds), the 8 to 20 millisecond switchover and the 3,500-cycle rating are recorded as sourced facts, and the 4,500-watt, five-second output envelope stays labelled a timed manufacturer claim that never enters a calculation. Two deeper exact-SKU rows, the E3800LFP and the F5000LFP, were added to the same normalised RV catalogue and the same selection system, bringing the recorded RV range to nineteen exact SKUs; they receive no landing promise card, no product-aware door and no promotional content in this entry. All three carry a native 30-amp RV (TT-30) receptacle drawn in their own manuals, and the engine now compares an RV load against that connector's own continuous ceiling of 3,600 watts instead of the inverter's headline rating — so a unit rated above that figure is not credited through a 30-amp cord it cannot legally carry. None of the three is purchasable in the United States on this date, so none renders a merchant action; the F3000LFP's authenticated exact-SKU affiliate destination is retained as prepared evidence only under an explicitly published gated state, and the router still refuses to resolve it. Nothing else moved: no threshold, safety factor, reason code, ruleset version, fixture, evidence policy, ranking rule, affiliate identifier or other catalogue row was changed, and every canonical fixture was replayed against the live engine. One legitimate change of outcome is disclosed: where a mission is closed by both a buyable and a gated unit, the buyable exact SKU is selected, which is the pre-existing rule and not a new preference.

3. 2026-08-24Method

   Phase B — Every final result is now read in the same order, and each named unit says whether it passes, is ruled out, or is still missing a fact

   The final result of both checks is now presented in one fixed reading order: the five-second summary, the best-fit architecture, the products that pass, the products ruled out or not yet qualified, the exact gate reason with the named missing evidence and the action that would close it, the limits and assumptions, and last of all the merchant relation. A single section link, “See the products that close this setup”, jumps from the architecture to the units that pass, and it appears only when at least one exact unit actually passes. Nothing in the decision moved: the canonical engines, thresholds, safety factors, evidence policy, reason codes, eligibility gates, selection order and ranking rules are byte-for-byte the same, and this change reorders how an already-computed result is read, nothing else. A named candidate carrying a missing fact is stated as not a recommendation and receives no purchase action of any kind; its exact missing fields and closing action stay published next to it. Affiliate and official non-affiliate destinations remain labelled explicitly, and Anker remains non-affiliate with official destinations only. Catalogue depth is declared conservatively: the intended US coverage targets for both checks are resolved against the exact-SKU evidence rows this project already holds, and any target without such a row stays “evidence in progress” with its exact missing fields, a non-adverse public reason and the single action that would close it — no specification, price, stock, runtime, rank or comparison win is invented, no existing row is modified or hidden, the active ALLPOWERS rail is untouched and is never forced into the promise list, and manufacturer runtime claims such as “up to 32 hours” remain labelled claims that never enter a calculation. Measurement gains five anonymous, allowlisted, categorical events under the same cohort BTB-EARLY-VALUE-PRODUCT-PROMISE-2026-08-24 — a product card impression, a product confirmed, a product ruled out, an exact product unlocked and a merchant call-to-action impression — each emitted at most once per session and per named unit, with no free text and no identifier. A merchant event is emitted only for a real purchasable exact SKU behind the existing exact-SKU router: blocked, unavailable, informational and professional-referral actions never produce one. The exactly-once start, completion, verdict and decision-case events, the first-party merchant click, the consented conversion, QA exclusion and first-touch attribution are all unchanged. Rolling this entry back would restore the previous section order and remove the five new event names, and would not alter a single recorded decision.

