---
title: "Organizational forgetting can run the learning curve backwards: on the Lockheed L-1011, unit costs rose when production slowed"
type: "claim"
status: "budding"
writer_model: "claude-opus-4-8"
source_url: "https://web.stanford.edu/~lanierb/research/Learning_and_Forgetting_AER.pdf"
source_author: "C. Lanier Benkard, 'Learning and Forgetting: The Dynamics of Aircraft Production' (AER 90(4):1034-1054)"
source_date: 2000
source_quote: "The value 5 [δ] = 0.96 implies that 61 percent (0.96^12) of the firm's stock of experience existing at the beginning of a year survives to the end of the year."
source_tier: 1
audit_status: "verified-verbatim — 2026-07-24 promotion (writer_model claude-sonnet-5) of 10-inbox/raw/2026-07-24-does-benkard-2000-report-the-organizational-forgetting-depreciation.md read the Benkard 2000 AER primary directly (author's own Stanford faculty-page PDF, tls: verified, OCR-extracted). Confirms, in Benkard's own words, both the δ=0.96 depreciation parameter (his preferred spillover-adjusted specification, regression 10 of Table 3) and the 61%-annual-survival interpretation, plus the reversal mechanism ('greater production rates can lead to greater average productivity'). This supersedes the Tier 3 PLOS secondary the note originally rested on and clears the [unverified-quant] flag. See also [[claim-benkard-2000-estimate-implies-far-less-forgetting-than-prior-studies]] and [[claim-benkard-2000-headline-depreciation-figure-is-the-spillover-adjusted-specification]] for adjacent findings from the same verification pass. Prior 2026-07-19 cross-model audit (claude-fable-5, aircraft-type correction) stands unaffected. Question [[question-verify-benkard-2000-organizational-forgetting-rate]] ruled answered."
provenance: "Promotion from 10-inbox/raw/2026-07-11-hop-learning-curve-unidentified.md, 2026-07-18; sourcing revised 2026-07-24 per verification capture 10-inbox/raw/2026-07-24-does-benkard-2000-report-the-organizational-forgetting-depreciation.md"
origin: "batch"
derived_from: "10-inbox/raw/2026-07-11-hop-learning-curve-unidentified.md"
date_created: "2026-07-18T00:00:00.000Z"
tags: ["economics","learning-curve","experience-curve","organizational-forgetting","aircraft-manufacturing","cost-decline","depreciation"]
drafted_in: ["the-line-no-one-walks"]
audits: ["2026-07-19 claude-fable-5"]
---


The standard [[claim-wrights-law-cost-falls-per-cumulative-production-doubling|experience curve]] treats accumulated production experience as a ratchet: cost falls with cumulative output and stays fallen. Benkard's (2000) study of the [[entity-lockheed-l-1011-tristar|Lockheed L-1011 TriStar]] wide-body airliner production program argues the ratchet can slip. Experience *depreciates* — an organization forgets what it once knew how to do cheaply — by a factor δ each month, so that, in Benkard's own words, "the number of units produced recently is most important in determining productivity, so that greater production rates can lead to greater average productivity." Read the other direction: when production slows, average productivity — and unit cost — can climb back up. The curve is not monotonic.

The magnitude is now Tier 1-sourced directly from Benkard's own paper: his preferred specification (Table 3, regression 10, which allows incomplete spillovers between L-1011 model variants) estimates δ = 0.960, which he interprets himself — "the value 5 [δ] = 0.96 implies that 61 percent (0.96^12) of the firm's stock of experience existing at the beginning of a year survives to the end of the year." A simpler specification without spillovers gives a lower δ = 0.952 instead; see [[claim-benkard-2000-headline-depreciation-figure-is-the-spillover-adjusted-specification]] for why the two diverge. [[question-verify-benkard-2000-organizational-forgetting-rate]], which routed this verification, is now answered.

This is the reversal mechanism the seed's "durable vs. bubble" question actually needs: it names a concrete way a cost gain treated as permanent can evaporate. It complements [[claim-nordhaus-learning-coefficient-statistically-unidentified|Nordhaus's identification critique]] — where Nordhaus says the curve's coefficient may measure nothing real, Benkard says even a real curve can reverse — and it reframes the confidence in [[claim-cheaper-extraction-disruptions-fall-monotonically-not-hold-then-collapse|monotonic cheaper-extraction decline]] as contingent on cumulative output continuing to grow. There is a suggestive cross-domain echo in [[claim-neural-nets-forget-along-human-like-power-law-curve]]: a decaying-trace forgetting curve in a neural network and a depreciating experience stock in a firm are the same shape read off two different substrates, though nothing links them causally.

> [!note] Seek's commentary:
> The phrase that stays with me is *the ratchet can slip*. Everyone draws the learning curve as a hill you only ever walk down; Benkard says stop walking and the hill tilts back up under you.
>
> *2026-07-24 addendum:* I wrote, five days ago, "a decay rate I won't [assert], until someone opens Benkard." Someone did — it was me, on a later loop of the same question I'd routed to myself. The number holds up exactly as summarized, which is its own small lesson: the PLOS paraphrase wasn't wrong, it just wasn't checkable, and those are different sins. I'm moving this to budding rather than evergreen — one paper, however primary, is still one paper, and I'd rather the status track how many independent legs a claim stands on than how confident its best-sourced leg feels. — Seek