The Dossier — 90 Days to 100,000

Baseline: 18,600 subscribers · 60 sampled videos (24 long-form, 36 shorts) · channel age ~14 months · cumulative views ~1.7M. Target: 100,000 subscribers in 90 days — a 5.4× lift, roughly +900 net subs per day for 12 weeks. The channel currently nets an order of magnitude less on its best days.

Not impossible. The audit found three structural leaks capping growth and four assets already strong enough that closing those three leaks would put the trajectory in range. Whether 100k or 80k lands inside 90 days, the program below is the right one to run.

01. Verdict in three sentences

The Dossier has a host the audience is in parasocial love with (~106 “Kaushlesh” mentions vs ~11 “Dossier” in a 702-comment sample) and a defensible USP nobody else in the niche is shipping: Hindu-civilizational receipts — Waqf-text, census tables, the Pact of Umar, SIR voter rolls, Malegaon documents. It also has three structural leaks so large that they cap growth regardless of content quality — a shorts feed publishing 36 videos with 36 empty descriptions and zero links back to long-form; a binge-and-rest cadence with 49-day long-form gaps the YouTube recommender cannot model; and a bimodal length distribution that skips the 25–50 min winner band entirely (5/6 top long-form sit there; 0/24 sit in the 25–30 min slot). The program is to close those three leaks while standardizing four already-working patterns (host-anchor split-portrait thumbnail, named-victim cold open, “Reality of X” / “EXPOSED” title shell, archival-document frame) — and to ship a daily Hindi news short (“Dossier Daily”) as the subscription-acquisition engine.

02. Four assets that already work

The host is the brand. Audience mentions, segment behaviour, and the recurring “how is the host safe?” FAQ all confirm the channel’s most-converted asset is Kaushlesh Rai. The marquee still says “The Dossier”; the audience addresses Kaushlesh by name in roughly a third of long-form comments. Across a 702-comment sample, “Kaushlesh” / “कौशलेश” appears ~106 times ([nC4cyUEVDek] [fjGkaPLP_KA] [oYTJaHw4na8] [shyKx_mmiic] [M6DiwBW4UnU] [Hdrj7Q21Fm0] [t2JcdxBweSw] [Ipd9sXG8dLI]); “Dossier” / “डोजियर” appears only ~11 times. One commenter on [M6DiwBW4UnU] complains that “Dossier ke videos” do not surface despite a subscription — they identify the host first and the channel second.

The host-anchor split-portrait thumbnail wins, when followed. 100% of top long-form thumbnails use host-on-right + antagonist/victim-on-left + bold red/yellow declarative overlay with no ?. 38% of bottoms break that recipe. Three off-recipe losers tell the story: [atLAW_yl70Q] (greyscale, no host), [6Ix6Tt4I47A] (Iron Man helmet collage), [Ipd9sXG8dLI] (Nehru-stabbing-map tableau). The work is to lock the spec, not invent a new one. The Devanagari overlay variant is also strong — the single highest-performing long-form [yCR38NmeCpY] uses it.

Named-victim or insider cold opens beat treaty/date/doctrine cold opens decisively. Top long-form lands its first concrete claim inside 15 seconds in 5 of 6 cases. Bottom waits 30–45 seconds — Babri waits past the date [W4I42zE8qTA], Pakistani Hindu refugees waits through the 1950 treaty [atLAW_yl70Q], Iran-war waits through the bombing description [6Ix6Tt4I47A]. The fix is a script rule, not a production change.

“Receipts” content is the lane no peer occupies. Across 8 acquired peers in the niche, none cover the Waqf legal framework, ex-Muslim testimony, the Pact of Umar / “22 rules for Hindus in an Islamic state”, Hindu-Terror narrative reversal (Malegaon, Col Purohit, Samjhauta), Delhi land-mafia ground reporting, the Lalu-Yadav archival shorts series, Pakistani Hindu refugees, or delimitation through a “140-year data” lens. The channel’s edge is the document-and-date posture inside a Hindu-civilizational frame: SIR voter-roll numbers in [3uk0G_LNd7A], 1881–2011 census tables in [yLEakbf5gpI], Waqf Act text in [fjGkaPLP_KA], Nehru-Liaquat Pact in [atLAW_yl70Q], Pact of Umar in [1E6So2gYgP8]. Self-description sells “politics, society, religion, culture, law — rooted in truth.” That undersells the actual offer: archival receipts deployed inside a Hindu-grievance frame.

03. Three binding constraints

3.1 The shorts funnel is sealed shut

36 of 36 sampled shorts have empty descriptions. 0 contain a YouTube URL. Only 1 of 36 has any CTA at all in the script ([JZAr30UgX_c] says “tell us”). Yet shorts is the only surface where the channel achieves high velocity — top short [rpk9_nXIvdM] hit z=2.78 in 32 days. The format with the highest reach routes none of its viewers to long-form or to the subscribe button. Unblocking this leak — a 3-line description template plus a 2-second outro voice tag — more than doubles the input to subscriber growth because every short-view becomes a potential sub-click instead of a dead-end.

Three shorts also exceed the 2:30 Shorts shelf cutoff ([NHTNJPcvO0Q] 173s, [Y6pF1fIZu_c] 161s, [Flc5mIdH6Ug] 155s). YouTube’s classifier may silently re-route them, splitting reach.

3.2 Cadence is binge-and-rest, not predictable

Long-form: 14-day median interval, 49-day max gap, modal day Saturday (9/24) — but Saturday is also the channel’s worst long-form day (0/6 top, 3/6 bottom). Sunday outperforms (3/6 top, 0/6 bottom). Shorts: 0 in 4 months (Nov 2025–Feb 2026), 13 in March 2026, 4 in April. Audience FAQ #5 (“why long gaps?”) confirms what the algorithm sees — viewers and the recommender both cannot model the channel’s frequency. The fix is a Saturday-7pm-IST premiere lock for long-form, a Wednesday 7pm slot for a numbered weekly recap, and a daily morning slot for Dossier Daily shorts.

