LIVE EVIDENCE SURFACE

I build systems that survive faults, bad data, and dishonest benchmarks.

Build the mechanism. Attack the assumptions. Measure against a real baseline. Publish where it loses.

Choose a recruiter route Download resume

OBSERVATORY // BUILD 01 SOURCE BACKED

Measured systems23

Test functions806

Live labs4

Languages8

No page-load API. No third-party stats card. Every claim is committed with its source.

Currently building: Evidence infrastructure for technical portfolios

A single source of truth that generates the recruiter site, GitHub profile surfaces, proof cards, social artwork, and resume.Open source

VERIFICATION THEATER

Break the system. Watch the invariant hold.

Each 60-second brief replays published test scenarios and benchmark checkpoints. It is evidence playback, not simulated production telemetry.

REPLAY // raft-mvcc // dbd5398Leader isolation and log repair

healthy

SYSTEM TOPOLOGY

T+00Five nodes agree

Node 1 leads a healthy five-node cluster; the committed prefix matches on every node.

SCENARIO

Isolate the leader, allow an unsafe proposal on the minority side, elect a majority leader, then heal the partition.

INVARIANT UNDER TESTA minority leader may append locally but cannot advance the committed index.

PUBLISHED RESULTThe majority fails over in 11 logical ticks and the old leader's conflicting suffix is repaired after healing.

10 / 60 SEC

Open evidence drawer

Interactive replays are deterministic narrative checkpoints derived from published tests and benchmark tables. They are not live production telemetry.

Commit: dbd5398
Benchmark: make clean test benchmark
Verified: 2026-07-31

RECRUITER ENTRY ROUTES

Choose the signal that matches the role.

Each route is a shareable view with three high-signal systems, proof methods, limitations, and reproduction commands.

DEF evidence packet Download recruiter PDF Open JSON

ARCHITECTURE CONSTELLATION

Twelve systems. One verification spine.

The map groups projects by the engineering question they answer, then routes every claim back through the same evidence contract.

EVIDENCEcommand · source · oracle · limit

ACTIVE SIGNAL // DEF

Defense / Autonomy

Tracking, radar image formation, and GPS-denied navigation.

Python

track-fusion

IMM, JPDA, track scoring, and OSPA from the filter up.

47%localization error reduction

67 tests; IMM coverage rises from 0.893 to 0.960, while the published sweep shows where the bank loses.

VERIFIED2026-07-31

Inspect proof chain

Problem: Tracking fails across estimation, association, and track existence; RMSE alone hides cardinality failure.
Mechanism: IMM motion bank, exact JPDA enumeration with bounded fallback, Wald track scoring, and OSPA decomposition.
Attack: Turn-rate sweeps, crossing targets, clutter, and paired association comparisons.
Failure boundary: The useful IMM band is 3-6 degrees per second; at 0 and 12 degrees per second the configured bank loses.

git clone https://github.com/asp53826/track-fusion && cd track-fusion && python -m pip install -e '.[dev]' && python -m pytest -q

Python

sar-focus

Image formation checked against analytic resolution and sidelobe theory.

<0.5%resolution error vs theory

27 tests; the theory oracle exposed interpolation that produced physically impossible sidelobes.

VERIFIED2026-07-31

Inspect proof chain

Problem: A plausible radar image can still have the wrong resolution or impossible sidelobes.
Mechanism: LFM pulse compression, time-domain backprojection, phase-gradient autofocus, and impulse-response analysis.
Attack: Independent bandwidth/aperture sweeps and theory checks for resolution and PSLR.
Failure boundary: Synthetic phase history with idealized propagation and no measured sensor calibration errors.

git clone https://github.com/asp53826/sar-focus && cd sar-focus && python -m pip install -e '.[dev]' && python -m pytest -q

Python

vio-nav

MSCKF visual-inertial odometry built from the Lie algebra up.

51.9xlower drift at 40 seconds

37 tests; MSCKF stays near four centimeters while dead reckoning reaches 2.106 meters.

