Published 07:13 AEST · Before US Open · Wednesday, 9 September 2026Pipeline 04:30 AESTVol. I  No. 121
The Grid
Independent Price Reporting for the Compute Market
thegridco.ai
EST. 2026 · SYDNEY
Standards · corrections of record

Corrections

A correction restates what a value or a narration should have been on the information set of the day— a clerical error, a calculation error, a misapplication of the stated method. It is never a re-assessment on later information, and it never rewrites the print: the published value stands, the correction is filed as its own record with a reason code, appended to the affected daily as a notice, and served inline on the permanent print URL. This page is the same registry /api/print/corrections serves, made readable.
7 corrections filed · 26 print-date records · 24 print dates touched · 2 value restatements (2026-07-30, 2026-07-31) · everything else annotates

Ⅰ · The corrections policy

what is corrected · what is not · how it is filed
The honesty frame

What is corrected.Information that was available when the print was determined: a value computed wrongly under the stated method, a methodology applied on the wrong date, a narration that attributed a number to a print it did not describe. The correction states the original, what stands corrected, and the print’s own move where the record states it.

What is not. A defective method is not corrected — it is changed by notice (/standards/notices), and every value computed correctly under it stands. New information learned after publication never restates a value. A missing edition is filed as a gap of record and is never backfilled.

How it is filed. One document per correction in the repository (docs/corrections/), one registry entry per (correction, print date, series), a CORRECTION NOTICE appended to the affected daily report (annotation, never a silent rewrite), and the entry served inline in the correctionsarray of the permanent print URL /api/print/<date>. Only two rows in the record carry a restated value; every other correction annotates.

Reason codes · verbatim from the registry
Reason codeMeaningRecords
methodology_effective_dateA methodology change governed a print before its announced effective date (ADR-003 §2 publish-before-effect); values stand, the effective date is corrected.1
rebase_contaminationA declared panel re-base leaked into a derived print as a market move; rows rewritten under a chain-linked methodology (v1.1), pre-correction values archived.6
narration_composition_as_pricePanel-composition change narrated as price behaviour; values stand, narration corrected by annotation.6
narration_absence_as_stillnessAbsence of a like-for-like measurement narrated as the print's stillness while the print moved on intra-panel repricing; values stand.4
narration_matched_pair_as_session_deltaThe 7-day matched-pair (like-for-like) figure narrated as the session's own move; values stand.2
print_missingNo print of record exists for the date; the gap stands and is never backfilled.7

