Solana mainnetmeasured 2026-10-31 to 2026-11-04, after Alpenglow

When two transactions compete,
who goes first?

In all ten validators the higher-fee transaction came first in more than half of the usable pairs, p from 0.59 to 1.00, but never strictly. Four reached the preregistered N of 100: AgaveBam-1 "fee rule holds, partial", JitoLabs-1 "mixed", FireBAM-1 "fee rule holds, partial", HarmonicAgave-1 "fee rule holds, strong". The other six are descriptive, N below 100.

one leader block, one of our pairs
3,286pairs sent into leader windows
683usable conflicting pairs
10validators, 8 client types
4,493transactions landed (68%)

Three regimes

The answer depends on when the two arrive

Pooled over all validators, from results/secondary.json. p is the share of pairs in which the higher fee came first; q the share in which the first-sent came first.

Buffered togetherfee decides

p =0.85
105 of 124 pairs95% interval 0.77 to 0.91

Pairs sent before the window that landed in its first slot: the higher fee came first.

Sent 200 ms apartarrival decides

q =1.00
13 of 13 pairs95% interval 0.75 to 1.00

Two requests in random order: the first-sent came first.

Equal feescoin flip

p =0.47
25 of 53 pairs95% interval 0.33 to 0.61

A randomly labelled pseudo-high transaction came first about half the time.

Results by validator

Ten validators, eight client types, one block at a time

Dots are point estimates, bars are exact 95% intervals. The reference line is 0.5, the faint band is the preregistered "no rule" zone, 0.35 to 0.65. Sorted by N. Hover a dot for the count.

00.250.50.751
AgaveBam-1
AgaveBam 4.4.0-rc.1N = 196
fee rule holds, partial
p 0.81 [0.74, 0.86]q 0.61 [0.54, 0.68]
JitoLabs-1
JitoLabs 4.4.0N = 138
mixed
fee rule partial and request-position rule partial
p 0.64 [0.56, 0.72]q 0.78 [0.70, 0.84]
FireBAM-1
FireBAM 26.10.5N = 119
fee rule holds, partial
p 0.77 [0.69, 0.84]q 0.52 [0.43, 0.61]
HarmonicAgave-1
HarmonicAgave 4.4.0N = 110
fee rule holds, strong
p 0.94 [0.87, 0.97]q 0.61 [0.51, 0.70]
Rakurai-1
Rakurai 4.3.2, 4.4.0N = 37
descriptive (N < 100)
p 0.59 [0.42, 0.75]q 0.70 [0.53, 0.84]
Firedancer-1
Firedancer 26.10.5N = 27
descriptive (N < 100)
p 0.93 [0.76, 0.99]q 0.41 [0.22, 0.61]
HarmonicFiredancer-1
HarmonicFiredancer 26.10.5N = 27
descriptive (N < 100)
p 0.89 [0.71, 0.98]q 0.63 [0.42, 0.81]
Agave-1
Agave 4.3.2, 4.4.0N = 23
descriptive (N < 100)
p 0.61 [0.39, 0.80]q 0.83 [0.61, 0.95]
HarmonicFiredancer-1
HarmonicFiredancer 26.10.5N = 5
descriptive (N < 100)
p 0.80 [0.28, 0.99]q 1.00 [0.48, 1.00]
HarmonicFiredancer-1
HarmonicFiredancer 26.10.5N = 1
descriptive (N < 100)
p 1.00 [0.03, 1.00]q 1.00 [0.03, 1.00]

Analysis: final: phase main only, real data; one Holm family of 20 tests; family alpha 0.025. Hollow ticks mark the first analysis of 2026-11-01 (2,649 pairs), kept unchanged in results/first_analysis_2026-11-01/. Verdict rules (N of at least 100; strict, strong, partial, mixed, no rule) are fixed in docs/PREREGISTRATION.md. Source: results/verdicts.json.

