Video ideation · @DrJamesCroft · 6 August 2026

Dr. James Croft — 9 video ideas, graded by evidence

Version 3 — rebuilt on peer-proven STRUCTURES, not just peer-proven topics. Version 2 fixed the evidence for the topics but left eight cards resting their structure claim on a single video from this channel's own first four weeks. Those two structures were then measured across the peer set and found zero times. Every card below now uses one of four title families measured across 17 peer channels (1.5K–126K subs), with the occupied slots named and rotated away from. Two cards survive from v2; eight are new; one more was cut by the new mechanical evidence gate.

subs 986videos 12avg views 21,156peer channels measured 17peer sub range 1.5K–126Kpeer outliers classified 42proven families 4survived from v2 2classification faceless persona-led single-discovery mini-docnews-driven yesdata via yt-api (Algrow quota exhausted)shipped 9cut by the evidence gate 1

How the score works

adjusted  =  raw  ×  evidence quality  ×  structure conformance

Evidence quality is graded from what was actually found: strong ×1.00 (two or more real videos, best above 150K) · partial ×0.75 (two or more, best above 35K) · thin ×0.45 (anything less). Structure conformance drops below 1.00 when the final title rewrote part of the proven structure instead of only filling its slots — the precedent then only partly applies. Every card shows its chain, so you can see exactly what was applied.

Ranking

Title-structure bank

This is the measured peer skeleton bank: every outlier in the top 5 of all 17 peer channels, classified into structural families. Outliers are computed against each peer channel's own median. Nothing is graded against a channel above 250K subs. The two structures this channel currently runs are at the bottom — they produced no peer outliers at all.

Show the six structures3 in use · 3 exhausted
Title structureProved byViewsOutlierUsed
{FAMOUS ARTIFACT} Was Reanalyzed / Decoded by AI — And the Findings Are {Hard to Explain}Stonehenge Was Reanalyzed by AI (Hidden Truths Uncovered, 52K subs)1,800,1896.5×3 ideas — 18 outliers across 10 of the 17 peer channels; pays from 5.35K subs to 52K
{Divers / An Underwater Drone} {Reached / Entered / Went Inside} {NAMED PLACE} — What {It Found}Divers Found Pharaoh's Army Beneath the Red Sea (QUANTARA, 48.1K subs)1,800,0005.0×4 ideas — 9 outliers across 5 peer channels, including 49.6× and 41.6× on a 4.6K-sub channel
{NAMED SITE / TOMB} Was Opened After {N} Years — What They Found InsideArchaeologists Just Found a Sealed Room in Petra (The Insight Archive, 32.2K subs)2,500,00021.0×2 ideas — 7 outliers across 6 peer channels
{At AGE / Before He Died}, {NAMED PERSON} FINALLY Reveals {WHAT}At 69, Jeremy Wade FINALLY Reveals Why River Monsters Was Cancelled (The Insight Archive, 32.2K subs)884,0007.4×1 idea — 7 outliers across 6 peer channels, including a 207× on a 7.1K-sub channel
They Found {X}. Then They Stopped Digging.They Found the World's First City. Then They Stopped Digging. (The Origin Vault, 22.1K subs)174,0003.2×not used — only 1 outlier on 1 peer channel; below the 2-channel bar, but worth watching
{SUPERLATIVE THING} Sat Untouched for {N} Years — Then {ACTOR} Finally Opened Itthis channel only (87K, 4.1×)87,0234.1×REMOVED — searched across the niche, the only discovery channel running this phrasing is @DrJamesCroft itself. Everything else it surfaces is abandoned-house urbex and romance-drama. Not a proven structure; a one-off that worked once on n=12.
{ORDINARY_ACTOR} Found {SPECIFIC_NUMBER} {THING} — Nobody Ever Noticedthis channel only (67K, 3.2×)66,6583.2×REMOVED — a peer-scale search collapses into a pop song called 'No One Noticed' and Disney-detail listicles. Zero peer outliers. See the findings: peers name the entity, they do not anonymise the finder.