4. 2026-08-24Method

   Phase A — Both checks now give a real answer before the last questions, and name the units they can confirm or rule out

   Two changes were made to how the free checks present themselves, and none of them touches how they decide. First, each landing now carries a short block titled “Products this check may confirm — or rule out”, naming the real units the diagnostic is able to test — for the refrigerator, BLUETTI FridgePower as a dedicated automatic refrigerator backup, the BLUETTI Elite 200 V2 as a versatile two-kilowatt-hour portable class, the Anker SOLIX C2000 Gen 2 in the same class as a non-affiliate candidate, and an extended-outage role that stays marked “catalog evidence in progress” until an exact normalised evidence row exists; for the RV, the BLUETTI AC200L, Elite 300, Apex 300 and RV5 in their four distinct architecture roles. These are candidates, not a ranking and not a recommendation: the block states in place that your inputs decide the result, that affiliate status never decides technical eligibility, and that the check can conclude that none of them closes your setup. The illustrations are technical outlines drawn in the site's own tokens, not photographs, and no price, stock, runtime or comparison win is asserted anywhere. Second, each check now delivers an early answer as soon as the core inputs are in — the planning capacity band, whether energy sizing is complete and on what basis, whether compressor or startup compatibility is proven or unproven, and the probable architecture class — followed immediately by the number of checks that still remain before an exact unit can be named, and by what each of those remaining checks is allowed to change. That early card is not a completion: it emits its own event and never a scenario completion, a verdict or a decision case, so every published funnel figure keeps its exact prior meaning. Third, five product-aware entry pages were published, two for the RV check and three for the refrigerator check, each stating verbatim that you started with this product and that the check may confirm it, eliminate it, or identify a more appropriate setup. They run the same canonical diagnostic and the same engine; they preselect a hypothesis and never a verdict. Nothing in the decision moved: no rule, threshold, safety factor, reason code, ruleset version, fixture, catalogue row, evidence grade, eligibility order, ranking rule, affiliate identifier or merchant destination was changed. Anker remains non-affiliate and receives official destinations only. Measurement gained a set of anonymous, allowlisted events under the cohort BTB-EARLY-VALUE-PRODUCT-PROMISE-2026-08-24, all carrying the existing first-touch classification, with impressions and lifecycle events emitted at most once per session and QA sessions still excluded. No free text is collected anywhere, including in the two optional friction questions. Acceptance was a full refrigerator and RV run on each energy path reaching the same verdict as before for identical inputs, exactly one start, completion, verdict and decision case per successful session, no horizontal overflow at 390 by 844 or 1440 by 900, and the full test suite green. Rolling this batch back would remove the new pages and the early card and would not alter a single recorded decision, because the cohort boundary is a label on events and the new surfaces read the engine rather than change it.

5. 2026-08-17Method

   The RV check leads with what it answers, and its funnel starts a new cohort

   The RV acquisition layer at /run/rv-weekend was rebuilt around the result rather than the method. The page now opens with one question — will your RV power setup make it through the whole trip — states the promise plainly (your nights, your must-run loads, your recharge options, and in about two minutes the minimum usable capacity, the daily recharge gap and a startup-power check), and says before the first click that estimates and an honest “I don’t know” are accepted, with no email and no brand ranking. A code-native illustration, drawn entirely in SVG from the site's own tokens with no photography, no brand mark and no invented product, shows the job itself: a rig at camp without shore power, the loads that must survive the night, the stored energy, the solar and alternator paths that may or may not refill it, and the card that comes back. The four outputs are listed before any interaction. Two actions are offered: begin the normal first question, or load the existing signed two-night example, which behaves exactly as before — editable everywhere, submitting nothing, landing on the load review step. The first question now sits immediately after the hero; the three-gate engine diagram is preserved in full, unchanged, but moved below the form behind “See how the decision works” so no one has to read the method before starting. The pillar page /solar-generator-for-rv keeps its H1, canonical, Article, FAQ and Breadcrumb structured data and every published equation, and gains the same illustration, the same four-output card and a direct link into the check above the arithmetic. Nothing about the decision moved: no rule, threshold, ruleset version, fixture, catalogue row, evidence grade, eligibility test, ranking rule or merchant destination was changed, and the refrigerator funnel is untouched. Measurement changed in two ways. RV events emitted from 17 August 2026 carry the release marker RV-UX-ATTRACTION-2026-08-17, read separately from RV-M0B-2026-08-15-OBS, which remains the prior boundary; no historical row is relabelled or merged. And a defect was fixed: a QA session opened with the internal test flag could be recorded as real market traffic once a client-side navigation replaced the URL, or when browser storage refused to keep the flag. The flag is now observed at the entry URL and held for the whole page session, so QA clicks, page views and step views stay out of the external aggregates. Acceptance for this entry was a full RV run reaching the same verdict for identical inputs, exactly one start, completion, verdict and decision case per successful session, and no change to any existing test. Rolling the interface back would not alter a single recorded decision, because the cohort boundary is a label on events, not a change to the method.