Show as table
ValidatorClient, versionNp [95% CI]q [95% CI]Preregistered verdict
AgaveBam-1AgaveBam 4.4.0-rc.11960.81 [0.74, 0.86]0.61 [0.54, 0.68]fee rule holds, partial
JitoLabs-1JitoLabs 4.4.01380.64 [0.56, 0.72]0.78 [0.70, 0.84]mixed: fee rule partial and request-position rule partial
FireBAM-1FireBAM 26.10.51190.77 [0.69, 0.84]0.52 [0.43, 0.61]fee rule holds, partial
HarmonicAgave-1HarmonicAgave 4.4.01100.94 [0.87, 0.97]0.61 [0.51, 0.70]fee rule holds, strong
Rakurai-1Rakurai 4.3.2, 4.4.0370.59 [0.42, 0.75]0.70 [0.53, 0.84]insufficient data (n=37)
Firedancer-1Firedancer 26.10.5270.93 [0.76, 0.99]0.41 [0.22, 0.61]insufficient data (n=27)
HarmonicFiredancer-1HarmonicFiredancer 26.10.5270.89 [0.71, 0.98]0.63 [0.42, 0.81]insufficient data (n=27)
Agave-1Agave 4.3.2, 4.4.0230.61 [0.39, 0.80]0.83 [0.61, 0.95]insufficient data (n=23)
HarmonicFiredancer-1HarmonicFiredancer 26.10.550.80 [0.28, 0.99]1.00 [0.48, 1.00]insufficient data (n=5)
HarmonicFiredancer-1HarmonicFiredancer 26.10.511.00 [0.03, 1.00]1.00 [0.03, 1.00]insufficient data (n=1)

The delivery path

Where our pairs actually landed

Share of probes whose pair landed together in the target's own leader window, all arms (results/runs.md). Firedancer-family and Harmonic leaders (Firedancer-1, HarmonicFiredancer-1) included our RPC-sent pair in their own window in only 7% to 9% of probes, against 16% to 49% for the others. Most of their pairs landed one or more windows later, in other leaders' blocks.

AgaveBam-1AgaveBam
49%246 / 506
FireBAM-1FireBAM
45%137 / 302
JitoLabs-1JitoLabs
33%157 / 471
Rakurai-1Rakurai
25%40 / 160
Agave-1Agave
20%27 / 137
HarmonicAgave-1HarmonicAgave
16%130 / 815
HarmonicFiredancer-1HarmonicFiredancer
9%37 / 426
Firedancer-1Firedancer
7%31 / 469

Landing itself was not the problem: 4,493 of 6,572 main-run transactions landed (68%), high and low fee alike. Source: results/runs.md.

Depth of the work

Everything here is counted by the build script

Nothing on this page is typed by hand. The numbers come from the result files, the documents and the code at build time.

6,866 mainnet transactions sent 3,433 pairs in 44 runs, pilot included
10 validators, one per client type Agave, AgaveBam, FireBAM, Firedancer, HarmonicAgave, HarmonicFiredancer, JitoLabs, Rakurai
4 source-code research documents pinned commits of Agave, Jito, BAM, Harmonic, Firedancer
518 code citations re-checked every quote re-read at its path and line, all PASS
7 preregistration amendments each dated, each before the data it governs
48 on-chain anchors 5 protocol memos, 43 per-run memos; 222 files hashed
128 unit tests analysis, planning, wallet, spend cap, parsers
60/60 localnet known answers, arrival and 57/60 for priority order on Agave
94 gossip snapshots client claims recorded at every run start and end
0.0364 SOL spent of 0.05 cap 73% of the hard cap, enforced in code before every send
30/30 random landed transactions re-verified public RPC, result PASS

Method

One request, two transactions, one block

01

A conflicting pair

Two of our own transactions, from two payers, both writing one fresh shared account. The scheduler must put one before the other. The compute unit price differs 5,500x: 5,500,000 vs 1,000 micro-lamports per CU.

02

Randomized fee and request position

Which transaction is high-fee and which goes first in the request are drawn independently. Sent as one JSON-RPC request about 270 ms before the target's leader window; a separate arm sends two requests 200 ms apart.

03

Block producer read from the block

A pair is usable only if both transactions sit in one block produced by the target, read from the block's Fee reward and never assumed from the schedule, inside the targeted window.

04

Preregistered and anchored on mainnet

Estimands, thresholds, N of at least 100 and the equivalence margin were fixed before the data. Hashes of the plan and of the data were written to mainnet as memo transactions.