The ideas

Demand = how much the topic is watched · Saturation = how crowded this structure×topic is (lower is better) · Recency = is it popping right now · Season = fit to the current window · Novelty = confidence no video exists at this structure×topic. Every structure below is proven on at least two independent peer channels; every card names the fills that are already occupied and rotates away from them.

01

An Underwater Drone Reached the Andros Wreck — 600 Amphorae Still Stacked Where the Crew Left Them

0.880 0.880×1.00
evidence strongnews-peggedSubstitute
Demand82
Saturation30
Recency92
Season70
Novelty0.95

Why: the only structure in the study that is proven on a full ladder of channel sizes — 4.3K, 8.4K, 48.1K and 52K subs, all paying, on five different wrecks. Bolted to a find announced 31 July that nobody has covered.

Thumbnail Overlay '600 STILL STACKED' — an ROV light raking across rows of amphorae in their original cargo pattern, necks all pointing the same way. Emotion: nobody has touched this arrangement in 2,300 years.

Occupied fills — never re-run these: SS Edmund Fitzgerald, Jacob's Well, Titanic, USS Yorktown, HMS Hood, U-47, Red Sea 'Pharaoh's Army'. Honest caveat: there is no drone footage of Andros yet. Build it on the survey imagery and the amphorae, and write it as the survey — not as a dive that did not happen.

Evidencestructure 1.9M, 1.6M, 706K · topic 210K, 69K, 40K, 275K · news source

Title-structure evidence

Topic evidence — drone reaches a named wreck · Andros

02

The Nazca Lines Were Reanalyzed by AI — And It Found Shapes Nobody Had Counted

0.860 0.860×1.00
evidence strongSubstitute
Demand88
Saturation35
Recency88
Season55
Novelty0.90

Why: by a wide margin the strongest family in the peer set — 18 outliers across 10 of the 17 channels, and it works at every size in the band. Stonehenge and the Shroud of Turin are now taken, so this rotates to the Nazca Lines, where the best existing video is 59.9K. Aerial geoglyphs are also the most renderable subject available.

Thumbnail Overlay 'IT COUNTED MORE' — the Nazca hummingbird from above with a faint machine-vision grid over it, several unfamiliar shapes ringed in the empty desert alongside. Emotion: it was always there and nobody saw it.

This card reverses my earlier advice. I originally hard-blocked this structure because this channel's own attempt did 701 views — a video published on its third day of existence, with neighbouring uploads reading 66,658 / 701 / 25,526 / 14,773 / 1,360 inside eight days. That is noise. I then tested my own follow-up theory that AI-decode needs a famous named artifact and disproved it (anonymous-artifact median 79K vs 28.5K). Other open fills: Göbekli Tepe (176K), Baalbek (149K), Voynich (176K).

Evidencestructure 1.8M, 707K, 315K · topic 755K, 627K

Title-structure evidence

Topic evidence — AI reanalysis · Nazca Lines

  • AI Found Something Impossible in the Shroud of Turin755K views · Galaxy Vault (48.6K subs) · 5.8× · 21:17 — same structure, rotated artifact
  • Stonehenge Mystery Finally Solved by AI and the Truth Is Terrifying627K views · The Silent Archive (23.5K subs) · 33.9× — the highest outlier in the AI family
03

Scientists Entered the Melt Cave Under the Rhône Glacier — And the Ice Handed Back a Man Missing 37 Years

0.830 0.830×1.00
evidence strongnews-peggedSubstitute
Demand78
Saturation30
Recency82
Season95
Novelty0.95

Why: the v2 version of this idea used the channel's own unproven 'A Hiker Found…' structure. Rebuilt on the entered-a-named-place family, which is peer-proven at 49.6× and 41.6× on a 4.5K-sub channel — the closest thing in the study to this channel's situation. Naming the glacier is the part that matters; 'a hiker' is exactly what the peers never write. Hard calendar: the Jul–Sep melt window closes at the autumn re-freeze.

Thumbnail Overlay 'THE ICE GAVE HIM BACK' — a headlamp beam inside a blue melt cave, a boot and a weathered hand emerging at the melt line, meltwater running past. Emotion: intimate dread, not gore.