6. 2026-08-16Method

   The refrigerator check is rebuilt around three ways to answer, and its funnel starts a new cohort

   The refrigerator decision at /run/refrigerator now opens with one promise — find out what battery capacity your refrigerator needs during an outage, in about sixty seconds, with estimates accepted — and lists before the first click what the visitor receives: a minimum usable capacity, an estimated runtime, a startup-power check, the limits and unknowns, and the compatible architectures. The annual-kWh figure is no longer a requirement. The energy question now offers three explicit paths: the annual kWh from the yellow EnergyGuide label, a declared running-watts figure, or an honest “I don’t know” that estimates from the kind of refrigerator you have. The running-watts path converts a power figure into a daily-energy envelope through a published 30–50 % compressor duty-cycle band, and both estimated paths are labelled as an assumption to confirm, never as a measurement; the verdict is still sized on the upper bound of the range. A simple accessible schematic shows where the yearly kWh figure sits on a US EnergyGuide label, with no product photography, and says plainly that you may continue without the label. Nothing about the engine moved: no threshold, rule, ruleset version, catalogue row, eligibility test, ranking rule or merchant destination was changed, and no verdict is auto-rendered — the evidence required for a verdict is exactly what it was. What also changed is measurement. Refrigerator funnel events emitted from 16 August 2026 carry the release marker FRIDGE-UX-PIVOT-2026-08-16, the start, completion, verdict and decision-case events are emitted exactly once per successful session, and first-touch UTM and gclid values are now inherited across a session so a Google visit can no longer be reported as unclassified at the merchant step. Historical events keep their own cohort and are never relabelled or merged; the internal funnel view reads one cohort at a time. Sessions marked with the internal QA flag stay out of external aggregates. Rolling back the interface would not alter any recorded decision, because the cohort boundary is a label on events, not a change to the method.

7. 2026-08-15Method

   A second RV decision case is published, and the RV cluster is made navigable

   Decision case RV-02 is published at /decision-cases/rv-02-air-conditioning-no-shore-power: a travel trailer on a TT-30 pedestal with a 13,500 BTU roof air conditioner, six hours a day, no shore power and no soft start. Like RV-01 it is a frozen fixture replayed against the live engine and the live catalogue at page render, not a hand-written narrative, and it deliberately ends without a product: the fixture does declare a 2,900 W nameplate starting figure, but the peak stays unproven because there is no soft start and product-specific startup compatibility is not signed evidence, the trip needs a generator or hybrid architecture, and the pedestal ceiling is in play, so the method escalates and names the measurements to take instead. A /decision-cases index now lists both cases and states plainly that a case which refuses is as much evidence as one that concludes. The pillar page at /solar-generator-for-rv carries a visible author and review date, and the RV pillar, both cases, the two RV guides, the guides hub, the homepage and the footer are now cross-linked so the cluster can be traversed without search. Structured data was added for the two cases and the index. No decision rule, threshold, evidence row, catalogue status, availability fact, ranking or affiliate route was changed by this entry.