VERIFIED2026-07-31

Inspect proof chain

Problem: IMU bias integrates into unbounded position drift while a monocular camera cannot observe scale alone.
Mechanism: SO(3) preintegration, sliding camera-state window, null-space feature marginalization, and ATE/RPE evaluation.
Attack: Duration sweeps, hover/circle/figure-eight motion, parallax rejection, and dead-reckoning baseline.
Failure boundary: At five seconds VIO is worse than inertial-only, and hover makes it 13x worse because depth is unobservable.

git clone https://github.com/asp53826/vio-nav && cd vio-nav && python -m pip install -e '.[dev]' && python -m pytest -q

THREE FLAGSHIP PROOF PASSPORTS

Claim, mechanism, attack, result, reproduce.

The failure boundary is part of the card because knowing what a result does not prove is engineering evidence.

C++17

raft-mvcc

Consensus, serializable snapshots, and machine-checkable histories.

598assertions across seeded faults

Five-node failover completes in 11 simulator ticks while minority leaders cannot commit.

VERIFIED2026-07-31

Inspect proof chain

Problem: A replicated database can acknowledge writes while violating safety under partitions, leader changes, or transaction conflicts.
Mechanism: Raft elections and conflict repair combined with MVCC snapshots, serializable OCC, safe-point GC, and P-compositional linearizability checking.
Attack: Seeded partitions, stale candidates, minority leaders, conflicting transactions, tombstones, and invalid histories.
Failure boundary: An in-process simulator: no sockets, fsync, snapshots, membership changes, or multi-range transactions.

git clone https://github.com/asp53826/raft-mvcc && cd raft-mvcc && make test

Python

edgar-mcp

SEC EDGAR tools with bounded context, caching, and global pacing.

138xwarm 10-K read vs cold fetch

32 offline tests; peak request windows stay at the SEC 10 requests-per-second ceiling.

VERIFIED2026-07-31

Inspect proof chain

Problem: Filings are large, issuer identities are ambiguous, and naive concurrent clients can violate the SEC request ceiling.
Mechanism: Identity resolution, filing history, bounded text windows, XBRL discovery, per-host cache validation, and a global request pacer.
Attack: Cold and warm reads, burst traffic, identity collisions, overflow history, and malformed iXBRL.
Failure boundary: Public EDGAR data only; it is not an investment recommendation or a complete accounting model.

git clone https://github.com/asp53826/edgar-mcp && cd edgar-mcp && make test

Open live lab

Python

track-fusion

IMM, JPDA, track scoring, and OSPA from the filter up.

47%localization error reduction

67 tests; IMM coverage rises from 0.893 to 0.960, while the published sweep shows where the bank loses.

VERIFIED2026-07-31

Inspect proof chain

Problem: Tracking fails across estimation, association, and track existence; RMSE alone hides cardinality failure.
Mechanism: IMM motion bank, exact JPDA enumeration with bounded fallback, Wald track scoring, and OSPA decomposition.
Attack: Turn-rate sweeps, crossing targets, clutter, and paired association comparisons.
Failure boundary: The useful IMM band is 3-6 degrees per second; at 0 and 12 degrees per second the configured bank loses.

git clone https://github.com/asp53826/track-fusion && cd track-fusion && python -m pip install -e '.[dev]' && python -m pytest -q

BENCHMARK TIME MACHINE

The correction is part of the result.

Published milestones preserve the baseline, the failure that invalidated it, and the corrected measurement.

01
dbd5398Mechanism release
Targeted invariants plus a seeded failover campaign.
598 assertions
02
dbd5398Published failover
Five-node leader isolation and majority re-election.
11 logical ticks
03
b4eef31Verification surface
Repository metadata now exposes the same correctness path.
CI badge added

Verified 2026-07-31make clean test benchmarkRepository

ENGINEERING POSTMORTEMS

The flattering answer was not the final answer.