Occupied fills: the Amazon forbidden zone and Lake Tahoe (both Beyond discovery). Competitive warning: two channels launched into frozen-discovery within the last eight days — @IceMysteries (3 videos) and @The.IceArchive (a dormant 16-year-old channel rebranded for it). Sourcing: pick a real named glacier with a real recovered-remains case; do not invent the 37-year figure.

Evidencestructure 273K, 229K, 519K · topic 2.0M, 318K, 434K, 54K · news source

Title-structure evidence

Topic evidence — entered a named place · glacier melt

04

An Underwater Drone Reached a Sealed Porcelain Cargo 2,000 Feet Down off Norway — Nothing Was Broken

0.760 0.760×1.00
evidence strongnews-peggedSubstitute
Demand68
Saturation25
Recency72
Season70
Novelty1.00

Why: the v2 version of this used the channel's own unproven 'Sat Untouched…' phrasing. Same subject, moved onto the drone ladder. 'Nothing was broken' is the specific hook — an intact porcelain cargo 2,000ft down is a physically improbable image, which is what this format needs.

Thumbnail Overlay 'NOTHING WAS BROKEN' — a submersible's light hitting stacked blue-and-white porcelain still in its crate, silt drifting off, every plate whole. Emotion: disbelief that it survived.

Third card on the drone family. That concentration is deliberate — it is the best-evidenced structure available — but space these across several weeks so your own thumbnails do not compete. Same footage caveat as card 1: build it on survey imagery, not on a dive that has not been filmed.

Evidencestructure 1.9M, 210K · topic 275K, 197K · news source

Title-structure evidence

Topic evidence — drone reaches a named wreck · Norway porcelain

05

Göbekli Tepe Was 3D-Scanned by AI — And the Enclosures Do Not Line Up the Way We Were Told

0.720 0.720×1.00
evidence strongSubstitute
Demand74
Saturation45
Recency80
Season55
Novelty0.75

Why: the same 18-outlier family, on a marquee site that is genuinely in the channel's declared territory (pre-agricultural Anatolian temples are named in its About page). The megalith imagery renders as well as Stonehenge does.

Thumbnail Overlay 'THEY DON'T LINE UP' — the T-pillars from directly overhead as a wireframe scan, with the enclosure axes drawn as lines that visibly fail to converge. Emotion: the diagram contradicts the caption.

Semi-occupied. ChronoFall already holds this fill at 176K — beatable, but it is not the clean open slot Nazca is. Rank it below card 2 and produce it second if the Nazca video performs.

Evidencestructure 1.8M, 755K, 707K · topic 176K, 315K

Title-structure evidence

Topic evidence — AI reanalysis · Göbekli Tepe

  • AI 3D-Scans FINALLY Exposed Göbekli Tepe's 12,000-Year Secret176K views · ChronoFall · the current ceiling on this fill — semi-open, beatable
  • AI FOUND an Impossible Signal in the Shroud of Turin315K views · Vectara (5.35K subs) · 13:12 — the family pays at 5.35K subs
06

A Sealed Egyptian Funerary Complex Was Opened After 3,000 Years — What Was Inside the Nile Delta Tombs

0.680 0.680×1.00
evidence strongnews-peggedSubstitute
Demand70
Saturation55
Recency90
Season60
Novelty0.85

Why: the freshest news peg in the set — three days old — moved from the channel's own unproven phrasing onto the peer-proven 'Opened After N Years' family, which is the structure The Silent Archive rode to 15.6× at 23.5K subs.

Thumbnail Overlay 'SEALED 3,000 YEARS' — a hand-held lamp throwing light through a just-broken mudbrick seal into a painted chamber, dust hanging in the beam. Emotion: the first look.

Occupied fills: Petra (twice), the Amazon sealed chamber, King Henry VIII's tomb, King Solomon's tomb, the Vietnam cave. Egypt is also the most-covered subject in the niche (saturation 55), so this rides on recency — ship while the find is days old or drop it.