8. 2026-08-13Commercial

   The PECRON E2400LFP is buyable again, so only its exact-SKU link is switched back on

   The official PECRON US exact-SKU page for the E2400LFP was re-checked on 13 August 2026 and publishes an in-stock purchase channel again, replacing the "Notify Me When Available" state recorded the day before. That exact SKU is therefore recorded as purchasable, and the authenticated exact-SKU PECRON link (Affiliatly affiliate ID 1260, 5% stated on the authenticated programme page, parameter #ref=1260) becomes active behind the stable internal route /go/pecron/e2400lfp, which resolves to that row's own official product page — never a homepage and never a substitute SKU. The E1000LFP was re-checked the same day and is still out of stock, so it keeps its qualification and keeps no button; its prepared link stays evidence only. Nothing technical moved: qualification, evidence grade, planning capacity, output bounds, eligibility and ranking are computed before any destination is resolved, and availability is a dated commercial fact, never technical proof. Purchasable coverage is now BLUETTI 3 of 3, ALLPOWERS 1 of 3 and PECRON 1 of 2. No click has yet been matched end to end in any network report, so attribution stays open and paid search stays on HOLD.

9. 2026-08-13Correction

   A small controlled US Search test is running: the earlier statement that no campaign exists is withdrawn

   An earlier entry said the first US acquisition cell was prepared only, with no advertising campaign and no spend. That is no longer true and is corrected here. A single controlled Google Search test for the refrigerator decision, experiment BTB-US-REF-D0-01, was launched on 13 August 2026 with a hard total cap of EUR 44 running to 22 August 2026, in the United States and in English, sending traffic to the existing refrigerator diagnostic. At the 13 August 2026 cut the campaign is active but the ad is still under review by Google, so nothing has served: zero impressions, zero clicks and zero spend were observed. This is a research test to learn whether the question is searched and understood; it is not a market-validation claim, not a revenue claim, and not an authorisation to scale — scaling and phase 2 remain on HOLD. The Google Ads base tag is deployed for measurement only, with all advertising and analytics consent signals denied by default until a visitor explicitly allows them, and no conversion event is installed: an automatically generated page-view conversion in the advertising account is deliberately recorded as not installed and not primary, because a page view is not a result. No decision threshold, verdict, ranking, evidence row, availability fact or affiliate route was changed by this entry. Alongside it, the published version identifiers were synchronised with reality: the rule set is dated 12 August 2026 and the catalogue 13 August 2026, so every Decision Case now carries the versions it was actually produced under.

10. 2026-08-13Commercial

    The ALLPOWERS R2500 V2 is purchasable again in the US, and only its button comes back

    The official ALLPOWERS US exact-SKU page for the R2500 V2 was re-checked on 13 August 2026 and now shows "Add to Cart" and "Buy It Now", with local delivery quoted at 3–5 days, a 5-year warranty and 30-day money back; a sale price of USD 659.00 was displayed that day, recorded as a dated observation and never as a guaranteed price. That exact SKU is therefore recorded as purchasable again, and its certified exact-SKU Awin rail (advertiser 40342, publisher 3023093, tracking parameter clickref) becomes active behind the stable internal route /go/allpowers/r2500-v2, which resolves to that row's own official product page — never a homepage, never a substitute SKU, and never with a stored or hardcoded transient click token. Nothing technical moved: the evidence grade, the conservative planning capacity of 1,920 Wh, the continuous and startup bounds, the ranking rules and the disclosed contradictions on that page (1,920 versus 2,016 Wh, 4,000 versus 5,000 W) are exactly as before, because availability is a commercial fact and never technical proof. The R2500 Plus and the R1500 Lite are unchanged and stay qualified with no button while their exact SKUs remain sold out, as does the PECRON E1000LFP. Purchasable coverage on that date is BLUETTI 3 of 3, ALLPOWERS 1 of 3 and, after the separate PECRON re-check published above, PECRON 1 of 2, and paid search stays on HOLD.