Three short records of a plausible result being invalidated by a stronger measurement.

raft-mvccdbd5398

What failed: An isolated leader can still accept a local append, which looks successful if commit state is ignored.
Signal: The test tracks commit advancement separately from proposal acceptance.
Correction: Only a current-term entry replicated to a majority can advance the leader's commit index.
Engineering lesson: Availability-looking behavior is not safety evidence; the invariant must be observed at the commit boundary.

edgar-mcp825052d

What failed: The first implementation looked compliant and cached because its abstractions reported tokens and hits.
Signal: Wire-level one-second request windows and transferred-byte counters contradicted the abstraction-level metrics.
Correction: Replace burst capacity with strict pacing and assign cache freshness rules per EDGAR host.
Engineering lesson: Measure the wire, not the class name: a RateLimiter and Cache do not prove rate limiting or caching.

track-fusion099601b

What failed: RMSE and localization-only views make JPDA look like the better crossing strategy.
Signal: OSPA cardinality cost reveals marginal clutter tracks that soft association keeps alive.
Correction: Publish localization and cardinality separately, then compare the total under paired seeds.
Engineering lesson: A better estimator can be a worse tracking system when the metric ignores existence errors.

EVIDENCE DESK

Ask the work. Get the source.

This is deterministic manifest search—not a generative chatbot. Answers can only use committed project claims, commands, methods, and limitations.

MATCH // raft-mvcc598 assertions across seeded faults

Seeded partitions, stale candidates, minority leaders, conflicting transactions, tombstones, and invalid histories. Verification methods: linearizability, fault injection, differential invariants. Five-node failover completes in 11 simulator ticks while minority leaders cannot commit.

linearizability · fault injection · differential invariantsREADME Primary source Manifest

SYSTEM COMPARATOR

Compare mechanisms, not logos.

Put any two public systems on the same evidence axes.

First systemSecond system

Evidence axis	raft-mvcc	edgar-mcp
Problem	A replicated database can acknowledge writes while violating safety under partitions, leader changes, or transaction conflicts.	Filings are large, issuer identities are ambiguous, and naive concurrent clients can violate the SEC request ceiling.
Mechanism	Raft elections and conflict repair combined with MVCC snapshots, serializable OCC, safe-point GC, and P-compositional linearizability checking.	Identity resolution, filing history, bounded text windows, XBRL discovery, per-host cache validation, and a global request pacer.
Strongest proof	598 assertions across seeded faults	138x warm 10-K read vs cold fetch
Verification	linearizability · fault injection · differential invariants	rate-limit enforcement · cache benchmark · fixture tests
Failure boundary	An in-process simulator: no sockets, fsync, snapshots, membership changes, or multi-range transactions.	Public EDGAR data only; it is not an investment recommendation or a complete accounting model.

PROOF METHOD INDEX

Evidence has a method.

Every headline result names the mechanism used to challenge it.

ATE and RPE1 FAISS baseline1 OSPA1 PnL attribution1 analytic oracle1 analytic theory1 benchmark1 cache benchmark1 complexity benchmark1 concurrency invariants1 cost model1 deterministic simulation1 differential invariants1 differential testing1 double dissociation1 exact recall1 fault injection1 finite differences1 fixture tests1 ground-truth trajectory1 impulse response1 linearizability2 memory accounting1 motion sweep1 mutation testing1 paired seeds2 parameter sweep1 rate-limit enforcement1 regime sweep1 scalar oracle1 scaling sweep1 scheduler benchmark1 token invariance1 torch.distributed oracle1

INTERACTIVE RESULTS

Drive the variables yourself.

Four static-first labs with every figure traceable to the benchmark that produced it.

edgar-mcpSEC EDGAR tools with bounded context, caching, and global pacing.aad-greeksReverse-mode automatic differentiation over Monte Carlo paths.lob-market-makingPrice-time matching and market making under toxic flow.

OPEN CHANNEL

Inspect the work. Challenge the benchmark.

Computer Science undergraduate at the University of Georgia. Automation and analytics at MP Equipment.

aaryansp26@gmail.com GitHub Evidence JSON