Ⅱ · The corrections of record

newest filing first · permalink per correction id
2026-08-31-pipeline-outage-missing-printsfiled 2026-08-31annotation only · values standapi record →
Print dateSeriesReasonPublished (original)Stands correctedPrint move / note
2026-08-25printprint_missingNo daily print exists for 2026-08-25 (would-be No. 112) — the scheduled cycle failed at its CI test gate before any scraper ran (dashboard node dependencies never installed in the workflow); nothing was collected or printed. First of six consecutive gap dates (2026-08-25 → 2026-08-30).The gap stands as recorded and is never backfilled (first-sight custody: the observations were never collected, so no honest retroactive print can exist). Workflow fix + resumption print No. 112 dated 2026-08-31; notice of record 2026-08-31-pipeline-outage-and-window-pause.—
2026-08-26printprint_missingNo daily print exists for 2026-08-26 (would-be No. 113) — the scheduled cycle failed at its CI test gate before any scraper ran; nothing was collected or printed. Second of six consecutive gap dates (2026-08-25 → 2026-08-30).The gap stands as recorded and is never backfilled. Notice of record 2026-08-31-pipeline-outage-and-window-pause.—
2026-08-27printprint_missingNo daily print exists for 2026-08-27 (would-be No. 114) — the scheduled cycle failed at its CI test gate before any scraper ran; nothing was collected or printed. Third of six consecutive gap dates (2026-08-25 → 2026-08-30).The gap stands as recorded and is never backfilled. Notice of record 2026-08-31-pipeline-outage-and-window-pause.—
2026-08-28printprint_missingNo daily print exists for 2026-08-28 (would-be No. 115) — the scheduled cycle failed at its CI test gate before any scraper ran; nothing was collected or printed. Fourth of six consecutive gap dates (2026-08-25 → 2026-08-30).The gap stands as recorded and is never backfilled. Notice of record 2026-08-31-pipeline-outage-and-window-pause.—
2026-08-29printprint_missingNo daily print exists for 2026-08-29 (would-be No. 116) — the scheduled cycle failed at its CI test gate before any scraper ran; nothing was collected or printed. Fifth of six consecutive gap dates (2026-08-25 → 2026-08-30).The gap stands as recorded and is never backfilled. Notice of record 2026-08-31-pipeline-outage-and-window-pause.—
2026-08-30printprint_missingNo daily print exists for 2026-08-30 (would-be No. 117) — the scheduled cycle failed at its CI test gate before any scraper ran; nothing was collected or printed. Sixth of six consecutive gap dates (2026-08-25 → 2026-08-30).The gap stands as recorded and is never backfilled. The pipeline resumed with print No. 112 dated 2026-08-31; the six numbers were never consumed. Notice of record 2026-08-31-pipeline-outage-and-window-pause.—
Document of record: docs/corrections/2026-08-31-pipeline-outage-missing-prints.md · records 6 · permanent print URLs: https://thegridco.ai/api/print/2026-08-25 · https://thegridco.ai/api/print/2026-08-26 · https://thegridco.ai/api/print/2026-08-27 · https://thegridco.ai/api/print/2026-08-28 · https://thegridco.ai/api/print/2026-08-29 · https://thegridco.ai/api/print/2026-08-30
2026-08-14-gap-composition-narration-sweepfiled 2026-08-16annotation only · values standapi record →
Print dateSeriesReasonPublished (original)Stands correctedPrint move / note
2026-07-30gap_h100_spotnarration_absence_as_stillness"GAP-H100 assessed spot holds at $2.45/hr — a zero-delta session for the executable-ask leg" · "GAP-H100 $2.45/hr unchanged" · "today's static GAP-H100 assessed ask is a pause in the catch-up cycle".Composition unchanged; Vast.ai's availability-verified ask fell $2.4292 → $2.2027 (−9.3%) day-over-day, reverting the prior session's spike. The 7-day matched window crossed the declared 2026-07-24 GAP v3 re-base — a re-base echo, not a market measure — so no honest like-for-like existed; that absence was rendered as stillness. Values stand as printed.$2.5596 → $2.4463 (−$0.1132, −4.42%)
2026-08-01gap_h100_spotnarration_matched_pair_as_session_deltaRegister: "Compute Supply — firm · rose +1.5%".The +1.5% is the like-for-like 7-day matched-pair measure (+1.51%, window 2026-07-25→2026-08-01) — coincidentally also Vast.ai's own day move (+1.50%) — attributed to the session's print without its label; the print's own move was +0.67%. The prose narration ("moved to $2.46/hr … a +$0.01 increment") was correct; the register line alone misattributed. Values stand as printed.$2.4460 → $2.4625 (+$0.0165, +0.67%)
2026-08-02gap_h100_spotnarration_composition_as_price"GAP-H100 assessed spot holds at $2.69/hr" · "GAP-H100 returns to zero-delta after the prior session's fractional +$0.01 increment" · register "balanced · unchanged".Vast.ai published no availability-verified offers; the panel narrowed to RunPod alone — a provider static at $2.69 across its entire stored history. The print rose to the sole remaining provider's posted price by panel re-composition. No like-for-like measurement was available; "zero-delta" was not a finding, it was the absence of one. Values stand as printed.$2.4625 → $2.6900 (+$0.2275, +9.24%)