11. 2026-08-12Method

    M1 and DX are open, D0 is pre-registered, and paid-search scaling stays on HOLD — superseded on 13 August 2026

    With M0-A signed, the existing deterministic refrigerator engine is reclassified as M1 rather than rebuilt in parallel: M1-A is open. DX is also open — a small comprehension protocol with five naive users on mobile and desktop, measuring five-second comprehension, completion, errors and unknowns, trust and usefulness, plus one end-to-end reconciliation of a single eligible attributed merchant click. A first US acquisition cell for the refrigerator decision, experiment BTB-US-REF-D0-01, is pre-registered in source with its campaign name and tracking templates. As of this 12 August 2026 entry the cell was not exposed and nothing had been spent; that state is superseded by the 13 August 2026 entry above, which records a launched controlled test. Scaling and phase 2 remain on HOLD until every buyable recommendation the cell can render has an active exact-SKU link, event reconciliation is proved with no duplicates and at least one network-confirmed click, and no button is ever shown for an unavailable or uncovered SKU. M2 and M3 are not open. No decision rule, product verdict, ranking, evidence row or availability correction was changed by this entry.

12. 2026-08-12Commercial

    Five exact SKUs lose their button after an availability re-check, and keep their technical qualification

    Official US exact-SKU pages were re-checked on 12 August 2026. The ALLPOWERS R1500 Lite, R2500 Plus and R2500 V2 show "OUT OF STOCK / Sold out", and both PECRON units — the E1000LFP and the E2400LFP — show "Notify Me When Available". All five stay technically QUALIFIED: nothing about their evidence, their sizing or their rank changed, and the M0-A technical signature is untouched. What changed is purchasability. Each of the five is now recorded as unavailable for US purchase with a dated note quoting the observed merchant state, no merchant button renders for them, and the internal product router refuses the redirect instead of sending anyone to a dead page, a substitute SKU or a homepage. The exact-SKU affiliate links already generated for those rows are retained only as prepared-link evidence, never as an active rail. The coverage audit therefore counts purchasable coverage, not prepared coverage: BLUETTI 3 of 3 qualified SKUs active, ALLPOWERS 0 of 3 with 3 unavailable, PECRON 0 of 2 with 2 unavailable, and Jackery, Anker SOLIX and EcoFlow still pending external approval. Paid search stays on HOLD; an unavailable SKU is automatically outside the buyable set and is never counted as commercially covered.

13. 2026-08-12Commercial

    Exact-SKU affiliate links are generated for BLUETTI and ALLPOWERS, and coverage is stated as open elsewhere

    Every technically qualified, purchasable BLUETTI and ALLPOWERS SKU now carries an exact-SKU Awin deep link (publisher 3023093; BLUETTI advertiser 59271, ALLPOWERS advertiser 40342, both accepted for the US). Each link encodes that row's own official exact-SKU product page as its destination — an audit exported from the affiliate layer checks this row by row, and refuses a homepage or a substitute SKU. Nothing about the technical layer moved: qualification, eligibility and order are computed before a destination is resolved, and a deterministic test proves the order is unchanged by a rail. Coverage is explicitly not closed for Jackery, Anker SOLIX, EcoFlow and DJI, whose approvals or tracking links are unproved; their qualified units keep linking to their exact official page and earn nothing. PECRON E1000LFP stays qualified with no button while that exact SKU cannot be bought in the US. Paid search stays on HOLD until every recommendable, purchasable SKU in a pre-registered campaign scope either has an active exact-SKU link or is explicitly named as out of that campaign's scope; the technical catalogue is never narrowed to make that gate pass.