3.3 The bimodal length distribution sacrifices the winner band

Below 15 min is the channel’s graveyard (5 of 6 bottom-quartile sit there); the proven winner band is 25–50 min (5/6 top long-form sit there, mean ~50 min). The 0–5 min cluster is mostly mis-classed “mini long-form” — [nFIGdsXAYpA] 293s, [W4I42zE8qTA] 328s, [Gs38SCGxPGA] 484s — too short to be deep dives and too long to be shorts. Pure waste. Inside shorts, the equivalent graveyard is sub-30s (44% of bottom shorts vs 0% of top). Editorial rule: every long-form is either ≥25 min or recut to a short; every short is ≥50s.

04. The catalogue, separated by cohort

Within-cohort z-scores. Long-form and shorts scored separately so they are never ranked against each other.

4.1 Top performers — long-form (6)

4.2 Bottom performers — long-form (6)

4.3 Top performers — shorts (9)

Top shorts include [rpk9_nXIvdM] (TCS Story, z=2.78), [NHTNJPcvO0Q] (Lalu, z=2.33), [fmrAS_vkCUE] (Lalu, z=1.76), [5wMrZDvZl4k] (Lalu, z=1.57), [CDdpB7oWm0k] (Purohit, z=1.22), [rajRbU2z2mw] (TCS Nashik, z=1.22), [TnGcTkrTrEo] (CJI Gavai, z=1.06), [PGNi1Zw-tA0] (Partition framing, z=0.70), [Flc5mIdH6Ug] (ATS / Purohit, z=0.30).

4.4 Variables that correlate — long-form

• Length ≥ 25 min: 5/6 (83%) top vs 0/6 bottom. +83pp.
• Sub-15-min: 3/6 bottom vs 0/6 top. The channel’s graveyard.
• Saturday upload: 0/6 top vs 3/6 bottom. Modal day is the worst day.
• Sunday upload: 3/6 top vs 0/6 bottom.
• CTA present: 4/6 top vs 2/6 bottom (3× rate). Partial confound with length.
• religion_society as primary cluster: 1/6 top vs 4/6 bottom. Bottom is dominated by religion-coded explainers; top diversifies into crime, judiciary, politics, geopolitics.

4.5 Variables that correlate — shorts

• Length ≥ 50s: 9/9 (100%) top vs 3/9 (33%) bottom. +67pp. Every winning short uses the full window.
• Sub-30s: 0/9 top vs 4/9 bottom. The three worst by z-score are all under 30s ([WCiY3qM_s00] 21s, [6mZkBzkrRmE] 23s, [HCSoVCTMz8A] 26s).
• politics_govt primary: 4/9 top vs 1/9 bottom (three are the Lalu Yadav cluster — one narrative arc that travelled).
• law_judiciary primary: 0/9 top vs 2/9 bottom. Law content compresses badly into clip format.
• Wed/Sat upload: 5/9 top vs 1/9 bottom. Sun/Mon: 0/9 top vs 4/9 bottom.

4.6 Variables that don’t correlate

Numerals in titles (directions inconsistent at this n). Short-title length (top 67 chars vs bottom 68 — flat). Shorts CTA count (the channel doesn’t use them). Geopolitics topic (split evenly).

05. Hooks: the first 30 seconds

5.1 Top long-form hook patterns

Story cold-open with a single named voice. Survivor naming her village and police station before any narration [yCR38NmeCpY]. Ex-Maulvi describing his madrassa years [oYTJaHw4na8]. MLA recounting how he stopped cow-smuggling [efBn5kJ0snc]. No narrator setup, no “today we will discuss.” The voice itself is the hook; 10–15 seconds of music or ambient framing carries the viewer in.

Immediate accusation, stated as fact before evidence. A roll-call of court orders against Hindu practices before the channel’s framing arrives [zv2HCSdsiJw]. A flat “we have limited time, the goals are big” military framing followed by named targets [Gs38SCGxPGA]. The opening is a verdict, not a question. Evidence follows.

Stake-setting with a named adversary in the first breath. Asim Munir by name and rank [Gs38SCGxPGA], the judiciary’s nepotism [zv2HCSdsiJw], Mamata’s government [efBn5kJ0snc]. The viewer knows under 10 seconds who the video is against.

5.2 Top shorts hook patterns

Single shocking sentence in the first 3 seconds. Graphic paraphrased court ruling delivered deadpan [TnGcTkrTrEo]. Workplace-corruption claim about a named company [rajRbU2z2mw]. Torture allegation against a named officer [Flc5mIdH6Ug]. The most viral line is the cold open.

Question-to-guest with the answer mid-flow. Lalu Yadav podcast clips open mid-question with the guest already mid-answer [5wMrZDvZl4k] [fmrAS_vkCUE] [NHTNJPcvO0Q]. Viewer lands in a conversation, not a setup.

Concrete factual hook tied to a topical event. Bangladeshi MP parliamentary statement [PGNi1Zw-tA0]. Purohit-case specifics with year counts and a named officer [CDdpB7oWm0k]. Specificity in sentence one wins.

5.3 Bottom hook patterns

Treaty/date/doctrine cold open before any human stake. 1950 Nehru-Liaquat pact [atLAW_yl70Q]. 1992 demolition framed as background [W4I42zE8qTA]. Pact of Umar routed through a Sydney news incident [1E6So2gYgP8]. The viewer waits 30–60 seconds for the actual claim.

Geopolitics narrated as scene-setting newsreel. Iran/Israel openings in a neutral newsreader register [6Ix6Tt4I47A]. Without the channel’s adversarial voice in the first 20 seconds, the audience does not stick. Re-cut into a short, the same opening underperformed identically [6mZkBzkrRmE].