2026-08-04gap_h100_spotnarration_composition_as_price"GAP-H100 assessed spot held at $2.54/hr — zero delta on the session, with the assessed executable-ask composite unchanged" · register "balanced · flat".Vast.ai returned to the bookable panel after two sessions absent; the composite moved down from the single-provider $2.69 as the panel re-widened. Like-for-like, Vast's H100 SXM 80GB ask returned at $2.3892, +5.3% above its last observation (2026-08-01) — the only price movement in the panel, upward, inside a downward composition move. Values stand as printed.$2.6900 → $2.5396 (−$0.1504, −5.59%)
2026-08-05gap_h100_spotnarration_composition_as_price"GAP-H100 assessed spot holds at $2.69/hr — zero delta for a second consecutive session" · register "balanced · unchanged".Vast.ai absent again; the panel narrowed to the static sole provider and the print rose to its posted price by re-composition. "Second consecutive zero-delta session" was false on both counts — the prior session's print moved −5.59%, and this session's moved +5.92%. Values stand as printed.$2.5396 → $2.6900 (+$0.1504, +5.92%)
2026-08-08gap_h100_spotnarration_absence_as_stillness"GAP-H100 assessed spot is unchanged at $2.45/hr — a zero-delta session following yesterday's +1.5% discrete step up" · register "BALANCED · edged up +0.0%".Composition unchanged; Vast.ai's ask fell $2.2693 → $2.2008 (−3.0%). The "zero delta" is the like-for-like 7-day measure (0.00%, window 2026-08-01→2026-08-08) narrated as the session's stillness without its label; "yesterday's +1.5% discrete step up" restates the corrected 2026-08-07 misstatement. Values stand as printed.$2.4797 → $2.4454 (−$0.0343, −1.38%)
2026-08-09gap_h100_spotnarration_absence_as_stillness"GAP-H100 assessed spot … holds at $2.48/hr — a +0.0% delta for the session" · "With GAP-H100 static, any basis movement in this session is fully attributed to OCPI velocity".Composition unchanged; Vast.ai's ask rose $2.2008 → $2.2687 (+3.1%). The narration quoted the print's new level while claiming a zero delta. No like-for-like measure existed (Vast absent at the baseline end of the 2026-08-02→2026-08-09 window); that absence was rendered as stillness. Values stand as printed.$2.4454 → $2.4793 (+$0.0339, +1.39%)
2026-08-10gap_h100_spotnarration_absence_as_stillness"GAP-H100 assessed spot holds at $2.24/hr — zero delta, a second consecutive pause following the prior discrete step-up sequence" · "GAP-H100 holds, second consecutive flat session".Composition unchanged; Vast.ai's ask fell $2.2687 → $1.7872 (−21.2%) day-over-day. The narration quoted the collapsed level while claiming stillness, and "second consecutive pause" was false on both counts — the prior session's print rose +1.39% and this session's fell −9.71% (the largest misstatement in the swept window). No like-for-like measure existed; that absence was rendered as stillness. Values stand as printed.$2.4793 → $2.2386 (−$0.2407, −9.71%)
2026-08-12gap_h100_spotnarration_composition_as_priceSIGNALS "GAP-H100 … printed: $2.69/hr [daily print, arrived today]" · register "balanced · unchanged".Vast.ai published no availability-verified offers; the panel narrowed to RunPod alone and the print rose to the static provider's posted price by re-composition. No like-for-like measurement available. Values stand as printed.$2.5802 → $2.6900 (+$0.1098, +4.25%)
2026-08-13gap_h100_spotnarration_composition_as_price"GAP-H100 assessed spot holds at $2.45/hr — zero delta from yesterday."Vast.ai returned to the bookable panel; the composite moved down from the single-provider $2.69 as the panel re-widened. Like-for-like, Vast's H100 SXM 80GB ask returned at $2.2027, −10.8% below its 2026-08-11 observation ($2.4705) — real downward price movement arriving inside the composition transition, which the "zero delta" narration erased entirely. Values stand as printed.$2.6900 → $2.4463 (−$0.2437, −9.06%)
Document of record: docs/corrections/2026-08-14-gap-composition-narration-sweep.md · records 10 · permanent print URLs: https://thegridco.ai/api/print/2026-07-30 · https://thegridco.ai/api/print/2026-08-01 · https://thegridco.ai/api/print/2026-08-02 · https://thegridco.ai/api/print/2026-08-04 · https://thegridco.ai/api/print/2026-08-05 · https://thegridco.ai/api/print/2026-08-08 · https://thegridco.ai/api/print/2026-08-09 · https://thegridco.ai/api/print/2026-08-10 · https://thegridco.ai/api/print/2026-08-12 · https://thegridco.ai/api/print/2026-08-13
2026-08-16-gap-2026-08-11-matched-as-incrementfiled 2026-08-16annotation only · values standapi record →
Print dateSeriesReasonPublished (original)Stands correctedPrint move / note
2026-08-11gap_h100_spotnarration_matched_pair_as_session_delta"GAP-H100 assessed spot … rising to $2.58/hr, a +1.7% increment, breaks the two-session zero-delta pause."The print rose $2.2386 → $2.5802 (+15.26%) day-over-day as Vast.ai's ask recovered $1.7872 → $2.4705 (+38.2%) from the prior session's collapse; panel composition unchanged. The narrated +1.7% is the 7-day matched-pair measure (window 2026-08-04→2026-08-11) standing in for the day's move — a genuine price move understated ~9×. "Two-session zero-delta pause" was false on both counts (08-09 +1.39%, 08-10 −9.71%). Values stand as printed.$2.2386 → $2.5802 (+$0.3417, +15.26%)
