Skip to the project journey

Three-day technical timeline · 29–31 August 2026

Three days. One defensible detector.

We began by deciding what evidence to trust. We ended with a protected evidence package, rejected shortcuts, a selected routed residual product, and a roadmap honest about what is still unproven.

  1. Day 1Choose what to trustBackbones · ROC-AUC · protected baseline
  2. Day 2Design for the real internetDistortions · residual heads · native-size routing
  3. Day 3Reject shortcuts. Build the product.Domain guardrails · integration · roadmap
Day 1Saturday · 29 August

Choose the metric, then protect the answer.

Question: Which representations look promising—and how do we stop ourselves from grading on the test?

Backbone exploration, metric selection, and a protected baseline

We screened several visual representations, then chose ROC-AUC as the primary ranking metric. The close exploratory scores were a direction finder—not a final leaderboard.

Evidence before excitement

Small score gaps need paired confidence intervals and an attached protocol before they can support a model-selection claim.

Team-recorded exploratory screen

Useful direction. Not protected-final evidence.

Descriptive
MoCo v3 TransformerNot selected
0.706
FatFormer
0.771
CLIP-RN50x64
0.828
DINOv2-L
0.836
DINOv3-7*Scope caveat if 7B
0.839
DINOv3-LHigh cost
0.842

*The recorded “DINOv3-7” label needs scope confirmation if it refers to a 7B model. Protocol details for this exploratory screen remain to be attached.

Why ROC-AUC?

An AUC of 0.842 means the system ranks a random generated image above a random real image about 84.2% of the time. It does not mean 84.2% accuracy at one threshold.

10,002balanced images5,001 real · 5,001 generated
2,004calibration rowsthreshold and development only
7,998protected final rowsduplicate-aware · hash-audited

Historical V1 protected-final result

Move the operating point—not the ranking model.

Historical V1
Threshold 0.50.7763balanced accuracy · recall 0.5924
Threshold 0.287470.8061balanced accuracy · recall 0.7127

The lower threshold caught 481 additional generated images and produced 238 more correct decisions overall. Precision moved from 0.9371 to 0.8764 and specificity from 0.9602 to 0.8995. ROC-AUC stayed 0.8505 because the ranking scores never changed. This was an explicit review tradeoff—not free accuracy—for the retired four-expert runtime. It is not the decision rule for the selected TEST1 graph.

Day 2Sunday · 30 August

Stress the pixels, then choose complementary views.

Question: What survives compression, resizing, screenshots, and unfamiliar generators?

01JPEG Q40−0.0306ROC-AUC change
02Half downscale−0.0239ROC-AUC change
03Screenshot-like−0.0198ROC-AUC change

Retrospective stress tests on the 2,004 calibration rows. Distortion losses varied by source dataset, so the pooled changes are not universal robustness guarantees.

Released architecture

One frozen encoder. Three residual heads.

Released
Expert 4SigLIP · 384 × 384
  • Frozen upstream Expert 4 checkpoint
  • 27 transformer blocks
  • 1,152-D pooled feature
  • Two-logit teacher margin
Residual route3 × 297,729 params
  • ≤64 px uses CIFAKE head at α 1.25
  • >64 px blends epoch 05 / 08
  • Fixed weights 0.65 / 0.35
  • Frozen margin boundary -1.55796

Tu et al. supplied Expert 4 and its detector training. SynthFlag does not claim to originate or train that checkpoint. The residual heads and routing are project artifacts with a collaborator rights attestation accepted by the project owner. Expert 4 redistribution and organizer eligibility remain separate.

A library is not a training policy.

12basic operations

19active extended operations

≈35candidate functions

The proposed 20% clean / 40% basic / 40% extended mix is a training design only when connected to a reproducible pipeline. Not every candidate transform reached that state during the sprint.

Selected model · Exact routed graph

Follow one image through your project-trained heads.

Native image size chooses one of two deterministic routes over a frozen Expert 4 representation. The contribution is the three residual heads, their training/evaluation stack, and this fixed routing policy—not the upstream teacher.

Selected SynthFlag TEST1 graph with frozen Expert 4 and three routed residual heads
Native longest side ≤64 uses the CIFAKE specialist at alpha 1.25; larger images use the fixed 0.65 / 0.35 two-head blend and boundary −1.557959395647049.
Need the exact route and boundary?Open the selected technical appendix and score derivation.
Day 3Monday · 31 August

Separate the selected heads from rejected shortcuts.

Question: Which adapter experiments failed—and what does the selected graph actually ship?

Historical four-expert adapter experiment

The pooled score rose. The transfer guardrails fell.

Rejected
Trainable parameters986,120≈0.0673% of 1.465B total
Grouped pooled OOF0.8661 → 0.9462tempting, but insufficient
Held-out CIFAKE−0.0553
Held-out WildFake−0.0465
Held-out SID-Set+0.0154
Mean held-out change−0.0288

Decision: do not promote this adapter. This older experiment belonged to the retired four-expert line; it is not one of the three selected residual heads. It learned shortcuts that looked excellent in pooled development but did not travel reliably. A separate WildFake-excluded run has training OOF numbers, but its actual holdout report is unfinished and is not a generalization result.

Product integration

Then turn evidence into a usable review tool.

Verified artifacts

Checkpoint hashes, model card, rights notes, release audit.

Reliable inference

FastAPI service, resumable batches, serialized accelerator jobs.

Judge-facing product

Upload workflow, architecture visuals, and technical appendix.

Deployment is separate

Service connectivity is a live operational state, not model evidence.

Roadmap · Questions worth testing

The next ideas stay visibly on the research side.

Proposed

Local camera statistics

Professor Ng Teck Khim’s interview prompted a forensic line of inquiry: examine 10 × 10 blocks, cross-channel correlations, and local variance associated with camera acquisition, Bayer sampling, and demosaicing.

Early prototype not stable enough for final use; no metric claimed.
Source complete

Sampled-video delivery

The source now extracts eight midpoint frames in the browser, scores them in two-frame microbatches, and returns a visible timeline with descriptive summaries. Public availability still depends on a connected inference worker.

No temporal understanding is claimed.

A staged adaptation ladder—heads, then LayerNorm, LoRA, and final blocks—was proposed for the retired four-expert line. Its head-only adapter was rejected; that result is separate from the three selected residual-head artifacts now shipped.

End-of-sprint truth

Shipped, unfinished, and deliberately not claimed.

Released

Routed residual image inference

Frozen Expert 4 → three residual corrections → native-size route → sigmoid score.

Measured

TEST1 public diagnostic

15,000 public sources were evaluated clean and under one deterministic composite view at reported threshold 0.5.

Incomplete

27,265-image benchmark

Extraction work does not equal a final benchmark. No final independent AUC is claimed here.

Blocked

V3 evaluation

The exact source needed for the planned evaluation was unavailable, so the metric remains unreported.

Score signal—not proof. SynthFlag is a triage aid for human review. It does not prove authorship, localize manipulation, or replace provenance and contextual evidence.

Continue with one clear choice

Try the product—or audit the technical details.

Try SynthFlag Open technical appendix