Tripwire #1 fired on one consistent basket: the "server DRAM" label was a vault citation error, not a basket switch

The question

"Is Tripwire #1's firing real or a basket artifact? The 58-63% and 13-18% readings are SERVER DRAM while the 10-15% that fired it is CONVENTIONAL DRAM - find one consistent ASP series spanning the whole tripwire window, or establish that no single free series exists and the tripwire must be rewritten to name its basket."

Context: [[2026-10-01-micron-fq4-fy2026-dram-hbm-tripwire-rescan]] scored Tripwire #1 as FIRED for the first time since the scoreboard was written, flagged a server-DRAM-to-conventional-DRAM basket switch as its one unresolved caveat, and queued this question as the test of whether the only fired indicator counts.

What we already know (from the vault)

• Tripwire #1 as currently written, verbatim from the scoreboard row [[2026-07-07-dram-hbm-phase2-phase3-early-signals]]: metric "ASP rate-of-change (2nd-derivative decel) - earliest, noisiest"; threshold "QoQ contract-price gain decelerates for 2+ consecutive quarters (leads the outright rollover)"; mid-2026 reading "2Q26 +58-63% → 3Q26 +13-18%; consumer demand weakening; 'flexible pricing'"; verdict "FLICKER (1 of 2 quarters). Watch 4Q26. Confounded by base effect - do NOT treat as Phase 3 alone". The row names no basket. That omission is the root cause of everything below.
• The scoreboard's own body text does name the basket, and names it correctly. The same brief's "what the web says" section reads "TrendForce (Jul 3 2026) projects 3Q26 conventional DRAM contract prices +13-18% QoQ - down from 2Q26's +58-63% QoQ" [[2026-07-07-dram-hbm-phase2-phase3-early-signals]]. The predecessor brief is equally explicit: "2Q26 conventional DRAM contract prices projected +58-63% QoQ" [[2026-06-28-chip-memory-cycle-phase2-phase3-indicators]].
• The "server DRAM" label enters the vault on 2026-09-25, and the same sentence admits it is unverified. [[2026-09-25-hbm-dram-phase-2-3-transition-q3-2026]] writes: "3Q26 server DRAM +13-18% QoQ ('Long-Term Agreements Cap Price Increases; Server DRAM Contract Prices Expected to Rise 13-18% QoQ in 3Q26,' TrendForce, 2026-07-09), against 2Q26's +58-63%. Both readings are carried from [the July 7 brief]; the July 9 headline was re-surfaced this run but the body was not fetched." The scoreboard row in that brief then reads "2Q26 +58-63% → 3Q26 +13-18% (server DRAM)".
• The 2026-10-01 rescan inherited the relabel and flagged the consequence. Its row 1 reads "2Q26 +58-63% → 3Q26 +13-18%, server DRAM (2026-07-09)" against "4Q26 conventional DRAM +10-15%", verdict "FIRED (qualified)", caveat "the basket changes from server DRAM to conventional DRAM" [[2026-10-01-micron-fq4-fy2026-dram-hbm-tripwire-rescan]].
• The vault's own rule covers this exact failure: a status written where the founder reads it must be re-checked at its source, not carried from notes (MEMORY: "Verify blockers against the source, not notes"; "git grep / headline-only verification" failure family).

What the web says

All four readings below were retrieved directly from TrendForce's free press centre on 2026-10-03 (HTTP 200, full article bodies parsed, not headline snippets). Quotes are verbatim.

• 1Q26, conventional DRAM, retrospective survey: "the memory industry experienced a significant boost in 1Q26 due to rapidly rising contract prices for conventional DRAM, which increased by approximately 93% to 98% QoQ" (TrendForce 2026-06-01).
• 2Q26, conventional DRAM, forecast: "TrendForce therefore expects conventional DRAM contract prices to rise another 58-63% QoQ in 2Q26" (same release, TrendForce 2026-06-01). This is the origin of the 58-63% figure, and it is labelled conventional DRAM at source.
• 3Q26, conventional DRAM, forecast: "Conventional DRAM contract prices are forecast to rise 13-18% QoQ in 3Q26, while NAND Flash contract prices are expected to increase 10-15%" (TrendForce 2026-07-03). So the 13-18% figure is also conventional DRAM at source.
• 3Q26 reaffirmed, two months into the quarter: "Conventional DRAM contract price growth, however, is projected to moderate to 13-18% QoQ" (TrendForce 2026-09-07, the 2Q26 DRAM industry revenue release, +59.5% QoQ to US$154.73B). The 3Q26 reading survived to within three weeks of quarter-end unrevised.
• 4Q26, conventional DRAM, forecast: "Conventional DRAM contract prices are projected to grow 10-15% QoQ in 4Q26... However, the pace of contract price increases is expected to moderate" (TrendForce 2026-09-30).
• The source of the confusion, confirmed: a separate TrendForce release does quote server DRAM at the same range: "TrendForce predicts that server DRAM contract prices will rise by 13-18% quarter-over-quarter in 3Q26" (TrendForce 2026-07-09). Two different baskets carry an identical range in the same quarter by coincidence. That numeric collision is what let a headline-only re-surface pass undetected. Note the same collision risk across metrics: 10-15% is NAND Flash in 3Q26 and conventional DRAM in 4Q26.
• Free per-segment percentage series do not exist on a regular cadence. Both quarterly pricing releases carry per-segment prose (PC DRAM, server DRAM, mobile DRAM, graphics DRAM, consumer DRAM) with no segment-level percentages - e.g. 4Q26 server DRAM gets "expected to remain undersupplied" and "increases for some suppliers are expected to lag the market average due to... long-term agreements (LTAs)", no number. The 2026-07-09 server-DRAM figure is an occasional one-off, not a series.