Document of record: docs/corrections/2026-08-16-gap-2026-08-11-matched-as-increment.md · records 1 · permanent print URL: https://thegridco.ai/api/print/2026-08-11
2026-08-14-gap-2026-08-07-composition-narrationfiled 2026-08-14annotation only · values standapi record →
Print dateSeriesReasonPublished (original)Stands correctedPrint move / note
2026-08-07gap_h100_spotnarration_composition_as_price"GAP-H100 assessed spot moved to $2.48/hr, a +1.5% step up — the first repricing after two zero-delta sessions…" · "today's move is the first signal that asks have adjusted to the accumulated pressure" · SIGNALS "$2.48/hr, +1.5%" · register "Compute Supply — FIRM · rose +1.5%".No repricing occurred. Vast.ai re-entered the bookable panel after two sessions absent; the assessed composite moved $2.69 → $2.48 (−$0.21, −7.8%) purely by panel re-composition (composition-fixed counterfactual delta $0.0000). The "+1.5%" is the 7-day matched-pair measure (+1.51%, window 2026-07-31→2026-08-07) — real and composition-immune, but not the print's same-day move; Vast's own ask returned −5.0% below its 2026-08-04 observation. The "step up" and "asks have adjusted" claims are withdrawn by annotation. Values stand as printed.$2.6900 → $2.4796 (−$0.2104, −7.8%) by panel re-composition
Document of record: docs/corrections/2026-08-14-gap-2026-08-07-composition-narration.md · records 1 · permanent print URL: https://thegridco.ai/api/print/2026-08-07
2026-08-09-editorial-gap-2026-08-06filed 2026-08-09annotation only · values standapi record →
Print dateSeriesReasonPublished (original)Stands correctedPrint move / note
2026-08-06printprint_missingNo daily print exists for 2026-08-06 — the scheduled cycle started ~5.5 h late, past UTC midnight, printed itself as 2026-08-07, and was then overwritten by the scheduled 2026-08-07 cycle.The gap stands as recorded and is never backfilled (a backfill would fabricate an observation cycle that never ran). Durable fix shipped 2026-08-09: the report date is resolved from the schedule anchor, not the run clock.—
Document of record: docs/corrections/2026-08-09-editorial-gap-2026-08-06.md · records 1 · permanent print URL: https://thegridco.ai/api/print/2026-08-06
2026-08-05-divergence-rebase-contaminationfiled 2026-08-05values restatedapi record →
Print dateSeriesReasonPublished (original)Stands correctedPrint move / note
2026-07-27gccx_divergencerebase_contamination-28.72-28.72Value unchanged; row re-versioned to divergence v1.1 with the GAP leg (+24.98) now measured within the v3 panel (07-24→07-27), chain detail recorded.
2026-07-29gccx_divergencerebase_contamination-41.57-41.57Value unchanged; row re-versioned to v1.1, GAP leg (+32.69) measured within the v3 panel (07-24→07-29).
2026-07-30gccx_divergencerebase_contamination-27.23-35.31restatedChain-linking restored the GAP leg (+24.98) the declared 2026-07-24 break had knocked out.
2026-07-31gccx_divergencerebase_contamination-46.63-21.65restatedHeadline correction: the cross-break matched-pair median (+49.96, a methodology step) replaced by the within-panel reading (0.00).
2026-08-03gccx_divergencerebase_contamination-16.33-16.33Value unchanged; DRAM-only physical leg both before and after (v3 segment below the pair floor — honest skip). Row re-versioned to v1.1.
2026-08-04gccx_divergencerebase_contamination-12.37-12.37Value unchanged; DRAM-only physical leg. Row re-versioned to v1.1.
Document of record: docs/corrections/2026-08-05-divergence-rebase-contamination.md · records 6 · permanent print URLs: https://thegridco.ai/api/print/2026-07-27 · https://thegridco.ai/api/print/2026-07-29 · https://thegridco.ai/api/print/2026-07-30 · https://thegridco.ai/api/print/2026-07-31 · https://thegridco.ai/api/print/2026-08-03 · https://thegridco.ai/api/print/2026-08-04
2026-07-25-gap-v3-effective-datefiled 2026-07-27annotation only · values standapi record →
Print dateSeriesReasonPublished (original)Stands correctedPrint move / note
2026-07-24gap_h100_spotmethodology_effective_dateGAP-H100 $2.38/hr narrated as a +$0.30 (+14.4%) step from $2.08/hr — 'an administrative repricing decision by one or more panel providers' — and counted by THE SCORE as confirmation of the founding memory call.The step was the GAP v3 executable-ask panel re-base taking effect on 2026-07-24, one print before its announced date (publish-before-effect breach, ADR-003 §2): panel-definition change, not market clearing. v3 effective date corrected to 2026-07-24; the call-confirmation claim is withdrawn by annotation. Values stand as printed.$2.0792 → $2.3797 (+14.45%) on panel-definition change
Document of record: docs/corrections/2026-07-25-gap-v3-effective-date.md · records 1 · permanent print URL: https://thegridco.ai/api/print/2026-07-24
Registry: dashboard/lib/corrections-registry.json — the read-side of docs/corrections/*.md, parity-tested in both directions on every build; served inline by /api/print. Add a correction by filing its document, appending the notice to the affected report, and registering the entry. Never edit a served entry except to fix a transcription error against its document.