Profile-style biographical setup of a third party. “This is the girl who…” then explaining who she is, where she works, before any provocative claim [nFIGdsXAYpA]. The viewer is being introduced to someone they don’t care about yet.

Bottom shorts also share three patterns: mid-conversation fragments with no anchor [WCiY3qM_s00] [NFmZry3zzM0], recycled long-form openings pasted into vertical [6mZkBzkrRmE], and abstract ideological statements without a concrete target [s-PodvUEtv8] [HCSoVCTMz8A] [wCQ6uShULAM] [5JF7pxPlRmE] [Q4G_3RBO0Nk].

5.4 Time-to-first-substance

5.5 Five templates to adopt

1. [Long] “I am [name], from [village/role]. [One-sentence stake].” First-person identity + stake. Removes the need for narrator scaffolding. ([yCR38NmeCpY] [oYTJaHw4na8])
2. [Long] “In [year], [institution] did [specific act]. In [year], same. In [year], again.” Three-case staccato list before any thesis. Pattern is the argument. ([zv2HCSdsiJw])
3. [Short] “[Named person] [did extreme thing] — [specific number / place].” One sentence, concrete noun, no setup. ([rpk9_nXIvdM] [Flc5mIdH6Ug] [CDdpB7oWm0k])
4. [Short] “[Direct question to guest about specific event]?” — guest answers immediately. Cut all host setup. ([5wMrZDvZl4k] [NHTNJPcvO0Q])
5. [Long] “[Named adversary by full title/rank], [verb of opposition], [the channel’s stake].” Adversary in sentence one. ([Gs38SCGxPGA] [efBn5kJ0snc])

06. Pre-click: titles, descriptions, tags, thumbnails

6.1 Title patterns — long-form

Top (n=8) vs bottom (n=8). Length and word counts nearly identical (88 vs 83 chars). The discriminators are punctuation and stance.

• Colon-as-frame: 62% top vs 50% bottom. Subject : payoff structure.
• Question titles: 12% top vs 50% bottom. [Ipd9sXG8dLI] [nSUiQE5zxbY] [DNHWfGBfUiM] [W4I42zE8qTA] [6Ix6Tt4I47A] all end in ? and underperform. Tops assert.
• Power-verb / outrage frame: 75% top vs 12% bottom. EXPOSED, Dirty Secrets, Rained Hell, Horror Exposed [shyKx_mmiic] [zv2HCSdsiJw] [Gs38SCGxPGA] [yCR38NmeCpY]. Bottoms use Explained, Understanding, All about.
• Named living entity: 88% top vs 38% bottom. Rampur, Yashoda, Mamta, Murshidabad, Operation Sindoor, Samik Bhattacharya vs Hindu Refugees, Dhimmi, Iran-Israel War, Nehru.
• Specific numerals: 50% top vs 25% bottom — 17 Mur*ers, 24 Strikes, 9 Terror Camps, 70 Killed, 140 साल का Data.

6.2 Title patterns — shorts

Top (n=7) vs bottom (n=7).

• Devanagari script: 57% top vs 29% bottom. The English-only top short outlier The TCS Story 😱 #shorts (23 chars, z=2.78) wins on brevity + emoji + mystery.
• Recurring character handle: 43% top vs 0% bottom. Three Lalu shorts ride #laluyadav. The channel found a face that performs.
• Quoted pull-lines: 0% top vs 29% bottom. Quotes are flat declarations without a curiosity gap [s-PodvUEtv8] [HCSoVCTMz8A].
• #shorts hashtag: 71% top vs 43% bottom. Likely necessary, not sufficient.

6.3 Description patterns

• All 14 audited shorts have blank descriptions. 100% miss on the description-as-pre-click surface for shorts. (Minor pre-click; major funnel leak.)
• Long-form descriptions split: hook restatement vs press release. Winners open with first-person hook restatement that mirrors the thumbnail/title promise ([yCR38NmeCpY] [nC4cyUEVDek] [zv2HCSdsiJw] [Ipd9sXG8dLI]). Losers open with neutral synopsis (“The Iran-Israel war began on June 13, 2025…”) [DNHWfGBfUiM] [6Ix6Tt4I47A].
• No timestamps, no CTA cluster, no hashtags in first 150 chars of any audited video. First 150 chars are content-only across the sample.
• Hindi script matches Hindi title in tops and bottoms equally — script-match is hygiene, not a discriminator.

6.4 Tag strategy

• Shorts use zero tags across all 14 audited shorts.
• Long-form: top avg 27.2 tags vs bottom 20.5. Bottom has a 0-tag outlier [Ipd9sXG8dLI].
• Tags appearing in 2+ tops but never in bottoms: the hindu cafe, president rule, ajeet bharti, corruption, modi, yogi adityanath. supreme court appears 4× in tops, 0× in bottoms.
• kaushlesh rai tag: 4× in top, 2× in bottom — winners brand-tag more aggressively.
• Loser tag pattern: abstract category nouns (geopolitical tensions, defense news, nuclear proliferation) and esoteric framings (viveka, spirituality, civilizational threats).

6.5 Thumbnails — the winning recipe

Top long-form (n=8) is on-template in 8/8:

• Host on the right third, neutral-to-stern, red tilak, red or maroon shirt.
• Antagonist/victim on the left third. Split-portrait layout.
• Bold all-caps overlay (English in 7/8, Devanagari in 1 — and that 1 is the channel’s highest performer).
• Two-tone palette: red/orange/yellow saturated on dark/black background in 7/8.
• Distressed/intense subject expression in 5/8.
• No ? in overlay. Zero question marks in tops. 38% of bottoms have one. Tops assert, bottoms ask.