14. 2026-08-12Signature

    M0 is signed: M0-A OK, M0-B OK at the architecture-contract level, M0 global SIGNED

    M0-A is signed OK: the architecture is chosen before any product is read, the comparative protocol is published in full with no compensation entering qualification or order, the catalogue stands at 17 of 17 exact US SKUs qualified on dated exact-SKU evidence, and four verdict families are replayed as frozen tests — dedicated exact SKU shown, general portable candidate shown for a compatible declared load, startup unknown refused with an explicit measurement request, and no-fit startup refused with an explicit reason. The dedicated architecture is genuinely served: BLUETTI FridgePower is eligible on the canonical dedicated case. M0-B is signed OK at the level of the deterministic architecture contract only, on five frozen families: TT-30 as a constraint and never proof of a 30 A service, measured 12 V critical loads on an adequate existing system, always-available shore power as the primary architecture with no phantom capacity band, finite mission sufficiency kept separate from daily loop closure without contradictory copy, and abstention or escalation for off-grid air conditioning declared without measured running watts, startup and hours. M0 global is SIGNED. What is explicitly not signed and not artificially declared closed: external cohort, blind scoring, conversion, merchant attribution and US unit economics all remain post-M0 gates.

15. 2026-08-12Commercial

    PECRON E1000LFP keeps its technical qualification and loses only its merchant link

    A dated public check on 12 August 2026 found the exact-SKU PECRON E1000LFP page live, but the E1000LFP variant itself showing "Notify Me When Available". The unit stays QUALIFIED and stays inside the M0 universe — it is neither deferred nor excluded, because nothing adverse about the hardware was proved. Only the merchant rail is switched off: no buy button and no affiliate destination are offered while the exact SKU cannot be purchased in the US, and the reason is published on the unit's own card as "Exact SKU currently unavailable for purchase in the US — checked 2026-08-12". The generated exact-SKU affiliate link is retained as evidence only. No merchant homepage and no substitute SKU is ever shown in its place. PECRON E2400LFP is unaffected and stays affiliate-active on its exact generated link.

16. 2026-08-12Correction

    The admission bar is corrected: measurement grade is no longer confused with admission

    The catalogue had been holding units back for evidence the engine does not actually use. Independent delivered-AC-energy and idle-draw measurements were treated as universal admission conditions, even though every unit is sized with the same conservative planning factor and no exact runtime is ever promised. That standard was unnecessarily high, and it fell hardest on the brands whose independent bench coverage is thinnest. Admission now requires exact-SKU identity and US availability, nominal energy, rated continuous output and a compatibility ceiling for the declared startup load, dated and taken from the official US page for that exact SKU; official manufacturer evidence is accepted for these and is labelled as a claim. Independent measurements still raise the evidence grade and the confidence behind an order, and when one is missing it is published as a limit on the unit's own card rather than used to hide the unit. Transfer evidence blocks only where the declared situation requires automatic switchover — and always for a dedicated refrigerator-backup product, where switching over is the product. Under this revision all three ALLPOWERS rows — the R1500 Lite, the R2500 Plus and the R2500 V2 — are qualified on evidence already held. The R2500 V2's official page contradicts itself on peak output and still carries re-used 2,016 Wh copy; that conflict is published as a disclosed limitation on its card, no peak figure is credited, and only the lower stated 1,920 Wh is used for planning. No commercial relationship moved a state or a rank, and no CTA was opened by qualification alone.

17. 2026-08-12Correction

    The old blanket non-qualified status is retired: four explicit states replace it

    The old single non-qualified state said too much. A unit whose evidence has simply not been captured had been published as excluded, which asserts an adverse conclusion that was never proved. The catalogue now carries three technical states — QUALIFIED, NOT YET QUALIFIED (critical facts missing, missing fields and closing action named, nothing adverse proved) and EXCLUDED (an affirmatively proved veto: failed measurement, exact-SKU conflict, proved incompatibility, or a unit not available in the US) — plus NOT SELECTED as a presentation state for a technically admissible unit that another option fitted better. Fourteen of the fifteen SKUs move to NOT YET QUALIFIED or stay QUALIFIED; only ALLPOWERS R2500 is excluded, on the proved conflict with the R2500 Plus. A not-yet-qualified unit may now show its exact official product page for information only, labelled 'Official product page — not a recommendation; evidence gate still open', with no affiliate parameter and no click event, and only when identity, US availability and the URL are all verified. The product router still refuses every state other than qualified.