Evidencestructure 2.5M, 289K, 371K · topic 261K, 1.1M · news source

Title-structure evidence

Topic evidence — named site opened · Nile Delta

07

Divers Found the Sunken City of the Pharaohs Beneath Abu Qir Bay — And It Is Still Standing

0.360 0.800×0.45
evidence thinSubstitute
Demand72
Saturation35
Recency58
Season65
Novelty0.90

Why: the Egypt-underwater fill is proven twice independently — QUANTARA at 1.8M and The Sealed Scroll at 292K, both on 'Divers Found Pharaoh's… beneath the Red Sea'. Thonis-Heracleion is the same emotional promise on a different named body of water, which is exactly the rotation the ladder rewards.

Thumbnail Overlay 'A CITY, 30 FEET DOWN' — a diver hovering beside a colossal submerged granite head half-buried in seabed, scale making the diver tiny. Emotion: awe at the scale.

The Red Sea 'Pharaoh's Army' fill is taken twice over — do not use it. Abu Qir Bay is the open rotation. Note the subject overlaps with the 'Pharaoh' framing enough that thumbnails must look clearly different from QUANTARA's.

Evidencestructure 1.8M, 292K, 229K · topic 399K, 1.8M

Title-structure evidence

Topic evidence — divers found a named submerged place · Thonis-Heracleion

08

Baalbek's Trilithon Was Reanalyzed by AI — And the Weight Numbers Stopped Making Sense

0.288 0.640×0.45
evidence thinSubstitute
Demand66
Saturation45
Recency72
Season55
Novelty0.75

Why: third rotation of the strongest family. Baalbek's megaliths are the most physically absurd artifact available — a 1,000-tonne block still in the quarry — which suits a format that needs a single image to carry the thumbnail.

Thumbnail Overlay '1,000 TONNES' — the Stone of the Pregnant Woman still lying in the quarry with a human figure at its end for scale, measurement lines overlaid. Emotion: this should not have been movable.

Weakest of the three AI cards and the fill already has a 149K holder. Produce it only after Nazca and Göbekli Tepe, and only if they work. Sourcing discipline: Baalbek attracts ancient-aliens content; stay on the engineering question or the channel inherits that audience.

Evidencestructure 1.8M, 707K · topic 149K, 315K

Title-structure evidence

Topic evidence — AI reanalysis · Baalbek

09

At 79, Zahi Hawass Finally Answered What Is Behind the Sealed Door in the Great Pyramid

0.234 0.520×0.45
evidence thinAdapt
Demand76
Saturation50
Recency65
Season55
Novelty0.70

Why: the fourth peer-proven family, and the one containing the single biggest outlier in the whole study (207× on a 7.1K-sub channel). It is also the family that fits an archaeology channel most naturally, because this niche has genuinely famous living figures.

Thumbnail Overlay 'THE DOOR' — the Gantenbrink door blocking stone at the end of the southern shaft, lit by a robot camera, with a red outline around the copper pins. No portrait of Hawass on the thumbnail. Emotion: the object, not the man.

Ranked last deliberately — this is the highest-risk card on the board. It names a real, living, litigious public figure, and the peer examples lean on framings ('Before He Dies…') that are borderline clickbait about real people's mortality. Only produce this from statements Hawass has actually made on the record, cite them on screen, and never imply an exclusive interview. If that discipline is not going to hold, cut the card — a fictional PhD persona putting words in a real archaeologist's mouth is the one failure mode this channel cannot recover from.

Evidencestructure 1.4M, 884K, 207K · topic 524K, 261K

Title-structure evidence

Topic evidence — named person reveals · Egyptology

  • Grok AI Was Asked Who Built the Pyramids - The Answer Shocked Everyone524K views · New Discovery (28.4K subs) · 5.1× — Great Pyramid interior questions at peer scale
  • What was found inside King Tut's Antechamber?261K views · HistoryFeels (228K subs) · 5:03

What the evidence changed

The evidence gate cut a card from this very report

The mechanical gate added to the skill after this run (_shared/evidence_gate.py) was pointed at this file and immediately blocked one card: 'The Empire's Largest Frontier Bathhouse Was Opened After 2,000 Years'. Its only topic proof was Odyssey - Ancient History at 1.04M subs — outside the peer band, so subject interest rather than transferable demand. No peer-band proof for Roman bathhouse discoveries exists, so the card is cut rather than propped up. That is the gate doing its job on its own author, which is the only real test of it.