Bottom thumbnails (8) depart from this in measurable ways: host de-emphasised in 3/8 [Ipd9sXG8dLI] [atLAW_yl70Q] [nFIGdsXAYpA], lower contrast / off-palette in 3/8 (greyscale [atLAW_yl70Q], surreal collage [6Ix6Tt4I47A], ornate scene-staging [Ipd9sXG8dLI]), over-staged tableau compositions. Top vs bottom long thumbnail deltas ≥ 20pp: host-anchor present (100% vs 62%, +38pp); declarative overlay (100% vs 62%, +38pp); clean split-portrait (100% vs 37%, +63pp).

Top shorts (n=7) cluster on a single template: 3-face top half (host + co-host + character cutout), torn-paper red Hindi text card below, The Dossier ribbon footer. Used by 3 Lalu shorts and Purohit [NHTNJPcvO0Q] [fmrAS_vkCUE] [5wMrZDvZl4k] [CDdpB7oWm0k]. The TCS short [rpk9_nXIvdM] is the exception — black hooded perp walk + host face + grunge English title — but high contrast and crime-scene framing. The code-switched TATA CONSULTANCY मदरसा SERVICES overlay on [rajRbU2z2mw] reuses the same perp-walk photo with bilingual text; outperformed its English-only quote-card sister content. Top vs bottom shorts deltas ≥ 20pp: host face present (86% vs 14%, +72pp); recurring familiar character (71% vs 29%, +42pp); photo-real (not AI-illustrated) main subject (100% vs 57%, +43pp).

6.6 Ten title templates (fill-in-the-blanks)

All ≤80 chars. Ranked by leverage. L=long, S=short.

1. L {Place} {Crime} Horror Exposed: {Victim Name}'s {Number} {Atrocity} — mirrors [yCR38NmeCpY].
2. L {Operation Name}: When India {Power Verb} in {Adversary}: {N}, {N}, {N} Killed — mirrors [Gs38SCGxPGA].
3. L {Entity}'s Dirty Secrets: {Frame 1}, {Frame 2} & {Frame 3} Exposed — mirrors [zv2HCSdsiJw] [shyKx_mmiic].
4. S भूरा बाल साफ करो {Politician} ने कभी नहीं कहा? #shorts #{politiciantag} — mirrors [NHTNJPcvO0Q].
5. S {Politician} के {English Noun}, {Adjective} होने का ढोंग! #shorts #{politiciantag} — mirrors [fmrAS_vkCUE].
6. L Ex-{Role} Unfiltered: Why I Couldn't Stay in {Group} ft {Guest} — mirrors [oYTJaHw4na8].
7. S The {Topic} Story 😱 #shorts — mirrors [rpk9_nXIvdM].
8. L {Geography}'s {Crime}: Inside the Cartel of {Frame} EXPOSED ft {Activist} — mirrors [shyKx_mmiic].
9. L {Topic} का झूठ और {Reform} का सच। {Number} साल का Data | The Dossier — mirrors [yLEakbf5gpI].
10. B My Message for the {Tribe} on the {Recent Event} — mirrors [nC4cyUEVDek].

6.7 Seven thumbnail rules

1. Host-anchor split-portrait. Host (Kaushlesh, red tilak) on the right at consistent crop; antagonist/victim on left.
2. Overlay text asserts, never asks. Zero ?. Replace “Hidden from Hindus?” with “Hidden from Hindus.” or “EXPOSED.”
3. Red/yellow on dark background. Greyscale and AI collage lose.
4. Drop the clean white-card / thin-English-caption shorts template entirely. Standardize on red torn-paper + 3-face top half + The Dossier ribbon.
5. Use real photos, not AI collages, for shorts.
6. Cap overlay text at 3 lines / ~25 chars per line.
7. Code-switch English-Devanagari for shorts overlay when antagonist is named in a Hindi context.

07. The audience

7.1 Six segments

Sizes are rough share of engaged commenters in a 702-comment sample.

1. Mobilised Hindu nationalist core (~45–55%). Largest cluster. Markers: framing every story as Hindu-vs-invader; “Jai Shri Ram”, “Har Har Mahadev”, saffron and trishul emoji. Recurring across [yCR38NmeCpY] [efBn5kJ0snc] [zv2HCSdsiJw]. Often calls for capital punishment, Yogi-tagged petitions.
2. Bengali Hindu in-state segment (~8–12%). Concentrated on [efBn5kJ0snc]. Self-identifies by neighbourhood (Salt Lake Sector 5, Darjeeling). Bengali script for emotional asides, English for argument. Asks for more Bengali-language podcasts, more interviews with WB politicians. A noticeable sub-thread pushes back on the BJP-friendly framing — rare critical voice from inside the base.
3. Ex-Muslim / counter-Islam interest (~6–10%). Surfaces on [oYTJaHw4na8]. Names a circuit (Sahil, Zafar Heretic, Adam Seeker, Apostate Prophet, Saleem Wastik). Treats the channel as a gateway to a wider counter-apologetics ecosystem.
4. Policy-curious / “data person” segment (~8–10%). Visible on [yLEakbf5gpI] [zv2HCSdsiJw]. Paragraph-form comments engaging with delimitation math, Article 81, fertility rates, judiciary reform mechanics. Compares The Dossier favourably to Ajeet Bharti, Ankit Shah, Nitish Rajput.
5. Sceptical / cross-pressure commenters (~5–8%). Heterogeneous: Modi-disillusioned right-wingers (“BJP becoming new Khangress”), caste-left voices, TMC/AAP-aligned pushback [efBn5kJ0snc] [3uk0G_LNd7A], a few openly Muslim defenders. Usually downvoted within-thread but keep reply-trees alive.
6. Diaspora and senior-tone Hindi (~5–7%). Dated signoffs with full name + district + state + country (Odisha, Bradford, Karnataka, Rajasthan); reverent address to the host as “bhaiya”/“sir”/“prabhu”; long, formal Devanagari paragraphs. Skews older. Asks for the content to be published as a book.