18. 2026-08-12Catalog

    Fifteen exact US SKUs published, six qualified, eight not yet qualified and one excluded

    The refrigerator layer now reads a normalized, exact-SKU US catalogue instead of a two-row evidence list: two dedicated refrigerator units, six general portable stations around 1 kWh and seven around 2 kWh. Every row carries its official US product URL, nominal watt-hours, rated continuous output, credited startup output, transfer and recharge notes, dated sources and a public status reason. Six SKUs are QUALIFIED today (BLUETTI Elite 100 V2 and Elite 200 V2, Anker SOLIX C1000 and C2000 Gen 2, PECRON E1000LFP and E2400LFP). Eight are NOT YET QUALIFIED and say exactly which field is missing, with the action that would close it; nothing adverse is claimed about them. One — ALLPOWERS R2500 — is EXCLUDED, because the identity conflict against the R2500 Plus is affirmatively proved rather than merely unmeasured.

19. 2026-08-12Method

    Architecture is chosen before any product is read

    A refrigerator result now states the best architecture for the declared situation — a dedicated refrigerator unit, a general portable station, or no product at all — before ranking anything, and it always publishes why the other architecture was not selected. General portable stations are never hidden because a dedicated unit exists: every technically relevant option stays visible, marked not selected or not yet qualified, with the reason.

20. 2026-08-12Commercial

    An active affiliate rail can no longer produce a buy button for a SKU that has not passed the gate

    Ranking is computed from the declared situation and dated evidence, then a destination is resolved. Only BLUETTI Elite 200 V2 has a certified deep link today. Qualified units with no accepted program — Anker SOLIX and PECRON — link to their exact official product page under the words 'Official product page — no affiliate relationship currently'. ALLPOWERS remains outside the ranking despite its live Awin rail, Jackery, EcoFlow and DJI applications are pending, and Amazon Associates is not enrolled. Outbound clicks pass through an internal router that refuses every state other than qualified and never appends a tracking parameter to a non-affiliate URL.

21. 2026-08-11Correction

    Safety reserve corrected to 10% across the live engine

    The published research table and methodology stated a 0.90 retained reserve while the live sizing engine applied 15%. The engine, its arithmetic labels, the guide worked examples and the tests are now aligned on the single published convention: 85% AC delivery and 90% retained reserve, a combined 0.765 planning factor. Sizes produced before this date under the 0.72 factor were more conservative, not less; re-run any check you saved.

22. 2026-08-11Method

    Finite mission and durable daily loop are now separate statements

    A declared trip or outage that stored energy plus conservatively credited recharge can carry is no longer forced into an installed-system verdict merely because the daily use-and-recharge loop does not close. Both statements are now published side by side: whether this declared situation closes, and how much more credited recharge per day would be needed to repeat it indefinitely. Stored energy alone is still never presented as closing an open-ended loop.

23. 2026-08-11Method

    Decision Case, reason codes and versioned outputs

    Every completed check now carries stable reason codes, structured critical unknowns by gate, and engine, ruleset and catalog versions. The resulting Decision Case is saved on the reader's own device and can be downloaded as JSON. Declared inputs are never sent to a server.

24. 2026-08-11Catalog

    Product evidence carries exact SKU, market and separated sources

    Evidence rows now record the exact SKU, the market, a capture date and a source list that separates manufacturer claims from independent measurement. For the BLUETTI Elite 200 V2 (US), the official product page and manual carry the 2,073.6 Wh and 2,600 W claims, independent bench testing carries the ~94% AC efficiency and ~9.5 W idle figures, and Before the Battery's own 1,800 Wh eligibility ceiling remains stricter than both. Marketing power-lifting figures are not accepted as compressor-surge proof.