Correction 5 — the structures this channel runs are not proven anywhere except on this channel

Version 2 fixed the topic evidence but left eight cards claiming their structure worked on the strength of one video from this channel's first four weeks. Both were then searched across the niche. '{THING} Sat Untouched for {N} Years — Then {ACTOR} Finally Opened It': the only discovery channel using this phrasing is @DrJamesCroft. The rest of the results are abandoned-house urbex (Bros Of Decay, 699K subs) and, oddly, romance-drama channels. '{ORDINARY_ACTOR} Found {NUMBER} {THING} — Nobody Ever Noticed': the search collapses into a pop song called 'No One Noticed' and Disney-detail listicles. Zero peer outliers between them. Both are now retired from the bank.

The pattern behind it: peers NAME the entity, this channel anonymises it

Look at what actually wins in the peer set — Petra · King Solomon's Tomb · King Henry VIII · Jacob's Well · Lake Tahoe · SS Edmund Fitzgerald · Titanic · Stonehenge · the Shroud of Turin · Jeremy Wade · Charles Duke. Now look at this channel — a Student · a Farmer · a Roman Fort Ditch · a Scroll · Construction · 3 Students. Every proven family in the bank puts a recognisable proper noun in the title, because that is what a viewer can decide to care about in the half-second the thumbnail is on screen. I earlier told you the non-expert finder was the load-bearing device; the peer data says the opposite — the named marquee entity is load-bearing, and anonymising it is the more likely reason this channel's titles underperform. Every title above names something.

The measured peer skeleton bank

42 outliers from the top 5 videos of all 17 peer channels, classified: AI reanalysis — 18 outliers across 10 channels (the runaway leader; pays from 5.35K subs to 52K). Divers/drone reached-or-entered a named place — 9 across 5 channels (including 49.6× and 41.6× on a 4.6K-sub channel). Named site opened after N years — 7 across 6 channels. Named person finally reveals — 7 across 6 channels (containing the single biggest outlier measured, 207×). One more was found but is below the bar: 'They Found X. Then They Stopped Digging.' — 1 outlier, 1 channel, 3.2×. Worth watching, not worth betting on yet.

Corrections carried over from version 2

Mega-channel evidence: v1 graded ideas against Kurzgesagt (25.6M), NatGeo (32.1M), Mark Rober (28.9M), BBC and Top Fives. Those prove a subject is interesting and nothing else. The AI block was wrong — see card 2. The '~26:40 target length' was wrong: it came from the three biggest channels in the lane. Inside the peer band, length does not correlate with performance — QUANTARA 7.4× views-per-sub at 25:47, but Discovery Future 3.9× at 8:11 and Sunken Realms Deep 1.8M at 8:34. Winners span 8–36 minutes. 'Petra is blocked forever' was too strong: @Beyonddiscovery (4,560 subs) cloned it for 348K, a 63.3× outlier. The correct rule is that an exact copy of a recent winner is a bad bet, while a swarmed skeleton with a rotated slot is exactly what you want.

What the fan-out got wrong, kept on the record

Independent of the evidence-population problem: the fan-out proved a porcelain-shipwreck idea with barn-find car videos ('This 1960 Chevy Apache sat untouched for 20 years', 45 views), proved a deepest-shipwreck idea with three caving videos, and shipped three titles with the specific number and the em-dash clause stripped out while self-scoring them 1.0 on structure conformance. One card — 'the Most Sealed-Off Valley on Earth' — was killed because a concept-level search for it returns music videos.

Data caveat

Algrow's daily quota was exhausted before this run and NexLev was not connected, so all figures come from the yt-api provider, which does not return comment counts. Everything is graded on views and relative outliers only. Peer outliers are computed against each peer channel's own median across its top 30 videos, so a 207× reflects a very low channel median (The Lost Mystery sits at 1,000 views) and should be read as 'this hugely outperformed its channel', not as an absolute.