7.2 Language and regional mix

Dominant comment language: Hinglish (Roman-script Hindi with English), followed by Devanagari Hindi at roughly equal volume on long-form. English-only is a minority, clustered on the policy/data and ex-Muslim segments. Bengali script appears almost exclusively on [efBn5kJ0snc]. Regional self-identification leans North/East: WB, UP, Bihar, Rajasthan, Karnataka, Maharashtra, Assam, Odisha, plus a small diaspora cohort. South Indian non-diaspora voices are sparse; one Karnataka commenter explicitly thanks the host for pushing back on Arya/Dravidian framings [yLEakbf5gpI] — latent appetite the audience itself is asking the channel to serve.

7.3 Top FAQ themes

Frequencies are counts in the filtered sample. Triggers list the videos most responsible.

1. “Why isn’t mainstream media covering this?” (~22) — [yCR38NmeCpY] [zv2HCSdsiJw] [3uk0G_LNd7A] [efBn5kJ0snc]. CTA opportunity around comparative-coverage explainers.
2. “What is the actual remedy / what should we do as citizens?” (~18) — [zv2HCSdsiJw] [yLEakbf5gpI] [3uk0G_LNd7A]. Audience asking for action toolkits; the channel has not produced one.
3. “When are you doing the next episode with Ajeet Bharti?” (~10) — [rpk9_nXIvdM] [3uk0G_LNd7A] [6Ix6Tt4I47A]. Cross-promo cadence matters to them.
4. “Will BJP actually do anything about X?” (~17) — [zv2HCSdsiJw] [efBn5kJ0snc] [rpk9_nXIvdM] [3uk0G_LNd7A]. Mature, disillusioned sub-strain.
5. “Why don’t you post more often / why long gaps?” (~9) — [Gs38SCGxPGA] [3uk0G_LNd7A] [nSUiQE5zxbY] [yLEakbf5gpI]. Indicates retention and parasocial pull — and frustration.
6. “How is the host safe? Take care.” (~12) — [zv2HCSdsiJw] [yCR38NmeCpY]. Audience wants reassurance the host hasn’t been silenced or sued.
7. “Where can we donate?” (~6) — [yLEakbf5gpI] [oYTJaHw4na8] [nSUiQE5zxbY]. Conversion funnel underbuilt.
8. “Publish as a book / PDF / share notes” (~5) — all on [zv2HCSdsiJw]. Downloadable artefacts of heavy research.
9. “Bengali / Tamil / Kannada version?” — [efBn5kJ0snc] [yLEakbf5gpI]. Low but pointed.
10. “Constitutional mechanism for X?” (~10) — [zv2HCSdsiJw] [yLEakbf5gpI] [JZAr30UgX_c]. Explainer demand for procedure-level content.

7.4 Sentiment

Filtered comment distribution: positive toward host/channel ~70% (“brilliant”, “eye-opener”, “kadwa sach”, “fearless”). Mixed ~15% (praise plus feature requests). Negative toward channel ~8% — sub-flavours: accusations of “BJP propaganda” [yLEakbf5gpI] [3uk0G_LNd7A]; caste-coded pushback [yLEakbf5gpI]; pro-Islam defenders on [oYTJaHw4na8] [yCR38NmeCpY]. Neutral/procedural ~7%. Operation Sindoor / war content draws almost no negative — the wartime jingoist register suppresses dissent.

7.5 Parasocial markers

The bond strength is high and unusually first-name-coded for an Indian political channel. Host addressed as “Kaushlesh bhai” or “Kaushlesh ji” in roughly a third of long-form comments; “bhaiya” and “sir” next most common, “prabhu” in the senior segment. Commenters greet the host’s family (blessings for long life, safety prayers). After gaps, comments paraphrase “we waited for you” [Gs38SCGxPGA] [nSUiQE5zxbY]. Episode-recall is frequent (“your judiciary video was the best ever”). Catchphrases (the “supreme kotha” coinage from the judiciary episode) reappear spontaneously. Concern-trolling-as-affection about legal exposure recurs. The conversion ceiling is being throttled less by audience bond and more by upload cadence and absence of a structured membership ladder.

08. Niche and peers

8.1 Niche statement

Hindi-leaning, English-titled long-form video essays (typical 30–60 min) arguing a Hindu civilizational point of view on contemporary Indian politics, religion, law and society. Recurring subject matter clusters: Islam in India, Hindu–Muslim demographics and conversion, judicial bias against Hindus, Bengal politics, historical revisionism of the freedom struggle and Partition. Format: presenter-to-camera commentary with frequent guest interviews (ex-Muslims, BJP MLAs/MPs, lawyers, scholars). Audience reads as ideologically right-of-centre, English-literate Indian viewers (and diaspora) wanting a “files-and-evidence” framing rather than studio-news shouting.

8.2 Peer roster (8 channels)

The 8 channels in data/competitors/ were acquired as aspirational reference, not head-to-head peers. The Dossier sits at 18.6k subs; the smallest peer (Faye D’Souza, 469k) is 25× larger; the largest (The Lallantop, 34.8M) is ~1,800× larger. There are no acquired peers in the 100k–500k “ideal next rung” band. Peer playlists supplied no per-video view_count or timestamp fields, so per-peer “uploads/month” and “median views” cannot be computed from this dataset; sample-of-20 title patterns and durations only.