The six considerations

How the design handles each point of the bounty, shortened from docs/DESIGN.md.

01Leader schedule and leader types

Windows are 4 slots and 4-aligned (measured). Targets are frozen identities, one per client type, chosen by a stated rule that requires a stable gossip claim (same clientId and version over at least 18 hours) because claims flap: 33 of about 3,900 nodes changed claim within 80 minutes on 2026-10-30.

02Path from our backend to the executing node

Our VM cannot send UDP or QUIC, so every transaction goes as an HTTPS JSON-RPC request to one provider.

03Interaction of our own transactions with each other

Within a pair the interaction is the design (shared W in arms P, E, S; none in arm D). Across pairs there is none by construction: one probe per window, a fresh W per probe, distinct payers, a payer reused only after confirmation or expiry, at least 2 s between probes.

04Competition inside the block

Other transactions compete with ours. Ours sit high in the priority queue (about 1.1 and 3.4 lamports per cost unit, while 18% to 50% of transactions pay no priority fee), so the leader can schedule both promptly; the order between them is still the leader's decision.

05Send order

The position of the high-fee transaction in the request is randomized independently of the fee assignment, giving q as a second estimand. The spaced arm sends two requests 200 ms apart in random order, the only arm with a known send order, to test for an arrival rule and for overtaking.

06Noise versus rule

Pre-specified estimands and verdict thresholds; an equal-fee arm as the noise floor; an equivalence test (90% intervals of both p and q inside [0.35, 0.65]) as the only route to "no rule detected"; Holm correction over all tests; synthetic data with planted rules and two localnet known answers as checks on the analysis code.

Timeline

Dated, in order, anchored

Every anchor is a memo transaction on mainnet carrying the SHA-256 of a file in this repository. Compare it with sha256sum of the file at the named commit.

  1. 2026-10-21 13:54 UTC
    Alpenglow boundary
    last block with vote transactions 459652999, first block after it 459653008 (docs/ALPENGLOW_BOUNDARY.md); all our data is above it
  2. 2026-10-30
    Localnet calibration on Agave 4.3.2
    docs/CALIBRATION.md
  3. 2026-10-30 17:09:26 UTC
    Preregistration v1.2 anchored
  4. 2026-10-30 17:10 to 18:16 UTC
    Pilot (58 probes) and re-pilot (89 probes)
    provisional targets, operational metrics only (Amendments 1 to 3)
  5. 2026-10-31 08:37:09 UTC
    Targets and plan frozen, manifest anchored
  6. 2026-10-31 08:38 to 22:31 UTC
    Main measurement (Amendments 4 to 6)
  7. 2026-11-01
    First confirmatory analysis (archived in results/first_analysis_2026-11-01/)
  8. 2026-11-04 08:3x UTC
    Amendment 7 anchored before any new data
  9. 2026-11-04 08:34 to 19:56 UTC
    Extension of the identical protocol (Amendment 7); pooled analysis at family alpha 0.025
  10. 2026-11-01
    Data manifest after the first period anchored
  11. 2026-11-04 20:4x UTC
    Final data manifest anchored
    data/anchors/data_manifest.txt

Verify in 60 seconds

Do not trust this page. Run it.

Pick a random sample of our landed transactions, fetch the blocks from a public RPC and confirm the recorded positions and producers. No wallet, no key, nothing is sent.

Clone
git clone https://github.com/recruitabx/Inside-The-Solana-Block.git
cd Inside-The-Solana-Block
Re-check our transactions against mainnet blocks (public RPC, no key)
docker compose run --rm verify
Without Docker
python3 -m venv .venv && . .venv/bin/activate && pip install -e . && itb verify

Last sample: 30 of 30 random landed transactions PASS, results/verify_sample.txt.

Spend36,440,074 of 50,000,000 lamports (0.0364 of 0.05 SOL)

The cap is enforced in code before every send; fees settle only for transactions that land. Audit: PASS, every fee-paying transaction of our accounts is in the spend ledger.

Honest limits

What we do not separate

The full list of problems, in order, is in What went wrong in the report.