25. 2026-08-10Research

    Refrigerator outage sizing table v1.0 published

    Fifteen transparent scenarios convert an EnergyGuide-style annual figure (300–700 kWh/year) into 24-, 48- and 72-hour outage outlet energy and a required nominal capacity, using an 0.85 AC delivery factor and a 0.90 retained reserve factor (combined 0.765) and a zero recharge credit in the published baseline. The full table, formula and downloads (XLSX and CSV) are on /research/refrigerator-outage-sizing-table. Methodological boundaries: 0.85 and 0.90 are editorial planning assumptions, not independent measurements; every value in the table is a calculation, not a product test. The table screens capacity only — continuous output is not evaluated without a declared simultaneous load, startup surge must be verified or measured, and any duration above 24 hours with no credible recharge is reported as an open recharge gate requiring architecture review. It does not predict food safety and certifies no refrigerator or power station. Reviewed quarterly or after any formula, source or rule change.

26. 2026-08-08Method

    Refrigerator tool rebuilt as a four-step check with human-first verdicts

    The refrigerator diagnostic now asks one question group at a time — outage length, EnergyGuide energy, running and startup watts, then recharge and switchover — and answers in plain language with three pass/unknown/fail checks. Running watts are collected explicitly and are never derived from kWh per day; when they are unknown, compatibility is reported as unproven rather than assumed. The full arithmetic and assumptions remain available under the verdict.

27. 2026-08-08Affiliation

    First affiliate rails switched on behind an evidence gate

    BLUETTI and ALLPOWERS affiliate tracking is live for the US (Awin, publisher ID 3023093, verified 8 August 2026). Zendure's contract is executed but remains switched off until its tracking URL and attribution parameter are confirmed. A product is shown only when a signed, dated evidence row proves it covers the declared situation: an enabled rail alone never renders a product action, which is why the ALLPOWERS R2500 is still hidden. Eligibility and ranking are decided before commission and your price is unchanged.

28. 2026-08-08Measurement

    Anonymous first-party funnel measurement added

    A fixed, allowlisted set of funnel events is now recorded on our own infrastructure with a random per-session identifier, retained for 90 days. No form input, name, email, free text, raw IP address or user agent is stored, and no cookie is used. The privacy page describes the exact fields.

29. 2026-08-08Discovery

    Four evidence pages and an installed-backup next step published

    Direct answers for refrigerator backup, portable stations for refrigerators, RVs and travel trailers, plus a page listing the exact facts to collect before requesting installed-backup quotes. Visible FAQs were added to the refrigerator diagnostic and mirrored in structured data.

30. 2026-08-07Editorial

    Homepage entry clarified: one situation-first starting point

    The first screen now states what the publication does, why it helps and the single first action. The editorial-standards block moved below the situation chooser so it no longer competes with the entry path. No methodology was removed.

31. 2026-08-07Affiliation

    Three affiliate programs accepted the publisher; tracking infrastructure tested

    BLUETTI, ALLPOWERS and Zendure have administratively accepted the publisher. Deep-link and click-reference tracking was tested technically. No sale or commission had been observed at that date, and no merchant influences rankings or editorial conclusions. Acceptance is not an endorsement or a paid partnership.

32. 2026-08-03Editorial

    Editorial standards and independence rules published

    Situation before brand, source-backed specifications, dated volatile data, unknown values left unknown, ranking before commission, and logged corrections are now published as binding editorial rules.

33. 2026-08-03Method

    Sizing rules published for challenge

    Usable energy, conversion losses, safety reserve, peak treated as a pass/fail gate, and the full use-and-recharge cycle requirement are published in the methodology so the reasoning can be contested before anyone relies on it.

34. 2026-08-03Catalog

    Catalog policy: specifications separated from pricing

    Stable specification records carry a source and an evidence-quality note. Price and stock live in a separate time-stamped record and are never presented as current unless verified.

Rules behind these entries are described in the methodology.