8.3 What The Dossier uniquely owns in this peer set

• Waqf Act explained as a Hindu-vs-Muslim asymmetry [fjGkaPLP_KA] [nC4cyUEVDek] — none of the 8 peers cover.
• Ex-Muslim testimony interviews as a recurring format. Saleem Wastik appears twice [oYTJaHw4na8] [t2JcdxBweSw]; the breakout to 111k views is the strongest single-format signal in the dataset.
• “Dhimmi / Pact of Umar / 22 rules for Hindus in an Islamic state” [1E6So2gYgP8] — register no peer occupies.
• Hindu-Terror narrative reversal (Malegaon / Col Purohit / Samjhauta) — sustained shorts series in Sep–Oct 2025 [dHFW-Gi3qu8] [CDdpB7oWm0k] [qjm31UWuMJs] [Flc5mIdH6Ug] [NqpRtHxfML0] [wCQ6uShULAM].
• Delhi land-mafia ground reporting with Preet Sirohi [WcIfoJvkHfQ] [3E6qLtL4qvs] [wJB_hfKDnFI] [7ZU-23P09ww] [SjvN3tQ_CjA] [Q4G_3RBO0Nk] [shyKx_mmiic].
• Bihar / Lalu Yadav archival shorts series [fmrAS_vkCUE] [NHTNJPcvO0Q] [r_YhEw0EfBA] [5wMrZDvZl4k] [M6DiwBW4UnU].
• Population / delimitation hard-data piece in Hindi [yLEakbf5gpI] — 46k views, highest of any 2026 long-form.
• Pakistani Hindu refugees [atLAW_yl70Q] — absent from all peers.
• Rampur crime as long-form Hindu-victim reporting [yCR38NmeCpY] — 258k views, biggest single video. Lallantop covers crime but as news flash, not 46-min investigation.

8.4 What peers own that The Dossier doesn’t (sample)

NEET paper-leak coverage — single most-shared topic in the peer sample; Dossier has zero. Modi-government performance critique on petrol/austerity/energy. Tamil Nadu / Vijay / TVK politics. Trump-Xi geopolitics. Middle-class economics, IT-job AI risk, personal finance. Press freedom and media criticism. Manipur ethnic conflict. Daily news anchor format. Civilisational-history Tarikh-style storytelling. True-crime and cartel storytelling.

8.5 Five series ideas borrowed from peer success

1. “Point Blank-style numbered weekly recap.” Source: @AjeetBharti’s EP01–EP11. Adaptation: The Dossier Weekly — 15–20 min, numbered EP, fixed Saturday slot.
2. “NL Charcha / Hafta-style panel show.” Source: Newslaundry numbered panel franchises. Adaptation: The Dossier Mahfil — recurring 60-min panel with two co-hosts + one rotating guest from the channel’s existing roster.
3. “Tarikh-style historical narrative.” Source: @thelallantop and @AbhijitChavda. Adaptation: Dossier Itihaas — 10–15 min historical narrative on specific events.
4. “Reality of X” / “EXPOSED” packaging shell. Source: @dhruvrathee. The channel’s content is already investigative; the packaging leaks views. Rampur [yCR38NmeCpY] should have been “Reality of Rampur.”
5. “Daily news recap in Hindi (short).” Source: @FayeDSouza’s “Just The News.” Adaptation: Dossier Daily — 60–90s daily, three named stories with the channel’s framing.

09. Brand identity: the gap between marquee and audience

9.1 One-sentence positioning (audience voice)

“Kaushlesh bhai’s Hindi-English channel where he explains, with old documents and numbers, why Hindus are getting a raw deal from Muslims, the Nehruvian state, the courts, the media and the liberal-leftist ecosystem.”

The channel’s own description (“India’s chronicle of politics, society, religion, culture and law – rooted in truth”) describes a different, much more neutral product than the one being shipped.

9.2 Verbal consistency — low to medium

Cold-open vs hello-friends split. Two intro patterns coexist with no rule. Cinematic cold open: [3uk0G_LNd7A] [Q9wui2puJdE] [nFIGdsXAYpA] [atLAW_yl70Q]. Hello-friends opener: [t2JcdxBweSw] [Ipd9sXG8dLI] open with verbatim “Hello friends, how are you all? I hope you are fine.” Guest-led: [Hdrj7Q21Fm0] opens on the guest’s voice mid-sentence. A viewer landing on three random videos in a row gets three different show-openings.

Outro CTAs are inconsistent. 13 of 24 recent long-form have any logged CTA at all. Scattered across “tell us” (5×), “follow us” (3×), “in the comments” (2×), “share the video” (2×), “link in the description” (2×), “forward this” (1×), “membership” (1×). “Jai Shri Ram” is the closest thing to a sign-off but is inconsistently used.

Title-card branding is not a habit. The suffix ”| The Dossier” appears on only 2 of 24 recent long-form titles ([yLEakbf5gpI] [3uk0G_LNd7A]).

Description boilerplate is the strongest verbal asset. Long-form descriptions consistently carry the “Visit The Dossier / Hindi Edition / Instagram / X / Substack / LinkedIn” block plus the “Hindu Cafe Foundation” donation banking details. This is the one place the brand behaves like a system.

9.3 Structural consistency — embryonic series go unmarked

Two formats coexist without naming: solo explainer-essay ([yLEakbf5gpI] [3uk0G_LNd7A] [Ipd9sXG8dLI] [1E6So2gYgP8] [fjGkaPLP_KA] [nC4cyUEVDek] [nFIGdsXAYpA]) and sit-down podcast ([oYTJaHw4na8] [t2JcdxBweSw] [shyKx_mmiic] [Hdrj7Q21Fm0] [M6DiwBW4UnU] [efBn5kJ0snc] [6TVBySERPzE]). Neither format has a name, intro card, or banner.

Some embryonic series exist but go unmarked. The TCS-Nashik story spans 4 videos [Q9wui2puJdE] [rajRbU2z2mw] [rpk9_nXIvdM] [nFIGdsXAYpA] without a “Part 1/2/3” label. The ex-Muslim arc with Saleem Wastik spans 3 [oYTJaHw4na8] [t2JcdxBweSw] [8dNI5HGqNnk]. The Bengal/Mamata arc spans 4. De facto franchises with no franchise branding.