Source inventory, with access status and actual basket coverage:

No DRAM-specific free index with monthly values and current coverage was found. The single consistent free series that spans the tripwire window is the TrendForce conventional-DRAM contract-price QoQ range.

Convergences and contradictions

• Convergence, and it is the answer: all three tripwire readings are the same basket. 58-63%, 13-18% and 10-15% are each labelled "conventional DRAM contract prices" in the TrendForce release that originated them. There was no basket switch. The 2026-10-01 caveat was describing a defect in the vault's citation chain, not in the data.
• Contradiction, inside the vault rather than between vault and web: three vault docs label the same number three ways. 2026-06-28 and 2026-07-07 say conventional DRAM (correct); 2026-09-25 says server DRAM (incorrect, and self-flagged as body-not-fetched); 2026-10-01 inherits the error and then reasons honestly from it. The mechanism was a coincidental range collision between two TrendForce releases published six days apart.
• Convergence on direction, with one weak link. On the consistent series, the deceleration is monotone on every statistic of the range - lower bound 93 → 58 → 13 → 10, upper bound 98 → 63 → 18 → 15, midpoint 95.5 → 60.5 → 15.5 → 12.5. The weak link is 3Q26 → 4Q26: those two ranges overlap between 13 and 15, so a realized 3Q26 at the bottom of its band against a realized 4Q26 at the top would be an acceleration. The 1Q→2Q and 2Q→3Q steps have no overlap at all.

Synthesis for RDCO

Outcome (a). One consistent free ASP series spans the entire tripwire window, Tripwire #1 re-scores as FIRED on that series, and the firing is slightly stronger than the 2026-10-01 rescan claimed. The series is TrendForce's conventional-DRAM contract-price QoQ range, published quarterly on the free press centre. Scored strictly forecast-to-forecast, the window is 2Q26 +58-63% → 3Q26 +13-18% → 4Q26 +10-15%: two consecutive decelerations, which is exactly the pre-registered threshold, on one basket with no substitution. Add the retrospective 1Q26 reading of +93-98% and it is three consecutive decelerations. The 3Q26 leg is additionally corroborated by TrendForce reaffirming 13-18% on 2026-09-07, two months into the quarter. So the one piece of movement on the scoreboard does not evaporate. It survives the test this question was written to apply.

What does change is the confidence attached to the vault's own labels, and that is the more expensive finding. A headline was re-surfaced without fetching the body, a coincidental numeric match (two TrendForce releases six days apart both quoting 13-18%, for different baskets) made the mislabel invisible, and the error then propagated into a scored row and generated a research question that cost a full brief to close. Two structural fixes follow. First, the scoreboard row for #1 never named its series, so there was nothing for a later reader to check the label against. Second, the 2026-09-25 brief disclosed "the body was not fetched" in the same sentence as the number it was labelling, and that disclosure did not stop the number from being scored. A disclosed-but-unenforced caveat is not a gate. The cheap version of the gate: any figure entering a scored tripwire row must carry a retrieved-body citation, and no scoreboard row may cite a source whose body was not fetched in the run that scored it.

