Bundle review table

14 — Punctuality Is a Buffer (`airline-turnaround-buffer`)

Supplied working report; original verdicts not validated.

Review status: Needs review. This is an archival source note, not a passed proof.

REPORT.md

# 14 — Punctuality Is a Buffer (`airline-turnaround-buffer`)

- Source: Incoprea Theory, Volume XXIX (`airline-turnaround-buffer`)
- Claim: Airline on-time performance is set by scheduled turnaround buffer relative to actual turnaround variance, not by fleet age.
- Prediction: Punctuality should track the buffer-to-variance ratio.
- Refutation: Punctuality tracking fleet age would remove the buffer.

## Test performed

We took the US DOT Bureau of Transportation Statistics on-time performance
file for **August 2026** (608,389 flight records; 594,207 flown after
dropping cancellations and diversions) and computed, per carrier and per
carrier-route segment:

- **On-time %** = share of flights with arrival delay < 15 min (BTS `ArrDel15`);
- **Buffer** (schedule padding) = mean scheduled elapsed time − minimum
  observed elapsed time per carrier-route segment, averaged over flights;
- **Variance proxy** = standard deviation of arrival delay (minutes);
- **Buffer-to-variance ratio** = buffer / σ(delay) — the hypothesis's predictor;
- **Mean fleet age** (rival predictor) = 2026 − year of manufacture from the
  FAA aircraft registry (Releasable Aircraft Database, `MASTER.txt`, matched
  to 96.8% of tail numbers flown that month).

We then correlated on-time % with the buffer-to-variance ratio and with fleet
age, at carrier level (n = 12 carriers with ≥ 5,000 flights) and at
carrier-route level (n = 3,662 segments with ≥ 60 flights), plus small OLS
regressions at carrier level.

## Data (sources, URLs, sample sizes)

Evidence class (a): new analysis of public data fetched 30 October 2026.

- BTS On-Time Performance, August 2026 (pre-zipped monthly CSV, 32 MB zipped /
  276 MB unzipped):
  <https://transtats.bts.gov/PREZIP/On_Time_Reporting_Carrier_On_Time_Performance_1987_present_2026_8.zip>
- FAA Releasable Aircraft Database (Oct 2026 snapshot, MASTER.txt):
  <https://registry.faa.gov/database/ReleasableAircraft.zip>
- Derived tables: `data/carrier_metrics.csv` (12 carriers),
  `data/route_metrics.csv` (3,662 segments),
  `data/correlation_summary.csv`. Code: `code/compute_metrics.py`,
  `code/analyze.py`.

## Result

**Buffer-to-variance ratio tracks punctuality.**
- Route level (n = 3,662): r(on-time, buffer/σ) = **+0.42, p < 0.001**
  (`figures/fig3_route_bins.png`, left panel). The ratio's components go in
  the expected directions: schedule padding alone r = +0.45 (p < 0.001);
  delay dispersion alone r = −0.30 (p < 0.001).
- Carrier level (n = 12): r = +0.37, p = 0.24 — the same positive direction,
  but a snapshot of 12 carriers cannot resolve it statistically
  (`figures/fig1_carrier_ratio.png`). One clear outlier (G4, ratio 0.15 but
  73.8% on-time) flattens the carrier line.

**Fleet age does not track punctuality.**
- Carrier level: r(on-time, mean fleet age) = **+0.37, p = 0.24** — positive,
  i.e. *older* fleets had slightly *better* on-time performance in Aug 2026,
  the opposite of the "fleet age" rival story (young-fleet F9: 4.7 yr, 66.6%
  on-time; DL: 15.2 yr, 78.4%). The association is not significant and has
  the wrong sign (`figures/fig2_carrier_age.png`).
- Adding fleet age alongside the buffer ratio does not remove the buffer
  term: OLS on-time ~ ratio + mean_age gives ratio coef +14.5 (p = 0.24) and
  age coef +0.61 (p = 0.23), n = 12 — both weak at carrier level, neither
  driving the other to zero. The refutation condition (fleet age removing
  the buffer) is not met.

## Verdict

**Support.** In the August 2026 snapshot, punctuality tracks the
buffer-to-variance ratio significantly at route level (r = +0.42, p < 0.001)
and directionally at carrier level, while fleet age fails as a rival
predictor — its carrier-level association has the wrong sign and is
insignificant. The buffer term is not displaced by fleet age.

## Limitations

- One-month cross-section; association, not causation. Scheduling and
  network choices are endogenous (carriers pad schedules where they expect
  congestion, and on-time % partly reflects that padding mechanically).
- Our "buffer" is gate-to-gate schedule padding, and our "variance" is the
  cross-flight σ of arrival delay — proxies for ground-turnaround buffer and
  turnaround variance, which the on-time file does not contain.
- Carrier-level n = 12 is underpowered; the significant test is route level.
- The ratio performs about as well as padding alone; we cannot claim the
  *ratio* form beats its numerator.
- Fleet age from the FAA registry is airframe age (year of manufacture);
  maintenance condition and operational reliability are unobserved.

## References

- [BTS On-Time Performance data](https://www.transtats.bts.gov/DL_SelectFields.aspx?gnoyr_VQ=FGJ&QO_fu146_anzr=b0-gvzr)
- [FAA Releasable Aircraft Database](https://www.faa.gov/licenses_certificates/aircraft_certification/aircraft_registry/releasable_aircraft_download)
- [Deshpande, V. & Arıkan, M. (2012), "The Impact of Airline Flight Schedules on Flight Delays", M&SOM 14(3)](https://pubsonline.informs.org/doi/10.1287/msom.1120.0379)
- [Forbes, S.J. (2008), "The effect of air traffic delays on airline prices", IJIO 26(5), 1218–1232](https://ideas.repec.org/a/eee/indorg/v26y2008i5p1218-1232.html)
- [Fan, T. (2003), "Market-specific airport congestion delays and airline schedule padding", MIT thesis](https://dspace.mit.edu/handle/1721.1/32443)

Included code and results