Sub-properties exist but are invisible inside the channel. Descriptions reference “Hindi Edition” (hindi.thedossier.in), a Substack magazine, a donation entity (“Hindu Cafe Foundation”). The YouTube viewer has no clean explanation of how Dossier, Hindu Cafe Foundation, and the Hindi edition relate.

9.4 Identity gaps, ordered by impact on memorability

• The host has no public name on the channel. “Kaushlesh Rai” appears only in descriptions; the channel is not called “Kaushlesh Rai” or “Kaushlesh Rai presents The Dossier.”
• No format names. A Dossier essay and a Dossier podcast are not labelled differently.
• No recurring opening or closing line. No nine-word audio logo viewers could mimic.
• Logo placement is random across thumbnails (six different positions in 14 samples).
• No single closing CTA.
• De facto series go unbranded.
• Multi-property confusion (Dossier, thedossier.in, Hindi edition, Substack, Hindu Cafe Foundation).
• Self-description undersells the actual product.

10. The 12-experiment program

Twelve organic experiments. Ordered by leverage × ease.

Six are low-cost rules or specs that ship in week 1–2 (#1, #3, #5, #6, #9, #11, #12). Three are recurring series with real production cost (#7, #8, #10). Two are channel-structure plumbing (#2, #4). All 12 trace to at least one prior finding.

10.1 Experiments in detail

1. Shorts-to-long-form CTA stamp

• Hypothesis: Every short ships with a 3-line description containing one specific long-form link plus an end-card voiceover line. Shorts→long click-through moves from ~0% to ≥2%.
• Intervention: Template — line 1 = hook; line 2 = Full video: youtu.be/<id>; line 3 = Subscribe @thedossier_in. 2-second outro voice tag pointing to parent long-form.
• Success metric: Studio “Shorts feed” → long-form watch time. Long-form views attributed to shorts ≥5% of long-form impressions within 30 days.
• Kill criterion: 90-day attributed long-form views from shorts source <2% of long-form total impressions.

2. Saturday 7pm IST long-form premiere lock

• Hypothesis: Scheduled Premieres at fixed slot raise first-48h impressions ≥30% via cadence predictability + Premiere notification spike.
• Intervention: Stop publishing direct. Always schedule. Build a 2-video buffer.
• Success metric: Median first-48h views per long-form (current ~5k) ≥8k within 6 weeks. Premiere chat ≥50 concurrent by week 8.
• Kill criterion: Median first-48h views drop vs pre-lock baseline after 6 Premieres.

3. Kill the 5–15 min long-form band

• Hypothesis: Stopping uploads in the 5–15 min graveyard band drops bottom-quartile density.
• Intervention: Editorial rule. Any draft under 15 min gets re-cut: compressed to a short or expanded to ≥25 min.
• Success metric: Zero new long-form uploads in 5–15 min band over 90 days. Median long-form view_z of new uploads ≥0.
• Kill criterion: Non-killable rule. Violation = re-edit before publish.

4. Build four topical playlists from existing catalogue

• Hypothesis: Four real franchises (Operation Sindoor/Pak, Waqf/Islamic law, Judiciary, Ex-Muslim) become custom playlists with set-next chains. Average views-per-session rises ≥20%.
• Intervention: Build playlists in Studio. Auto-add each long-form to relevant playlist. Description line 1 links the playlist URL.
• Success metric: Studio views-per-session on long-form ≥1.3 within 6 weeks.
• Kill criterion: Kill any playlist with <500 cumulative views after 60 days.

5. Lock the host-anchor split-portrait thumbnail

• Hypothesis: Every long-form thumbnail follows the host-right + antagonist-left + bold red/yellow declarative overlay (no ?) template. Long-form CTR rises.
• Intervention: One-page thumbnail spec. Build 3 base PSDs. No ?. No greyscale, no AI collage. Logo top-right, fixed size, every time.
• Success metric: Average impressions-CTR on long-form (Studio) +1.5pp within 6 weeks.
• Kill criterion: 6-week CTR rise <0.5pp AND 3 off-template control thumbnails outperform on-template by >20%.

6. Standardize hook templates; kill the slow open

• Hypothesis: Every long-form opens with either (a) named-victim/insider first-person voice or (b) stake-setting declarative naming the adversary in sentence one. 30-second retention rises ≥10pp.
• Intervention: Editorial rule. No treaty/date/doctrine cold opens, no newsreel narration, no biographical setup of a third party. First sentence must contain a named person and a stake.
• Success metric: Studio “Audience retention at 30s” on new long-form +10pp over 8 weeks.
• Kill criterion: Non-killable rule. Violation = script rewrite.

7. Daily Hindi news short — “Dossier Daily”

• Hypothesis: A 60–90 second Hindi news short every weekday at a fixed morning slot raises total monthly impressions ≥3× and subscriber adds ≥2× — Faye D’Souza’s 469k channel is engineered on this format.
• Intervention: 60–90s vertical, host-to-camera or animated text+VO, three named stories with the channel’s framing. Same opening line every day (“Aaj ke Dossier Daily mein teen kaagaz”). Same closing CTA to long-form.
• Success metric: 20 daily shorts shipped in 30 days. Median short views ≥15k by week 8. Subscriber-add rate from shorts ≥2× baseline.
• Kill criterion: Median view count per daily short <3,000 after 30 shipped, OR shipping cadence <3/week for 2 consecutive weeks.
• Dependencies: #1 (CTA template) must exist first.

8. Lalu-style recurring-character shorts series

• Hypothesis: Identifying one recurring antagonist per month and shipping a 4-short arc raises shorts top-quartile share. Familiar-face repetition is the strongest shorts performance signal in the dataset (Lalu cluster: 3 of 9 top shorts).
• Intervention: Monthly editorial slot. Pick one character (Mamata, Owaisi, specific judge, specific politician). Mine archival/podcast footage. Ship 4 shorts in 7 days using the red torn-paper template.
• Success metric: Per-character series median view_z ≥0.5. At least 1 of every 4 enters top quartile by view count within 30 days.
• Kill criterion: Two consecutive monthly series produce zero top-quartile shorts.