The firing should not be upgraded in meaning, only in cleanliness. Two reasons to keep conviction where [[2026-10-01-micron-fq4-fy2026-dram-hbm-tripwire-rescan]] put it. First, vintage asymmetry: the 4Q26 reading was published 2026-09-30, one day before the quarter it forecasts, making it the least-informed point in the series and the most likely to be revised, while 2Q26's figure was published two months into its quarter. Second, and more important, a growth rate decaying from +95% is partly arithmetic. On midpoints the cumulative contract-price level (my arithmetic, index 4Q25 = 100, product of midpoints) runs 100 → 196 → 314 → 362 → 408. The level is still roughly 4.1x its 4Q25 base and still making new highs at +12.5% a quarter. A second derivative turning while the first derivative sits at +12.5% is precisely the state the tripwire set was designed to detect and precisely why #1 was never allowed to flip the phase alone. The composite rule is untouched: #4 has zero months of decline, #2 has one quarter of DIO rise, so no Phase-3 declaration is arithmetically available. Posture unchanged; nothing here is a position action.

A rewrite of Tripwire #1 is still warranted, for a different reason than the question anticipated. Outcome (b) was the escape hatch if no consistent series existed. It does exist, so the rewrite is not a substitute for scoring - it is hygiene that prevents the same ambiguity recurring, because the row's silence about its basket is what allowed the drift in the first place. Proposal below, not applied.

PROPOSED replacement for Tripwire #1 - awaiting the founder's read, NOT applied

I have not edited [[2026-07-07-dram-hbm-phase2-phase3-early-signals]]. Clause A is the basket fix and I hold it with high confidence. Clause B is a separable annotation suggestion I hold with lower confidence; accept or reject it independently.

Clause A (basket fix), verbatim replacement row:

| 1 | ASP rate-of-change (2nd-derivative decel) - earliest, noisiest. SERIES (binding): TrendForce "conventional DRAM" contract price, QoQ % range, as published in the quarterly memory-pricing release on the free press centre. "Conventional DRAM" = all DRAM excluding HBM, blended across server, PC, mobile, graphics and consumer DRAM. CONTRACT only, never spot. Do NOT substitute a segment figure (server DRAM, PC DRAM, mobile DRAM, consumer DRAM), a company ASP (Micron, SK hynix, Samsung), a spot index (DRAMeXchange DXI), or a PPI aggregate into this row; those are logged as corroboration or under #4, never as a reading of #1. If TrendForce stops publishing this range free, the row goes UNREADABLE - it is never scored on a substitute basket. | The QoQ gain decelerates on all three statistics of the published range (lower bound, midpoint, upper bound) for 2+ consecutive quarters, compared forecast-to-forecast within the same series. A quarter whose range overlaps the prior quarter's range counts as a WEAK deceleration and must be annotated as such in the reading cell. | 1Q26 +93-98% (retrospective) → 2Q26 +58-63% → 3Q26 +13-18% (reaffirmed 2026-09-07) → 4Q26 +10-15%. All four conventional DRAM, TrendForce free press centre, bodies retrieved 2026-10-03 | FIRED. Three consecutive decelerations, two of them forecast-to-forecast with non-overlapping ranges; the 3Q26→4Q26 step is WEAK (ranges overlap 13-15). Confounded by a compounding base - do NOT treat as Phase 3 alone; the composite rule still governs |

Clause B (annotation, separable, lower confidence): add a standing annotation requirement, not a threshold change - changing the threshold after seeing the data would break the pre-registration discipline the scoreboard exists to enforce. "Each #1 reading is recorded alongside the cumulative contract-price index (product of midpoints, 4Q25 = 100) and the current QoQ gain. A firing while the cumulative index is still making new highs at a gain above +5% QoQ is annotated ARITHMETIC-DOMINATED - a growth rate decaying from a near-doubling quarter falls partly by construction."

Why this is in the vault

It closes the single open caveat on the only fired indicator in the memory-cycle Phase 2→3 tripwire set [[2026-10-01-micron-fq4-fy2026-dram-hbm-tripwire-rescan]], and it does so by re-deriving the series from primary sources rather than from vault notes: the "server DRAM" basket switch never happened, so the memory-cycle-v1.1 phase-tracker keeps its one bearish signal and the Phase-2 posture needs no revision. It also supplies the verbatim text for a basket-naming rewrite of Tripwire #1 in [[2026-07-07-dram-hbm-phase2-phase3-early-signals]], pending the founder's read.

Open follow-ups

• Capture the realized 4Q26 conventional-DRAM print when it lands (expected in TrendForce's 1Q27-forecast release, roughly early January 2027, and in the 4Q26 DRAM industry revenue release, roughly early March 2027). The 4Q26 figure now in the row is a forecast published one day before its quarter, and the 3Q26→4Q26 ranges overlap, so the realized values are what settle whether that step is a deceleration at all.
• Backfill anchors/dram-spot/ with the conventional-DRAM contract QoQ series (1Q26 through 4Q26, four rows, all sourced here). The anchor's newest row is 2026-04-24, so the series the tripwire is scored on is not actually stored anywhere machine-readable. Carried from the 2026-10-01 brief; this brief supplies the values.
• Decide whether a second independent ASP series is worth constructing from Micron's sequential DRAM ASP disclosures (free, company-level, includes HBM, fiscal quarters offset ~1 month). It would be an imperfect but genuinely independent check on TrendForce, and TrendForce is currently a single point of failure for the earliest tripwire in the set.
• Decide whether a paid TrendForce DRAM contract-price subscription is worth it now that free coverage has been mapped. The free tier gives one blended QoQ range per quarter and no per-segment percentages; the per-segment and per-part series are behind either a subscription or a DRAMeXchange login. This is a spend decision, not a Ray execution item.
• Add the citation gate named in the synthesis to the tripwire-rescan procedure: no scoreboard row may cite a source whose body was not fetched in the run that scored it. The 2026-09-25 brief disclosed exactly that condition and the number was scored anyway.

Related

• [[2026-10-01-micron-fq4-fy2026-dram-hbm-tripwire-rescan]] - the parent brief; fired #1, flagged the basket caveat, queued this question
• [[2026-07-07-dram-hbm-phase2-phase3-early-signals]] - the five-row tripwire scoreboard; row 1 is what the proposal above would replace
• [[2026-09-25-hbm-dram-phase-2-3-transition-q3-2026]] - where the "server DRAM" label entered the vault, self-flagged as body-not-fetched
• [[2026-06-28-chip-memory-cycle-phase2-phase3-indicators]] - the predecessor indicator set; labels 2Q26 correctly as conventional DRAM
• [[2026-05-18-memory-cycle-v1.1]] - the executable thesis whose exit anchors these tripwires instrument
• [[2026-09-08-cxmt-ymtc-bit-share-phase3-threshold]] - sibling re-keying of tripwire row 5; same pattern of fixing an under-specified row

Sources

Vault

• ~/rdco-vault/06-reference/research/2026-10-01-micron-fq4-fy2026-dram-hbm-tripwire-rescan.md
• ~/rdco-vault/06-reference/research/2026-07-07-dram-hbm-phase2-phase3-early-signals.md
• ~/rdco-vault/06-reference/research/2026-09-25-hbm-dram-phase-2-3-transition-q3-2026.md
• ~/rdco-vault/06-reference/research/2026-06-28-chip-memory-cycle-phase2-phase3-indicators.md
• ~/rdco-vault/01-projects/investing/theses/2026-05-18-memory-cycle-v1.1.md
• ~/rdco-vault/06-reference/research/2026-09-08-cxmt-ymtc-bit-share-phase3-threshold.md

Web - retrieved 2026-10-03, full article bodies parsed

• Rapid Contract Price Surge Drives 1Q26 DRAM Industry Up 81% QoQ - TrendForce, 2026-06-01 - 1Q26 conventional DRAM +93-98% QoQ; 2Q26 forecast +58-63% QoQ
• AI Server Demand Continues to Support Memory Prices in 3Q26... - TrendForce, 2026-07-03 - 3Q26 conventional DRAM +13-18% QoQ; NAND +10-15%
• Long-Term Agreements Cap Price Increases; Server DRAM Contract Prices Expected to Rise 13-18% QoQ in 3Q26 - TrendForce, 2026-07-09 - the server-DRAM release whose coincident range caused the vault mislabel
• DRAM Industry Revenue Rises 59.5% QoQ in 2Q26... - TrendForce, 2026-09-07 - reaffirms 3Q26 conventional DRAM at +13-18% QoQ
• AI Server Demand Sustains Memory Contract Price Increases in 4Q26... - TrendForce, 2026-09-30 - 4Q26 conventional DRAM +10-15% QoQ; NAND +15-20%; pace "expected to moderate"
• FRED PCU334413334413, PPI: Semiconductor and Related Device Manufacturing - free monthly; 29.620 (Jan 2026) → 28.472 (Aug 2026); wrong basket and wrong sign, ruled out
• DRAMeXchange homepage and /Price/ pages - contract and spot detail pages return a member login form; DXI live value renders via iframe, static-HTML value is commented-out stale markup. Exists, values not retrieved

Paywalled or gated, not retrieved, not retried

• TrendForce DRAM Market Bulletin (RP260916TA) and DRAM Contract Price dataset - per-part monthly contract series
• DRAMeXchange per-part contract and spot price tables - login-gated
• Statista South Korea DRAM PPI - values gated and coverage ends 2024, cannot span the 2026 window