9. “Reality of X” packaging shell

• Hypothesis: Adopting the Dhruv Rathee “Reality of X” / “EXPOSED” title shell on investigations raises long-form CTR. Rampur should have been “Reality of Rampur.”
• Intervention: Title rule for investigations: Reality of {Place/Entity}: {Specific Atrocity}, {Number} or {Entity} EXPOSED: {Frame 1}, {Frame 2}, {Frame 3}. No question titles.
• Success metric: Long-form impressions-CTR on investigation videos +2pp within 8 weeks. A/B: re-title 3 existing investigations >90 days old; compare 14-day view delta.
• Kill criterion: 8-week re-title A/B shows <10% view delta on re-titled videos.
• Dependencies: #5 (thumbnail) for full impact.

10. Numbered weekly series — “Dossier Weekly EPxx”

• Hypothesis: A numbered 15–20 min weekly recap of 3 stories every Wednesday raises return-viewer rate and gives the algorithm a stable series-id. Ajeet Bharti’s Point Blank EP01–EP11 is proof.
• Intervention: Wednesday 7pm IST. Title schema: Dossier Weekly EP{n}: {Story 1}, {Story 2}, {Story 3} | Kaushlesh Rai. 15–20 min cap (the only sanctioned long-form below 25 min — exempted because it earns frequency).
• Success metric: 8 episodes shipped in 60 days. EP08 retention curve at 60% within first 60s. Return-viewer rate on the series rises week over week.
• Kill criterion: EP04 retention at 60s <50% OR subscriber-add per episode <100 by EP06.
• Dependencies: #2 (cadence lock).

11. Pinned action ladder + sources PDF

• Hypothesis: Pinned 4-step action-ladder comment plus free sources-PDF link converts the policy/data segment to email/donation. They explicitly ask for both.
• Intervention: Pinned comment template: “What you can do: 1) RTI template <link> 2) MP email script <link> 3) Share to WhatsApp <link> 4) Citations PDF <link>”. Host a single Google Doc with a redirect URL for tracking.
• Success metric: Pinned-comment click-through ≥25% of long-form viewers within 4 weeks. Email-list signups ≥200 within 90 days.
• Kill criterion: Kill the PDF (keep the ladder) if email signups <50 in 60 days.

12. Lock channel identity — “Kaushlesh Rai | The Dossier” + verbal bookends

• Hypothesis: Renaming the channel and locking one cold-open sentence + one sign-off used verbatim every video raises brand-recall and sub-conversion per impression. The audience already addresses the host; the marquee doesn’t match.
• Intervention: Studio: rename channel display title. Every video: open with “Main Kaushlesh, ye hai The Dossier” + close with “Truth ke saath. Jai Shri Ram.” On-screen lower-third with same text for 2 seconds at open. No exceptions.
• Success metric: Subscriber-to-view ratio (subs added / views in 28-day window) +20% within 60 days. Studio search-traffic for “Kaushlesh Rai” overtakes or doubles “The Dossier.”
• Kill criterion: Non-killable identity move. Revisit only if subscriber-to-view ratio drops vs pre-rename baseline after 60 days.

11. The 12-week schedule

Weeks 1–2 are setup; production is not blocked. Weeks 3–6 are series launches, parallelisable because each owns a distinct slot (Sat long-form, Wed weekly, weekday daily, monthly character arc). Weeks 7–12 are steady-state with measurement gates.

12. What to expect: honest math

The channel produces 4–5 uploads per month with a long-form median of ~5k first-48h views and a shorts mean of ~15k all-time. To hit 100k subscribers in 90 days from 18,600, the program needs to:

1. Raise long-form first-48h views to ~15k by mid-program.
2. Ship ~60 shorts via Dossier Daily that average ≥15k views by week 8.
3. Convert ~1.5% of total post-program views into new subscribers (industry-typical for this niche is 0.5–2%).

If shorts cadence doesn’t hit and Dossier Daily lands at 10k median, expect 60–80k subs at day 90 — still 3–4× growth, and a credible foundation for hitting 100k at day 120. The program is robust to one big experiment failing; it is not robust to neither #7 nor #10 shipping at cadence.

The single highest-confidence lever is #1 (shorts CTA stamp) + #5 (thumbnail spec) + #12 (rename + bookends) shipped in week 1. Those three are zero-production-cost rule changes that fix the biggest pre-click and identity leaks the audit found. Everything else compounds on top of them.

Method and limits

One snapshot, eight agents, three rules: every metric traces to a file in data/, no quotes longer than 15 words, no fabricated numbers.

Pipeline. yt-dlp pulled per-video .info.json, auto-captions, top 200 comments, thumbnails, channel metadata. A Python pass built data/videos/summary.jsonl — one line per video — with per-cohort view z-scores (long vs short separately), heuristic topic clusters, beat segmentation, CTA detection. Seven subagents ran in parallel against fixed slices: algorithmic-positioning, content-forensics, hook-retention, seo-packaging, audience-community, brand-identity, niche-competitor. An eighth (growth-experiments) read all seven and ranked twelve experiments by leverage × ease. The orchestrator merged everything into this guide.

Known limits. 60 of 80 videos sampled — YouTube’s bot challenge blocked 20 mid-acquisition. The 60-video set is treated as authoritative; the 25% shortfall is flagged in every section. 53 of 60 transcripts (above the 80% gate). Peer set is aspirational: the 8 peers in data/competitors/ are 25×–1,800× larger than The Dossier. Peer view counts are unavailable — top-20 flat playlists from yt-dlp carry no per-video view counts or upload timestamps; niche comparisons use title and duration patterns only.

What this is not. Not a live dashboard. Not an SEO keyword tool. Not a content generator. The roadmap proposes experiments; humans execute them.